xref: /linux/drivers/dma/amba-pl08x.c (revision c9895ed5a84dc3cbc86a9d6d5656d8c187f53380)
1 /*
2  * Copyright (c) 2006 ARM Ltd.
3  * Copyright (c) 2010 ST-Ericsson SA
4  *
5  * Author: Peter Pearse <peter.pearse@arm.com>
6  * Author: Linus Walleij <linus.walleij@stericsson.com>
7  *
8  * This program is free software; you can redistribute it and/or modify it
9  * under the terms of the GNU General Public License as published by the Free
10  * Software Foundation; either version 2 of the License, or (at your option)
11  * any later version.
12  *
13  * This program is distributed in the hope that it will be useful, but WITHOUT
14  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
15  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
16  * more details.
17  *
18  * You should have received a copy of the GNU General Public License along with
19  * this program; if not, write to the Free Software Foundation, Inc., 59
20  * Temple Place - Suite 330, Boston, MA  02111-1307, USA.
21  *
22  * The full GNU General Public License is in this distribution in the file
23  * called COPYING.
24  *
25  * Documentation: ARM DDI 0196G == PL080
26  * Documentation: ARM DDI 0218E == PL081
27  *
28  * PL080 & PL081 both have 16 sets of DMA signals that can be routed to any
29  * channel.
30  *
31  * The PL080 has 8 channels available for simultaneous use, and the PL081
32  * has only two channels. So on these DMA controllers the number of channels
33  * and the number of incoming DMA signals are two totally different things.
34  * It is usually not possible to theoretically handle all physical signals,
35  * so a multiplexing scheme with possible denial of use is necessary.
36  *
37  * The PL080 has a dual bus master, PL081 has a single master.
38  *
39  * Memory to peripheral transfer may be visualized as
40  *	Get data from memory to DMAC
41  *	Until no data left
42  *		On burst request from peripheral
43  *			Destination burst from DMAC to peripheral
44  *			Clear burst request
45  *	Raise terminal count interrupt
46  *
47  * For peripherals with a FIFO:
48  * Source      burst size == half the depth of the peripheral FIFO
49  * Destination burst size == the depth of the peripheral FIFO
50  *
51  * (Bursts are irrelevant for mem to mem transfers - there are no burst
52  * signals, the DMA controller will simply facilitate its AHB master.)
53  *
54  * ASSUMES default (little) endianness for DMA transfers
55  *
56  * The PL08x has two flow control settings:
57  *  - DMAC flow control: the transfer size defines the number of transfers
58  *    which occur for the current LLI entry, and the DMAC raises TC at the
59  *    end of every LLI entry.  Observed behaviour shows the DMAC listening
60  *    to both the BREQ and SREQ signals (contrary to documented),
61  *    transferring data if either is active.  The LBREQ and LSREQ signals
62  *    are ignored.
63  *
64  *  - Peripheral flow control: the transfer size is ignored (and should be
65  *    zero).  The data is transferred from the current LLI entry, until
66  *    after the final transfer signalled by LBREQ or LSREQ.  The DMAC
67  *    will then move to the next LLI entry.
68  *
69  * Only the former works sanely with scatter lists, so we only implement
70  * the DMAC flow control method.  However, peripherals which use the LBREQ
71  * and LSREQ signals (eg, MMCI) are unable to use this mode, which through
72  * these hardware restrictions prevents them from using scatter DMA.
73  *
74  * Global TODO:
75  * - Break out common code from arch/arm/mach-s3c64xx and share
76  */
77 #include <linux/device.h>
78 #include <linux/init.h>
79 #include <linux/module.h>
80 #include <linux/interrupt.h>
81 #include <linux/slab.h>
82 #include <linux/dmapool.h>
83 #include <linux/dmaengine.h>
84 #include <linux/amba/bus.h>
85 #include <linux/amba/pl08x.h>
86 #include <linux/debugfs.h>
87 #include <linux/seq_file.h>
88 
89 #include <asm/hardware/pl080.h>
90 
91 #define DRIVER_NAME	"pl08xdmac"
92 
93 /**
94  * struct vendor_data - vendor-specific config parameters for PL08x derivatives
95  * @channels: the number of channels available in this variant
96  * @dualmaster: whether this version supports dual AHB masters or not.
97  */
98 struct vendor_data {
99 	u8 channels;
100 	bool dualmaster;
101 };
102 
103 /*
104  * PL08X private data structures
105  * An LLI struct - see PL08x TRM.  Note that next uses bit[0] as a bus bit,
106  * start & end do not - their bus bit info is in cctl.  Also note that these
107  * are fixed 32-bit quantities.
108  */
109 struct pl08x_lli {
110 	u32 src;
111 	u32 dst;
112 	u32 lli;
113 	u32 cctl;
114 };
115 
116 /**
117  * struct pl08x_driver_data - the local state holder for the PL08x
118  * @slave: slave engine for this instance
119  * @memcpy: memcpy engine for this instance
120  * @base: virtual memory base (remapped) for the PL08x
121  * @adev: the corresponding AMBA (PrimeCell) bus entry
122  * @vd: vendor data for this PL08x variant
123  * @pd: platform data passed in from the platform/machine
124  * @phy_chans: array of data for the physical channels
125  * @pool: a pool for the LLI descriptors
126  * @pool_ctr: counter of LLIs in the pool
127  * @lli_buses: bitmask to or in to LLI pointer selecting AHB port for LLI fetches
128  * @mem_buses: set to indicate memory transfers on AHB2.
129  * @lock: a spinlock for this struct
130  */
131 struct pl08x_driver_data {
132 	struct dma_device slave;
133 	struct dma_device memcpy;
134 	void __iomem *base;
135 	struct amba_device *adev;
136 	const struct vendor_data *vd;
137 	struct pl08x_platform_data *pd;
138 	struct pl08x_phy_chan *phy_chans;
139 	struct dma_pool *pool;
140 	int pool_ctr;
141 	u8 lli_buses;
142 	u8 mem_buses;
143 	spinlock_t lock;
144 };
145 
146 /*
147  * PL08X specific defines
148  */
149 
150 /*
151  * Memory boundaries: the manual for PL08x says that the controller
152  * cannot read past a 1KiB boundary, so these defines are used to
153  * create transfer LLIs that do not cross such boundaries.
154  */
155 #define PL08X_BOUNDARY_SHIFT		(10)	/* 1KB 0x400 */
156 #define PL08X_BOUNDARY_SIZE		(1 << PL08X_BOUNDARY_SHIFT)
157 
158 /* Minimum period between work queue runs */
159 #define PL08X_WQ_PERIODMIN	20
160 
161 /* Size (bytes) of each LLI buffer allocated for one transfer */
162 # define PL08X_LLI_TSFR_SIZE	0x2000
163 
164 /* Maximum times we call dma_pool_alloc on this pool without freeing */
165 #define PL08X_MAX_ALLOCS	0x40
166 #define MAX_NUM_TSFR_LLIS	(PL08X_LLI_TSFR_SIZE/sizeof(struct pl08x_lli))
167 #define PL08X_ALIGN		8
168 
169 static inline struct pl08x_dma_chan *to_pl08x_chan(struct dma_chan *chan)
170 {
171 	return container_of(chan, struct pl08x_dma_chan, chan);
172 }
173 
174 static inline struct pl08x_txd *to_pl08x_txd(struct dma_async_tx_descriptor *tx)
175 {
176 	return container_of(tx, struct pl08x_txd, tx);
177 }
178 
179 /*
180  * Physical channel handling
181  */
182 
183 /* Whether a certain channel is busy or not */
184 static int pl08x_phy_channel_busy(struct pl08x_phy_chan *ch)
185 {
186 	unsigned int val;
187 
188 	val = readl(ch->base + PL080_CH_CONFIG);
189 	return val & PL080_CONFIG_ACTIVE;
190 }
191 
192 /*
193  * Set the initial DMA register values i.e. those for the first LLI
194  * The next LLI pointer and the configuration interrupt bit have
195  * been set when the LLIs were constructed.  Poke them into the hardware
196  * and start the transfer.
197  */
198 static void pl08x_start_txd(struct pl08x_dma_chan *plchan,
199 	struct pl08x_txd *txd)
200 {
201 	struct pl08x_driver_data *pl08x = plchan->host;
202 	struct pl08x_phy_chan *phychan = plchan->phychan;
203 	struct pl08x_lli *lli = &txd->llis_va[0];
204 	u32 val;
205 
206 	plchan->at = txd;
207 
208 	/* Wait for channel inactive */
209 	while (pl08x_phy_channel_busy(phychan))
210 		cpu_relax();
211 
212 	dev_vdbg(&pl08x->adev->dev,
213 		"WRITE channel %d: csrc=0x%08x, cdst=0x%08x, "
214 		"clli=0x%08x, cctl=0x%08x, ccfg=0x%08x\n",
215 		phychan->id, lli->src, lli->dst, lli->lli, lli->cctl,
216 		txd->ccfg);
217 
218 	writel(lli->src, phychan->base + PL080_CH_SRC_ADDR);
219 	writel(lli->dst, phychan->base + PL080_CH_DST_ADDR);
220 	writel(lli->lli, phychan->base + PL080_CH_LLI);
221 	writel(lli->cctl, phychan->base + PL080_CH_CONTROL);
222 	writel(txd->ccfg, phychan->base + PL080_CH_CONFIG);
223 
224 	/* Enable the DMA channel */
225 	/* Do not access config register until channel shows as disabled */
226 	while (readl(pl08x->base + PL080_EN_CHAN) & (1 << phychan->id))
227 		cpu_relax();
228 
229 	/* Do not access config register until channel shows as inactive */
230 	val = readl(phychan->base + PL080_CH_CONFIG);
231 	while ((val & PL080_CONFIG_ACTIVE) || (val & PL080_CONFIG_ENABLE))
232 		val = readl(phychan->base + PL080_CH_CONFIG);
233 
234 	writel(val | PL080_CONFIG_ENABLE, phychan->base + PL080_CH_CONFIG);
235 }
236 
237 /*
238  * Overall DMAC remains enabled always.
239  *
240  * Disabling individual channels could lose data.
241  *
242  * Disable the peripheral DMA after disabling the DMAC in order to allow
243  * the DMAC FIFO to drain, and hence allow the channel to show inactive
244  */
245 static void pl08x_pause_phy_chan(struct pl08x_phy_chan *ch)
246 {
247 	u32 val;
248 
249 	/* Set the HALT bit and wait for the FIFO to drain */
250 	val = readl(ch->base + PL080_CH_CONFIG);
251 	val |= PL080_CONFIG_HALT;
252 	writel(val, ch->base + PL080_CH_CONFIG);
253 
254 	/* Wait for channel inactive */
255 	while (pl08x_phy_channel_busy(ch))
256 		cpu_relax();
257 }
258 
259 static void pl08x_resume_phy_chan(struct pl08x_phy_chan *ch)
260 {
261 	u32 val;
262 
263 	/* Clear the HALT bit */
264 	val = readl(ch->base + PL080_CH_CONFIG);
265 	val &= ~PL080_CONFIG_HALT;
266 	writel(val, ch->base + PL080_CH_CONFIG);
267 }
268 
269 
270 /* Stops the channel */
271 static void pl08x_stop_phy_chan(struct pl08x_phy_chan *ch)
272 {
273 	u32 val;
274 
275 	pl08x_pause_phy_chan(ch);
276 
277 	/* Disable channel */
278 	val = readl(ch->base + PL080_CH_CONFIG);
279 	val &= ~PL080_CONFIG_ENABLE;
280 	val &= ~PL080_CONFIG_ERR_IRQ_MASK;
281 	val &= ~PL080_CONFIG_TC_IRQ_MASK;
282 	writel(val, ch->base + PL080_CH_CONFIG);
283 }
284 
285 static inline u32 get_bytes_in_cctl(u32 cctl)
286 {
287 	/* The source width defines the number of bytes */
288 	u32 bytes = cctl & PL080_CONTROL_TRANSFER_SIZE_MASK;
289 
290 	switch (cctl >> PL080_CONTROL_SWIDTH_SHIFT) {
291 	case PL080_WIDTH_8BIT:
292 		break;
293 	case PL080_WIDTH_16BIT:
294 		bytes *= 2;
295 		break;
296 	case PL080_WIDTH_32BIT:
297 		bytes *= 4;
298 		break;
299 	}
300 	return bytes;
301 }
302 
303 /* The channel should be paused when calling this */
304 static u32 pl08x_getbytes_chan(struct pl08x_dma_chan *plchan)
305 {
306 	struct pl08x_phy_chan *ch;
307 	struct pl08x_txd *txd;
308 	unsigned long flags;
309 	size_t bytes = 0;
310 
311 	spin_lock_irqsave(&plchan->lock, flags);
312 	ch = plchan->phychan;
313 	txd = plchan->at;
314 
315 	/*
316 	 * Follow the LLIs to get the number of remaining
317 	 * bytes in the currently active transaction.
318 	 */
319 	if (ch && txd) {
320 		u32 clli = readl(ch->base + PL080_CH_LLI) & ~PL080_LLI_LM_AHB2;
321 
322 		/* First get the remaining bytes in the active transfer */
323 		bytes = get_bytes_in_cctl(readl(ch->base + PL080_CH_CONTROL));
324 
325 		if (clli) {
326 			struct pl08x_lli *llis_va = txd->llis_va;
327 			dma_addr_t llis_bus = txd->llis_bus;
328 			int index;
329 
330 			BUG_ON(clli < llis_bus || clli >= llis_bus +
331 				sizeof(struct pl08x_lli) * MAX_NUM_TSFR_LLIS);
332 
333 			/*
334 			 * Locate the next LLI - as this is an array,
335 			 * it's simple maths to find.
336 			 */
337 			index = (clli - llis_bus) / sizeof(struct pl08x_lli);
338 
339 			for (; index < MAX_NUM_TSFR_LLIS; index++) {
340 				bytes += get_bytes_in_cctl(llis_va[index].cctl);
341 
342 				/*
343 				 * A LLI pointer of 0 terminates the LLI list
344 				 */
345 				if (!llis_va[index].lli)
346 					break;
347 			}
348 		}
349 	}
350 
351 	/* Sum up all queued transactions */
352 	if (!list_empty(&plchan->pend_list)) {
353 		struct pl08x_txd *txdi;
354 		list_for_each_entry(txdi, &plchan->pend_list, node) {
355 			bytes += txdi->len;
356 		}
357 	}
358 
359 	spin_unlock_irqrestore(&plchan->lock, flags);
360 
361 	return bytes;
362 }
363 
364 /*
365  * Allocate a physical channel for a virtual channel
366  *
367  * Try to locate a physical channel to be used for this transfer. If all
368  * are taken return NULL and the requester will have to cope by using
369  * some fallback PIO mode or retrying later.
370  */
371 static struct pl08x_phy_chan *
372 pl08x_get_phy_channel(struct pl08x_driver_data *pl08x,
373 		      struct pl08x_dma_chan *virt_chan)
374 {
375 	struct pl08x_phy_chan *ch = NULL;
376 	unsigned long flags;
377 	int i;
378 
379 	for (i = 0; i < pl08x->vd->channels; i++) {
380 		ch = &pl08x->phy_chans[i];
381 
382 		spin_lock_irqsave(&ch->lock, flags);
383 
384 		if (!ch->serving) {
385 			ch->serving = virt_chan;
386 			ch->signal = -1;
387 			spin_unlock_irqrestore(&ch->lock, flags);
388 			break;
389 		}
390 
391 		spin_unlock_irqrestore(&ch->lock, flags);
392 	}
393 
394 	if (i == pl08x->vd->channels) {
395 		/* No physical channel available, cope with it */
396 		return NULL;
397 	}
398 
399 	return ch;
400 }
401 
402 static inline void pl08x_put_phy_channel(struct pl08x_driver_data *pl08x,
403 					 struct pl08x_phy_chan *ch)
404 {
405 	unsigned long flags;
406 
407 	/* Stop the channel and clear its interrupts */
408 	pl08x_stop_phy_chan(ch);
409 	writel((1 << ch->id), pl08x->base + PL080_ERR_CLEAR);
410 	writel((1 << ch->id), pl08x->base + PL080_TC_CLEAR);
411 
412 	/* Mark it as free */
413 	spin_lock_irqsave(&ch->lock, flags);
414 	ch->serving = NULL;
415 	spin_unlock_irqrestore(&ch->lock, flags);
416 }
417 
418 /*
419  * LLI handling
420  */
421 
422 static inline unsigned int pl08x_get_bytes_for_cctl(unsigned int coded)
423 {
424 	switch (coded) {
425 	case PL080_WIDTH_8BIT:
426 		return 1;
427 	case PL080_WIDTH_16BIT:
428 		return 2;
429 	case PL080_WIDTH_32BIT:
430 		return 4;
431 	default:
432 		break;
433 	}
434 	BUG();
435 	return 0;
436 }
437 
438 static inline u32 pl08x_cctl_bits(u32 cctl, u8 srcwidth, u8 dstwidth,
439 				  size_t tsize)
440 {
441 	u32 retbits = cctl;
442 
443 	/* Remove all src, dst and transfer size bits */
444 	retbits &= ~PL080_CONTROL_DWIDTH_MASK;
445 	retbits &= ~PL080_CONTROL_SWIDTH_MASK;
446 	retbits &= ~PL080_CONTROL_TRANSFER_SIZE_MASK;
447 
448 	/* Then set the bits according to the parameters */
449 	switch (srcwidth) {
450 	case 1:
451 		retbits |= PL080_WIDTH_8BIT << PL080_CONTROL_SWIDTH_SHIFT;
452 		break;
453 	case 2:
454 		retbits |= PL080_WIDTH_16BIT << PL080_CONTROL_SWIDTH_SHIFT;
455 		break;
456 	case 4:
457 		retbits |= PL080_WIDTH_32BIT << PL080_CONTROL_SWIDTH_SHIFT;
458 		break;
459 	default:
460 		BUG();
461 		break;
462 	}
463 
464 	switch (dstwidth) {
465 	case 1:
466 		retbits |= PL080_WIDTH_8BIT << PL080_CONTROL_DWIDTH_SHIFT;
467 		break;
468 	case 2:
469 		retbits |= PL080_WIDTH_16BIT << PL080_CONTROL_DWIDTH_SHIFT;
470 		break;
471 	case 4:
472 		retbits |= PL080_WIDTH_32BIT << PL080_CONTROL_DWIDTH_SHIFT;
473 		break;
474 	default:
475 		BUG();
476 		break;
477 	}
478 
479 	retbits |= tsize << PL080_CONTROL_TRANSFER_SIZE_SHIFT;
480 	return retbits;
481 }
482 
483 struct pl08x_lli_build_data {
484 	struct pl08x_txd *txd;
485 	struct pl08x_driver_data *pl08x;
486 	struct pl08x_bus_data srcbus;
487 	struct pl08x_bus_data dstbus;
488 	size_t remainder;
489 };
490 
491 /*
492  * Autoselect a master bus to use for the transfer this prefers the
493  * destination bus if both available if fixed address on one bus the
494  * other will be chosen
495  */
496 static void pl08x_choose_master_bus(struct pl08x_lli_build_data *bd,
497 	struct pl08x_bus_data **mbus, struct pl08x_bus_data **sbus, u32 cctl)
498 {
499 	if (!(cctl & PL080_CONTROL_DST_INCR)) {
500 		*mbus = &bd->srcbus;
501 		*sbus = &bd->dstbus;
502 	} else if (!(cctl & PL080_CONTROL_SRC_INCR)) {
503 		*mbus = &bd->dstbus;
504 		*sbus = &bd->srcbus;
505 	} else {
506 		if (bd->dstbus.buswidth == 4) {
507 			*mbus = &bd->dstbus;
508 			*sbus = &bd->srcbus;
509 		} else if (bd->srcbus.buswidth == 4) {
510 			*mbus = &bd->srcbus;
511 			*sbus = &bd->dstbus;
512 		} else if (bd->dstbus.buswidth == 2) {
513 			*mbus = &bd->dstbus;
514 			*sbus = &bd->srcbus;
515 		} else if (bd->srcbus.buswidth == 2) {
516 			*mbus = &bd->srcbus;
517 			*sbus = &bd->dstbus;
518 		} else {
519 			/* bd->srcbus.buswidth == 1 */
520 			*mbus = &bd->dstbus;
521 			*sbus = &bd->srcbus;
522 		}
523 	}
524 }
525 
526 /*
527  * Fills in one LLI for a certain transfer descriptor and advance the counter
528  */
529 static void pl08x_fill_lli_for_desc(struct pl08x_lli_build_data *bd,
530 	int num_llis, int len, u32 cctl)
531 {
532 	struct pl08x_lli *llis_va = bd->txd->llis_va;
533 	dma_addr_t llis_bus = bd->txd->llis_bus;
534 
535 	BUG_ON(num_llis >= MAX_NUM_TSFR_LLIS);
536 
537 	llis_va[num_llis].cctl = cctl;
538 	llis_va[num_llis].src = bd->srcbus.addr;
539 	llis_va[num_llis].dst = bd->dstbus.addr;
540 	llis_va[num_llis].lli = llis_bus + (num_llis + 1) * sizeof(struct pl08x_lli);
541 	if (bd->pl08x->lli_buses & PL08X_AHB2)
542 		llis_va[num_llis].lli |= PL080_LLI_LM_AHB2;
543 
544 	if (cctl & PL080_CONTROL_SRC_INCR)
545 		bd->srcbus.addr += len;
546 	if (cctl & PL080_CONTROL_DST_INCR)
547 		bd->dstbus.addr += len;
548 
549 	BUG_ON(bd->remainder < len);
550 
551 	bd->remainder -= len;
552 }
553 
554 /*
555  * Return number of bytes to fill to boundary, or len.
556  * This calculation works for any value of addr.
557  */
558 static inline size_t pl08x_pre_boundary(u32 addr, size_t len)
559 {
560 	size_t boundary_len = PL08X_BOUNDARY_SIZE -
561 			(addr & (PL08X_BOUNDARY_SIZE - 1));
562 
563 	return min(boundary_len, len);
564 }
565 
566 /*
567  * This fills in the table of LLIs for the transfer descriptor
568  * Note that we assume we never have to change the burst sizes
569  * Return 0 for error
570  */
571 static int pl08x_fill_llis_for_desc(struct pl08x_driver_data *pl08x,
572 			      struct pl08x_txd *txd)
573 {
574 	struct pl08x_bus_data *mbus, *sbus;
575 	struct pl08x_lli_build_data bd;
576 	int num_llis = 0;
577 	u32 cctl;
578 	size_t max_bytes_per_lli;
579 	size_t total_bytes = 0;
580 	struct pl08x_lli *llis_va;
581 
582 	txd->llis_va = dma_pool_alloc(pl08x->pool, GFP_NOWAIT,
583 				      &txd->llis_bus);
584 	if (!txd->llis_va) {
585 		dev_err(&pl08x->adev->dev, "%s no memory for llis\n", __func__);
586 		return 0;
587 	}
588 
589 	pl08x->pool_ctr++;
590 
591 	/* Get the default CCTL */
592 	cctl = txd->cctl;
593 
594 	bd.txd = txd;
595 	bd.pl08x = pl08x;
596 	bd.srcbus.addr = txd->src_addr;
597 	bd.dstbus.addr = txd->dst_addr;
598 
599 	/* Find maximum width of the source bus */
600 	bd.srcbus.maxwidth =
601 		pl08x_get_bytes_for_cctl((cctl & PL080_CONTROL_SWIDTH_MASK) >>
602 				       PL080_CONTROL_SWIDTH_SHIFT);
603 
604 	/* Find maximum width of the destination bus */
605 	bd.dstbus.maxwidth =
606 		pl08x_get_bytes_for_cctl((cctl & PL080_CONTROL_DWIDTH_MASK) >>
607 				       PL080_CONTROL_DWIDTH_SHIFT);
608 
609 	/* Set up the bus widths to the maximum */
610 	bd.srcbus.buswidth = bd.srcbus.maxwidth;
611 	bd.dstbus.buswidth = bd.dstbus.maxwidth;
612 	dev_vdbg(&pl08x->adev->dev,
613 		 "%s source bus is %d bytes wide, dest bus is %d bytes wide\n",
614 		 __func__, bd.srcbus.buswidth, bd.dstbus.buswidth);
615 
616 
617 	/*
618 	 * Bytes transferred == tsize * MIN(buswidths), not max(buswidths)
619 	 */
620 	max_bytes_per_lli = min(bd.srcbus.buswidth, bd.dstbus.buswidth) *
621 		PL080_CONTROL_TRANSFER_SIZE_MASK;
622 	dev_vdbg(&pl08x->adev->dev,
623 		 "%s max bytes per lli = %zu\n",
624 		 __func__, max_bytes_per_lli);
625 
626 	/* We need to count this down to zero */
627 	bd.remainder = txd->len;
628 	dev_vdbg(&pl08x->adev->dev,
629 		 "%s remainder = %zu\n",
630 		 __func__, bd.remainder);
631 
632 	/*
633 	 * Choose bus to align to
634 	 * - prefers destination bus if both available
635 	 * - if fixed address on one bus chooses other
636 	 */
637 	pl08x_choose_master_bus(&bd, &mbus, &sbus, cctl);
638 
639 	if (txd->len < mbus->buswidth) {
640 		/* Less than a bus width available - send as single bytes */
641 		while (bd.remainder) {
642 			dev_vdbg(&pl08x->adev->dev,
643 				 "%s single byte LLIs for a transfer of "
644 				 "less than a bus width (remain 0x%08x)\n",
645 				 __func__, bd.remainder);
646 			cctl = pl08x_cctl_bits(cctl, 1, 1, 1);
647 			pl08x_fill_lli_for_desc(&bd, num_llis++, 1, cctl);
648 			total_bytes++;
649 		}
650 	} else {
651 		/* Make one byte LLIs until master bus is aligned */
652 		while ((mbus->addr) % (mbus->buswidth)) {
653 			dev_vdbg(&pl08x->adev->dev,
654 				"%s adjustment lli for less than bus width "
655 				 "(remain 0x%08x)\n",
656 				 __func__, bd.remainder);
657 			cctl = pl08x_cctl_bits(cctl, 1, 1, 1);
658 			pl08x_fill_lli_for_desc(&bd, num_llis++, 1, cctl);
659 			total_bytes++;
660 		}
661 
662 		/*
663 		 * Master now aligned
664 		 * - if slave is not then we must set its width down
665 		 */
666 		if (sbus->addr % sbus->buswidth) {
667 			dev_dbg(&pl08x->adev->dev,
668 				"%s set down bus width to one byte\n",
669 				 __func__);
670 
671 			sbus->buswidth = 1;
672 		}
673 
674 		/*
675 		 * Make largest possible LLIs until less than one bus
676 		 * width left
677 		 */
678 		while (bd.remainder > (mbus->buswidth - 1)) {
679 			size_t lli_len, target_len, tsize, odd_bytes;
680 
681 			/*
682 			 * If enough left try to send max possible,
683 			 * otherwise try to send the remainder
684 			 */
685 			target_len = min(bd.remainder, max_bytes_per_lli);
686 
687 			/*
688 			 * Set bus lengths for incrementing buses to the
689 			 * number of bytes which fill to next memory boundary,
690 			 * limiting on the target length calculated above.
691 			 */
692 			if (cctl & PL080_CONTROL_SRC_INCR)
693 				bd.srcbus.fill_bytes =
694 					pl08x_pre_boundary(bd.srcbus.addr,
695 						target_len);
696 			else
697 				bd.srcbus.fill_bytes = target_len;
698 
699 			if (cctl & PL080_CONTROL_DST_INCR)
700 				bd.dstbus.fill_bytes =
701 					pl08x_pre_boundary(bd.dstbus.addr,
702 						target_len);
703 			else
704 				bd.dstbus.fill_bytes = target_len;
705 
706 			/* Find the nearest */
707 			lli_len	= min(bd.srcbus.fill_bytes,
708 				      bd.dstbus.fill_bytes);
709 
710 			BUG_ON(lli_len > bd.remainder);
711 
712 			if (lli_len <= 0) {
713 				dev_err(&pl08x->adev->dev,
714 					"%s lli_len is %zu, <= 0\n",
715 						__func__, lli_len);
716 				return 0;
717 			}
718 
719 			if (lli_len == target_len) {
720 				/*
721 				 * Can send what we wanted.
722 				 * Maintain alignment
723 				 */
724 				lli_len	= (lli_len/mbus->buswidth) *
725 							mbus->buswidth;
726 				odd_bytes = 0;
727 			} else {
728 				/*
729 				 * So now we know how many bytes to transfer
730 				 * to get to the nearest boundary.  The next
731 				 * LLI will past the boundary.  However, we
732 				 * may be working to a boundary on the slave
733 				 * bus.  We need to ensure the master stays
734 				 * aligned, and that we are working in
735 				 * multiples of the bus widths.
736 				 */
737 				odd_bytes = lli_len % mbus->buswidth;
738 				lli_len -= odd_bytes;
739 
740 			}
741 
742 			if (lli_len) {
743 				/*
744 				 * Check against minimum bus alignment:
745 				 * Calculate actual transfer size in relation
746 				 * to bus width an get a maximum remainder of
747 				 * the smallest bus width - 1
748 				 */
749 				/* FIXME: use round_down()? */
750 				tsize = lli_len / min(mbus->buswidth,
751 						      sbus->buswidth);
752 				lli_len	= tsize * min(mbus->buswidth,
753 						      sbus->buswidth);
754 
755 				if (target_len != lli_len) {
756 					dev_vdbg(&pl08x->adev->dev,
757 					"%s can't send what we want. Desired 0x%08zx, lli of 0x%08zx bytes in txd of 0x%08zx\n",
758 					__func__, target_len, lli_len, txd->len);
759 				}
760 
761 				cctl = pl08x_cctl_bits(cctl,
762 						       bd.srcbus.buswidth,
763 						       bd.dstbus.buswidth,
764 						       tsize);
765 
766 				dev_vdbg(&pl08x->adev->dev,
767 					"%s fill lli with single lli chunk of size 0x%08zx (remainder 0x%08zx)\n",
768 					__func__, lli_len, bd.remainder);
769 				pl08x_fill_lli_for_desc(&bd, num_llis++,
770 					lli_len, cctl);
771 				total_bytes += lli_len;
772 			}
773 
774 
775 			if (odd_bytes) {
776 				/*
777 				 * Creep past the boundary, maintaining
778 				 * master alignment
779 				 */
780 				int j;
781 				for (j = 0; (j < mbus->buswidth)
782 						&& (bd.remainder); j++) {
783 					cctl = pl08x_cctl_bits(cctl, 1, 1, 1);
784 					dev_vdbg(&pl08x->adev->dev,
785 						"%s align with boundary, single byte (remain 0x%08zx)\n",
786 						__func__, bd.remainder);
787 					pl08x_fill_lli_for_desc(&bd,
788 						num_llis++, 1, cctl);
789 					total_bytes++;
790 				}
791 			}
792 		}
793 
794 		/*
795 		 * Send any odd bytes
796 		 */
797 		while (bd.remainder) {
798 			cctl = pl08x_cctl_bits(cctl, 1, 1, 1);
799 			dev_vdbg(&pl08x->adev->dev,
800 				"%s align with boundary, single odd byte (remain %zu)\n",
801 				__func__, bd.remainder);
802 			pl08x_fill_lli_for_desc(&bd, num_llis++, 1, cctl);
803 			total_bytes++;
804 		}
805 	}
806 	if (total_bytes != txd->len) {
807 		dev_err(&pl08x->adev->dev,
808 			"%s size of encoded lli:s don't match total txd, transferred 0x%08zx from size 0x%08zx\n",
809 			__func__, total_bytes, txd->len);
810 		return 0;
811 	}
812 
813 	if (num_llis >= MAX_NUM_TSFR_LLIS) {
814 		dev_err(&pl08x->adev->dev,
815 			"%s need to increase MAX_NUM_TSFR_LLIS from 0x%08x\n",
816 			__func__, (u32) MAX_NUM_TSFR_LLIS);
817 		return 0;
818 	}
819 
820 	llis_va = txd->llis_va;
821 	/* The final LLI terminates the LLI. */
822 	llis_va[num_llis - 1].lli = 0;
823 	/* The final LLI element shall also fire an interrupt. */
824 	llis_va[num_llis - 1].cctl |= PL080_CONTROL_TC_IRQ_EN;
825 
826 #ifdef VERBOSE_DEBUG
827 	{
828 		int i;
829 
830 		for (i = 0; i < num_llis; i++) {
831 			dev_vdbg(&pl08x->adev->dev,
832 				 "lli %d @%p: csrc=0x%08x, cdst=0x%08x, cctl=0x%08x, clli=0x%08x\n",
833 				 i,
834 				 &llis_va[i],
835 				 llis_va[i].src,
836 				 llis_va[i].dst,
837 				 llis_va[i].cctl,
838 				 llis_va[i].lli
839 				);
840 		}
841 	}
842 #endif
843 
844 	return num_llis;
845 }
846 
847 /* You should call this with the struct pl08x lock held */
848 static void pl08x_free_txd(struct pl08x_driver_data *pl08x,
849 			   struct pl08x_txd *txd)
850 {
851 	/* Free the LLI */
852 	dma_pool_free(pl08x->pool, txd->llis_va, txd->llis_bus);
853 
854 	pl08x->pool_ctr--;
855 
856 	kfree(txd);
857 }
858 
859 static void pl08x_free_txd_list(struct pl08x_driver_data *pl08x,
860 				struct pl08x_dma_chan *plchan)
861 {
862 	struct pl08x_txd *txdi = NULL;
863 	struct pl08x_txd *next;
864 
865 	if (!list_empty(&plchan->pend_list)) {
866 		list_for_each_entry_safe(txdi,
867 					 next, &plchan->pend_list, node) {
868 			list_del(&txdi->node);
869 			pl08x_free_txd(pl08x, txdi);
870 		}
871 	}
872 }
873 
874 /*
875  * The DMA ENGINE API
876  */
877 static int pl08x_alloc_chan_resources(struct dma_chan *chan)
878 {
879 	return 0;
880 }
881 
882 static void pl08x_free_chan_resources(struct dma_chan *chan)
883 {
884 }
885 
886 /*
887  * This should be called with the channel plchan->lock held
888  */
889 static int prep_phy_channel(struct pl08x_dma_chan *plchan,
890 			    struct pl08x_txd *txd)
891 {
892 	struct pl08x_driver_data *pl08x = plchan->host;
893 	struct pl08x_phy_chan *ch;
894 	int ret;
895 
896 	/* Check if we already have a channel */
897 	if (plchan->phychan)
898 		return 0;
899 
900 	ch = pl08x_get_phy_channel(pl08x, plchan);
901 	if (!ch) {
902 		/* No physical channel available, cope with it */
903 		dev_dbg(&pl08x->adev->dev, "no physical channel available for xfer on %s\n", plchan->name);
904 		return -EBUSY;
905 	}
906 
907 	/*
908 	 * OK we have a physical channel: for memcpy() this is all we
909 	 * need, but for slaves the physical signals may be muxed!
910 	 * Can the platform allow us to use this channel?
911 	 */
912 	if (plchan->slave &&
913 	    ch->signal < 0 &&
914 	    pl08x->pd->get_signal) {
915 		ret = pl08x->pd->get_signal(plchan);
916 		if (ret < 0) {
917 			dev_dbg(&pl08x->adev->dev,
918 				"unable to use physical channel %d for transfer on %s due to platform restrictions\n",
919 				ch->id, plchan->name);
920 			/* Release physical channel & return */
921 			pl08x_put_phy_channel(pl08x, ch);
922 			return -EBUSY;
923 		}
924 		ch->signal = ret;
925 
926 		/* Assign the flow control signal to this channel */
927 		if (txd->direction == DMA_TO_DEVICE)
928 			txd->ccfg |= ch->signal << PL080_CONFIG_DST_SEL_SHIFT;
929 		else if (txd->direction == DMA_FROM_DEVICE)
930 			txd->ccfg |= ch->signal << PL080_CONFIG_SRC_SEL_SHIFT;
931 	}
932 
933 	dev_dbg(&pl08x->adev->dev, "allocated physical channel %d and signal %d for xfer on %s\n",
934 		 ch->id,
935 		 ch->signal,
936 		 plchan->name);
937 
938 	plchan->phychan_hold++;
939 	plchan->phychan = ch;
940 
941 	return 0;
942 }
943 
944 static void release_phy_channel(struct pl08x_dma_chan *plchan)
945 {
946 	struct pl08x_driver_data *pl08x = plchan->host;
947 
948 	if ((plchan->phychan->signal >= 0) && pl08x->pd->put_signal) {
949 		pl08x->pd->put_signal(plchan);
950 		plchan->phychan->signal = -1;
951 	}
952 	pl08x_put_phy_channel(pl08x, plchan->phychan);
953 	plchan->phychan = NULL;
954 }
955 
956 static dma_cookie_t pl08x_tx_submit(struct dma_async_tx_descriptor *tx)
957 {
958 	struct pl08x_dma_chan *plchan = to_pl08x_chan(tx->chan);
959 	struct pl08x_txd *txd = to_pl08x_txd(tx);
960 	unsigned long flags;
961 
962 	spin_lock_irqsave(&plchan->lock, flags);
963 
964 	plchan->chan.cookie += 1;
965 	if (plchan->chan.cookie < 0)
966 		plchan->chan.cookie = 1;
967 	tx->cookie = plchan->chan.cookie;
968 
969 	/* Put this onto the pending list */
970 	list_add_tail(&txd->node, &plchan->pend_list);
971 
972 	/*
973 	 * If there was no physical channel available for this memcpy,
974 	 * stack the request up and indicate that the channel is waiting
975 	 * for a free physical channel.
976 	 */
977 	if (!plchan->slave && !plchan->phychan) {
978 		/* Do this memcpy whenever there is a channel ready */
979 		plchan->state = PL08X_CHAN_WAITING;
980 		plchan->waiting = txd;
981 	} else {
982 		plchan->phychan_hold--;
983 	}
984 
985 	spin_unlock_irqrestore(&plchan->lock, flags);
986 
987 	return tx->cookie;
988 }
989 
990 static struct dma_async_tx_descriptor *pl08x_prep_dma_interrupt(
991 		struct dma_chan *chan, unsigned long flags)
992 {
993 	struct dma_async_tx_descriptor *retval = NULL;
994 
995 	return retval;
996 }
997 
998 /*
999  * Code accessing dma_async_is_complete() in a tight loop may give problems.
1000  * If slaves are relying on interrupts to signal completion this function
1001  * must not be called with interrupts disabled.
1002  */
1003 static enum dma_status
1004 pl08x_dma_tx_status(struct dma_chan *chan,
1005 		    dma_cookie_t cookie,
1006 		    struct dma_tx_state *txstate)
1007 {
1008 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1009 	dma_cookie_t last_used;
1010 	dma_cookie_t last_complete;
1011 	enum dma_status ret;
1012 	u32 bytesleft = 0;
1013 
1014 	last_used = plchan->chan.cookie;
1015 	last_complete = plchan->lc;
1016 
1017 	ret = dma_async_is_complete(cookie, last_complete, last_used);
1018 	if (ret == DMA_SUCCESS) {
1019 		dma_set_tx_state(txstate, last_complete, last_used, 0);
1020 		return ret;
1021 	}
1022 
1023 	/*
1024 	 * This cookie not complete yet
1025 	 */
1026 	last_used = plchan->chan.cookie;
1027 	last_complete = plchan->lc;
1028 
1029 	/* Get number of bytes left in the active transactions and queue */
1030 	bytesleft = pl08x_getbytes_chan(plchan);
1031 
1032 	dma_set_tx_state(txstate, last_complete, last_used,
1033 			 bytesleft);
1034 
1035 	if (plchan->state == PL08X_CHAN_PAUSED)
1036 		return DMA_PAUSED;
1037 
1038 	/* Whether waiting or running, we're in progress */
1039 	return DMA_IN_PROGRESS;
1040 }
1041 
1042 /* PrimeCell DMA extension */
1043 struct burst_table {
1044 	int burstwords;
1045 	u32 reg;
1046 };
1047 
1048 static const struct burst_table burst_sizes[] = {
1049 	{
1050 		.burstwords = 256,
1051 		.reg = (PL080_BSIZE_256 << PL080_CONTROL_SB_SIZE_SHIFT) |
1052 			(PL080_BSIZE_256 << PL080_CONTROL_DB_SIZE_SHIFT),
1053 	},
1054 	{
1055 		.burstwords = 128,
1056 		.reg = (PL080_BSIZE_128 << PL080_CONTROL_SB_SIZE_SHIFT) |
1057 			(PL080_BSIZE_128 << PL080_CONTROL_DB_SIZE_SHIFT),
1058 	},
1059 	{
1060 		.burstwords = 64,
1061 		.reg = (PL080_BSIZE_64 << PL080_CONTROL_SB_SIZE_SHIFT) |
1062 			(PL080_BSIZE_64 << PL080_CONTROL_DB_SIZE_SHIFT),
1063 	},
1064 	{
1065 		.burstwords = 32,
1066 		.reg = (PL080_BSIZE_32 << PL080_CONTROL_SB_SIZE_SHIFT) |
1067 			(PL080_BSIZE_32 << PL080_CONTROL_DB_SIZE_SHIFT),
1068 	},
1069 	{
1070 		.burstwords = 16,
1071 		.reg = (PL080_BSIZE_16 << PL080_CONTROL_SB_SIZE_SHIFT) |
1072 			(PL080_BSIZE_16 << PL080_CONTROL_DB_SIZE_SHIFT),
1073 	},
1074 	{
1075 		.burstwords = 8,
1076 		.reg = (PL080_BSIZE_8 << PL080_CONTROL_SB_SIZE_SHIFT) |
1077 			(PL080_BSIZE_8 << PL080_CONTROL_DB_SIZE_SHIFT),
1078 	},
1079 	{
1080 		.burstwords = 4,
1081 		.reg = (PL080_BSIZE_4 << PL080_CONTROL_SB_SIZE_SHIFT) |
1082 			(PL080_BSIZE_4 << PL080_CONTROL_DB_SIZE_SHIFT),
1083 	},
1084 	{
1085 		.burstwords = 1,
1086 		.reg = (PL080_BSIZE_1 << PL080_CONTROL_SB_SIZE_SHIFT) |
1087 			(PL080_BSIZE_1 << PL080_CONTROL_DB_SIZE_SHIFT),
1088 	},
1089 };
1090 
1091 static int dma_set_runtime_config(struct dma_chan *chan,
1092 				  struct dma_slave_config *config)
1093 {
1094 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1095 	struct pl08x_driver_data *pl08x = plchan->host;
1096 	struct pl08x_channel_data *cd = plchan->cd;
1097 	enum dma_slave_buswidth addr_width;
1098 	dma_addr_t addr;
1099 	u32 maxburst;
1100 	u32 cctl = 0;
1101 	int i;
1102 
1103 	if (!plchan->slave)
1104 		return -EINVAL;
1105 
1106 	/* Transfer direction */
1107 	plchan->runtime_direction = config->direction;
1108 	if (config->direction == DMA_TO_DEVICE) {
1109 		addr = config->dst_addr;
1110 		addr_width = config->dst_addr_width;
1111 		maxburst = config->dst_maxburst;
1112 	} else if (config->direction == DMA_FROM_DEVICE) {
1113 		addr = config->src_addr;
1114 		addr_width = config->src_addr_width;
1115 		maxburst = config->src_maxburst;
1116 	} else {
1117 		dev_err(&pl08x->adev->dev,
1118 			"bad runtime_config: alien transfer direction\n");
1119 		return -EINVAL;
1120 	}
1121 
1122 	switch (addr_width) {
1123 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
1124 		cctl |= (PL080_WIDTH_8BIT << PL080_CONTROL_SWIDTH_SHIFT) |
1125 			(PL080_WIDTH_8BIT << PL080_CONTROL_DWIDTH_SHIFT);
1126 		break;
1127 	case DMA_SLAVE_BUSWIDTH_2_BYTES:
1128 		cctl |= (PL080_WIDTH_16BIT << PL080_CONTROL_SWIDTH_SHIFT) |
1129 			(PL080_WIDTH_16BIT << PL080_CONTROL_DWIDTH_SHIFT);
1130 		break;
1131 	case DMA_SLAVE_BUSWIDTH_4_BYTES:
1132 		cctl |= (PL080_WIDTH_32BIT << PL080_CONTROL_SWIDTH_SHIFT) |
1133 			(PL080_WIDTH_32BIT << PL080_CONTROL_DWIDTH_SHIFT);
1134 		break;
1135 	default:
1136 		dev_err(&pl08x->adev->dev,
1137 			"bad runtime_config: alien address width\n");
1138 		return -EINVAL;
1139 	}
1140 
1141 	/*
1142 	 * Now decide on a maxburst:
1143 	 * If this channel will only request single transfers, set this
1144 	 * down to ONE element.  Also select one element if no maxburst
1145 	 * is specified.
1146 	 */
1147 	if (plchan->cd->single || maxburst == 0) {
1148 		cctl |= (PL080_BSIZE_1 << PL080_CONTROL_SB_SIZE_SHIFT) |
1149 			(PL080_BSIZE_1 << PL080_CONTROL_DB_SIZE_SHIFT);
1150 	} else {
1151 		for (i = 0; i < ARRAY_SIZE(burst_sizes); i++)
1152 			if (burst_sizes[i].burstwords <= maxburst)
1153 				break;
1154 		cctl |= burst_sizes[i].reg;
1155 	}
1156 
1157 	plchan->runtime_addr = addr;
1158 
1159 	/* Modify the default channel data to fit PrimeCell request */
1160 	cd->cctl = cctl;
1161 
1162 	dev_dbg(&pl08x->adev->dev,
1163 		"configured channel %s (%s) for %s, data width %d, "
1164 		"maxburst %d words, LE, CCTL=0x%08x\n",
1165 		dma_chan_name(chan), plchan->name,
1166 		(config->direction == DMA_FROM_DEVICE) ? "RX" : "TX",
1167 		addr_width,
1168 		maxburst,
1169 		cctl);
1170 
1171 	return 0;
1172 }
1173 
1174 /*
1175  * Slave transactions callback to the slave device to allow
1176  * synchronization of slave DMA signals with the DMAC enable
1177  */
1178 static void pl08x_issue_pending(struct dma_chan *chan)
1179 {
1180 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1181 	unsigned long flags;
1182 
1183 	spin_lock_irqsave(&plchan->lock, flags);
1184 	/* Something is already active, or we're waiting for a channel... */
1185 	if (plchan->at || plchan->state == PL08X_CHAN_WAITING) {
1186 		spin_unlock_irqrestore(&plchan->lock, flags);
1187 		return;
1188 	}
1189 
1190 	/* Take the first element in the queue and execute it */
1191 	if (!list_empty(&plchan->pend_list)) {
1192 		struct pl08x_txd *next;
1193 
1194 		next = list_first_entry(&plchan->pend_list,
1195 					struct pl08x_txd,
1196 					node);
1197 		list_del(&next->node);
1198 		plchan->state = PL08X_CHAN_RUNNING;
1199 
1200 		pl08x_start_txd(plchan, next);
1201 	}
1202 
1203 	spin_unlock_irqrestore(&plchan->lock, flags);
1204 }
1205 
1206 static int pl08x_prep_channel_resources(struct pl08x_dma_chan *plchan,
1207 					struct pl08x_txd *txd)
1208 {
1209 	struct pl08x_driver_data *pl08x = plchan->host;
1210 	unsigned long flags;
1211 	int num_llis, ret;
1212 
1213 	num_llis = pl08x_fill_llis_for_desc(pl08x, txd);
1214 	if (!num_llis) {
1215 		kfree(txd);
1216 		return -EINVAL;
1217 	}
1218 
1219 	spin_lock_irqsave(&plchan->lock, flags);
1220 
1221 	/*
1222 	 * See if we already have a physical channel allocated,
1223 	 * else this is the time to try to get one.
1224 	 */
1225 	ret = prep_phy_channel(plchan, txd);
1226 	if (ret) {
1227 		/*
1228 		 * No physical channel was available.
1229 		 *
1230 		 * memcpy transfers can be sorted out at submission time.
1231 		 *
1232 		 * Slave transfers may have been denied due to platform
1233 		 * channel muxing restrictions.  Since there is no guarantee
1234 		 * that this will ever be resolved, and the signal must be
1235 		 * acquired AFTER acquiring the physical channel, we will let
1236 		 * them be NACK:ed with -EBUSY here. The drivers can retry
1237 		 * the prep() call if they are eager on doing this using DMA.
1238 		 */
1239 		if (plchan->slave) {
1240 			pl08x_free_txd_list(pl08x, plchan);
1241 			pl08x_free_txd(pl08x, txd);
1242 			spin_unlock_irqrestore(&plchan->lock, flags);
1243 			return -EBUSY;
1244 		}
1245 	} else
1246 		/*
1247 		 * Else we're all set, paused and ready to roll, status
1248 		 * will switch to PL08X_CHAN_RUNNING when we call
1249 		 * issue_pending(). If there is something running on the
1250 		 * channel already we don't change its state.
1251 		 */
1252 		if (plchan->state == PL08X_CHAN_IDLE)
1253 			plchan->state = PL08X_CHAN_PAUSED;
1254 
1255 	spin_unlock_irqrestore(&plchan->lock, flags);
1256 
1257 	return 0;
1258 }
1259 
1260 /*
1261  * Given the source and destination available bus masks, select which
1262  * will be routed to each port.  We try to have source and destination
1263  * on separate ports, but always respect the allowable settings.
1264  */
1265 static u32 pl08x_select_bus(struct pl08x_driver_data *pl08x, u8 src, u8 dst)
1266 {
1267 	u32 cctl = 0;
1268 
1269 	if (!(dst & PL08X_AHB1) || ((dst & PL08X_AHB2) && (src & PL08X_AHB1)))
1270 		cctl |= PL080_CONTROL_DST_AHB2;
1271 	if (!(src & PL08X_AHB1) || ((src & PL08X_AHB2) && !(dst & PL08X_AHB2)))
1272 		cctl |= PL080_CONTROL_SRC_AHB2;
1273 
1274 	return cctl;
1275 }
1276 
1277 static struct pl08x_txd *pl08x_get_txd(struct pl08x_dma_chan *plchan,
1278 	unsigned long flags)
1279 {
1280 	struct pl08x_txd *txd = kzalloc(sizeof(struct pl08x_txd), GFP_NOWAIT);
1281 
1282 	if (txd) {
1283 		dma_async_tx_descriptor_init(&txd->tx, &plchan->chan);
1284 		txd->tx.flags = flags;
1285 		txd->tx.tx_submit = pl08x_tx_submit;
1286 		INIT_LIST_HEAD(&txd->node);
1287 
1288 		/* Always enable error and terminal interrupts */
1289 		txd->ccfg = PL080_CONFIG_ERR_IRQ_MASK |
1290 			    PL080_CONFIG_TC_IRQ_MASK;
1291 	}
1292 	return txd;
1293 }
1294 
1295 /*
1296  * Initialize a descriptor to be used by memcpy submit
1297  */
1298 static struct dma_async_tx_descriptor *pl08x_prep_dma_memcpy(
1299 		struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
1300 		size_t len, unsigned long flags)
1301 {
1302 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1303 	struct pl08x_driver_data *pl08x = plchan->host;
1304 	struct pl08x_txd *txd;
1305 	int ret;
1306 
1307 	txd = pl08x_get_txd(plchan, flags);
1308 	if (!txd) {
1309 		dev_err(&pl08x->adev->dev,
1310 			"%s no memory for descriptor\n", __func__);
1311 		return NULL;
1312 	}
1313 
1314 	txd->direction = DMA_NONE;
1315 	txd->src_addr = src;
1316 	txd->dst_addr = dest;
1317 	txd->len = len;
1318 
1319 	/* Set platform data for m2m */
1320 	txd->ccfg |= PL080_FLOW_MEM2MEM << PL080_CONFIG_FLOW_CONTROL_SHIFT;
1321 	txd->cctl = pl08x->pd->memcpy_channel.cctl &
1322 			~(PL080_CONTROL_DST_AHB2 | PL080_CONTROL_SRC_AHB2);
1323 
1324 	/* Both to be incremented or the code will break */
1325 	txd->cctl |= PL080_CONTROL_SRC_INCR | PL080_CONTROL_DST_INCR;
1326 
1327 	if (pl08x->vd->dualmaster)
1328 		txd->cctl |= pl08x_select_bus(pl08x,
1329 					pl08x->mem_buses, pl08x->mem_buses);
1330 
1331 	ret = pl08x_prep_channel_resources(plchan, txd);
1332 	if (ret)
1333 		return NULL;
1334 
1335 	return &txd->tx;
1336 }
1337 
1338 static struct dma_async_tx_descriptor *pl08x_prep_slave_sg(
1339 		struct dma_chan *chan, struct scatterlist *sgl,
1340 		unsigned int sg_len, enum dma_data_direction direction,
1341 		unsigned long flags)
1342 {
1343 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1344 	struct pl08x_driver_data *pl08x = plchan->host;
1345 	struct pl08x_txd *txd;
1346 	u8 src_buses, dst_buses;
1347 	int ret;
1348 
1349 	/*
1350 	 * Current implementation ASSUMES only one sg
1351 	 */
1352 	if (sg_len != 1) {
1353 		dev_err(&pl08x->adev->dev, "%s prepared too long sglist\n",
1354 			__func__);
1355 		BUG();
1356 	}
1357 
1358 	dev_dbg(&pl08x->adev->dev, "%s prepare transaction of %d bytes from %s\n",
1359 		__func__, sgl->length, plchan->name);
1360 
1361 	txd = pl08x_get_txd(plchan, flags);
1362 	if (!txd) {
1363 		dev_err(&pl08x->adev->dev, "%s no txd\n", __func__);
1364 		return NULL;
1365 	}
1366 
1367 	if (direction != plchan->runtime_direction)
1368 		dev_err(&pl08x->adev->dev, "%s DMA setup does not match "
1369 			"the direction configured for the PrimeCell\n",
1370 			__func__);
1371 
1372 	/*
1373 	 * Set up addresses, the PrimeCell configured address
1374 	 * will take precedence since this may configure the
1375 	 * channel target address dynamically at runtime.
1376 	 */
1377 	txd->direction = direction;
1378 	txd->len = sgl->length;
1379 
1380 	txd->cctl = plchan->cd->cctl &
1381 			~(PL080_CONTROL_SRC_AHB2 | PL080_CONTROL_DST_AHB2 |
1382 			  PL080_CONTROL_SRC_INCR | PL080_CONTROL_DST_INCR |
1383 			  PL080_CONTROL_PROT_MASK);
1384 
1385 	/* Access the cell in privileged mode, non-bufferable, non-cacheable */
1386 	txd->cctl |= PL080_CONTROL_PROT_SYS;
1387 
1388 	if (direction == DMA_TO_DEVICE) {
1389 		txd->ccfg |= PL080_FLOW_MEM2PER << PL080_CONFIG_FLOW_CONTROL_SHIFT;
1390 		txd->cctl |= PL080_CONTROL_SRC_INCR;
1391 		txd->src_addr = sgl->dma_address;
1392 		if (plchan->runtime_addr)
1393 			txd->dst_addr = plchan->runtime_addr;
1394 		else
1395 			txd->dst_addr = plchan->cd->addr;
1396 		src_buses = pl08x->mem_buses;
1397 		dst_buses = plchan->cd->periph_buses;
1398 	} else if (direction == DMA_FROM_DEVICE) {
1399 		txd->ccfg |= PL080_FLOW_PER2MEM << PL080_CONFIG_FLOW_CONTROL_SHIFT;
1400 		txd->cctl |= PL080_CONTROL_DST_INCR;
1401 		if (plchan->runtime_addr)
1402 			txd->src_addr = plchan->runtime_addr;
1403 		else
1404 			txd->src_addr = plchan->cd->addr;
1405 		txd->dst_addr = sgl->dma_address;
1406 		src_buses = plchan->cd->periph_buses;
1407 		dst_buses = pl08x->mem_buses;
1408 	} else {
1409 		dev_err(&pl08x->adev->dev,
1410 			"%s direction unsupported\n", __func__);
1411 		return NULL;
1412 	}
1413 
1414 	txd->cctl |= pl08x_select_bus(pl08x, src_buses, dst_buses);
1415 
1416 	ret = pl08x_prep_channel_resources(plchan, txd);
1417 	if (ret)
1418 		return NULL;
1419 
1420 	return &txd->tx;
1421 }
1422 
1423 static int pl08x_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
1424 			 unsigned long arg)
1425 {
1426 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1427 	struct pl08x_driver_data *pl08x = plchan->host;
1428 	unsigned long flags;
1429 	int ret = 0;
1430 
1431 	/* Controls applicable to inactive channels */
1432 	if (cmd == DMA_SLAVE_CONFIG) {
1433 		return dma_set_runtime_config(chan,
1434 					      (struct dma_slave_config *)arg);
1435 	}
1436 
1437 	/*
1438 	 * Anything succeeds on channels with no physical allocation and
1439 	 * no queued transfers.
1440 	 */
1441 	spin_lock_irqsave(&plchan->lock, flags);
1442 	if (!plchan->phychan && !plchan->at) {
1443 		spin_unlock_irqrestore(&plchan->lock, flags);
1444 		return 0;
1445 	}
1446 
1447 	switch (cmd) {
1448 	case DMA_TERMINATE_ALL:
1449 		plchan->state = PL08X_CHAN_IDLE;
1450 
1451 		if (plchan->phychan) {
1452 			pl08x_stop_phy_chan(plchan->phychan);
1453 
1454 			/*
1455 			 * Mark physical channel as free and free any slave
1456 			 * signal
1457 			 */
1458 			release_phy_channel(plchan);
1459 		}
1460 		/* Dequeue jobs and free LLIs */
1461 		if (plchan->at) {
1462 			pl08x_free_txd(pl08x, plchan->at);
1463 			plchan->at = NULL;
1464 		}
1465 		/* Dequeue jobs not yet fired as well */
1466 		pl08x_free_txd_list(pl08x, plchan);
1467 		break;
1468 	case DMA_PAUSE:
1469 		pl08x_pause_phy_chan(plchan->phychan);
1470 		plchan->state = PL08X_CHAN_PAUSED;
1471 		break;
1472 	case DMA_RESUME:
1473 		pl08x_resume_phy_chan(plchan->phychan);
1474 		plchan->state = PL08X_CHAN_RUNNING;
1475 		break;
1476 	default:
1477 		/* Unknown command */
1478 		ret = -ENXIO;
1479 		break;
1480 	}
1481 
1482 	spin_unlock_irqrestore(&plchan->lock, flags);
1483 
1484 	return ret;
1485 }
1486 
1487 bool pl08x_filter_id(struct dma_chan *chan, void *chan_id)
1488 {
1489 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1490 	char *name = chan_id;
1491 
1492 	/* Check that the channel is not taken! */
1493 	if (!strcmp(plchan->name, name))
1494 		return true;
1495 
1496 	return false;
1497 }
1498 
1499 /*
1500  * Just check that the device is there and active
1501  * TODO: turn this bit on/off depending on the number of physical channels
1502  * actually used, if it is zero... well shut it off. That will save some
1503  * power. Cut the clock at the same time.
1504  */
1505 static void pl08x_ensure_on(struct pl08x_driver_data *pl08x)
1506 {
1507 	u32 val;
1508 
1509 	val = readl(pl08x->base + PL080_CONFIG);
1510 	val &= ~(PL080_CONFIG_M2_BE | PL080_CONFIG_M1_BE | PL080_CONFIG_ENABLE);
1511 	/* We implicitly clear bit 1 and that means little-endian mode */
1512 	val |= PL080_CONFIG_ENABLE;
1513 	writel(val, pl08x->base + PL080_CONFIG);
1514 }
1515 
1516 static void pl08x_unmap_buffers(struct pl08x_txd *txd)
1517 {
1518 	struct device *dev = txd->tx.chan->device->dev;
1519 
1520 	if (!(txd->tx.flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
1521 		if (txd->tx.flags & DMA_COMPL_SRC_UNMAP_SINGLE)
1522 			dma_unmap_single(dev, txd->src_addr, txd->len,
1523 				DMA_TO_DEVICE);
1524 		else
1525 			dma_unmap_page(dev, txd->src_addr, txd->len,
1526 				DMA_TO_DEVICE);
1527 	}
1528 	if (!(txd->tx.flags & DMA_COMPL_SKIP_DEST_UNMAP)) {
1529 		if (txd->tx.flags & DMA_COMPL_DEST_UNMAP_SINGLE)
1530 			dma_unmap_single(dev, txd->dst_addr, txd->len,
1531 				DMA_FROM_DEVICE);
1532 		else
1533 			dma_unmap_page(dev, txd->dst_addr, txd->len,
1534 				DMA_FROM_DEVICE);
1535 	}
1536 }
1537 
1538 static void pl08x_tasklet(unsigned long data)
1539 {
1540 	struct pl08x_dma_chan *plchan = (struct pl08x_dma_chan *) data;
1541 	struct pl08x_driver_data *pl08x = plchan->host;
1542 	struct pl08x_txd *txd;
1543 	unsigned long flags;
1544 
1545 	spin_lock_irqsave(&plchan->lock, flags);
1546 
1547 	txd = plchan->at;
1548 	plchan->at = NULL;
1549 
1550 	if (txd) {
1551 		/* Update last completed */
1552 		plchan->lc = txd->tx.cookie;
1553 	}
1554 
1555 	/* If a new descriptor is queued, set it up plchan->at is NULL here */
1556 	if (!list_empty(&plchan->pend_list)) {
1557 		struct pl08x_txd *next;
1558 
1559 		next = list_first_entry(&plchan->pend_list,
1560 					struct pl08x_txd,
1561 					node);
1562 		list_del(&next->node);
1563 
1564 		pl08x_start_txd(plchan, next);
1565 	} else if (plchan->phychan_hold) {
1566 		/*
1567 		 * This channel is still in use - we have a new txd being
1568 		 * prepared and will soon be queued.  Don't give up the
1569 		 * physical channel.
1570 		 */
1571 	} else {
1572 		struct pl08x_dma_chan *waiting = NULL;
1573 
1574 		/*
1575 		 * No more jobs, so free up the physical channel
1576 		 * Free any allocated signal on slave transfers too
1577 		 */
1578 		release_phy_channel(plchan);
1579 		plchan->state = PL08X_CHAN_IDLE;
1580 
1581 		/*
1582 		 * And NOW before anyone else can grab that free:d up
1583 		 * physical channel, see if there is some memcpy pending
1584 		 * that seriously needs to start because of being stacked
1585 		 * up while we were choking the physical channels with data.
1586 		 */
1587 		list_for_each_entry(waiting, &pl08x->memcpy.channels,
1588 				    chan.device_node) {
1589 		  if (waiting->state == PL08X_CHAN_WAITING &&
1590 			    waiting->waiting != NULL) {
1591 				int ret;
1592 
1593 				/* This should REALLY not fail now */
1594 				ret = prep_phy_channel(waiting,
1595 						       waiting->waiting);
1596 				BUG_ON(ret);
1597 				waiting->phychan_hold--;
1598 				waiting->state = PL08X_CHAN_RUNNING;
1599 				waiting->waiting = NULL;
1600 				pl08x_issue_pending(&waiting->chan);
1601 				break;
1602 			}
1603 		}
1604 	}
1605 
1606 	spin_unlock_irqrestore(&plchan->lock, flags);
1607 
1608 	if (txd) {
1609 		dma_async_tx_callback callback = txd->tx.callback;
1610 		void *callback_param = txd->tx.callback_param;
1611 
1612 		/* Don't try to unmap buffers on slave channels */
1613 		if (!plchan->slave)
1614 			pl08x_unmap_buffers(txd);
1615 
1616 		/* Free the descriptor */
1617 		spin_lock_irqsave(&plchan->lock, flags);
1618 		pl08x_free_txd(pl08x, txd);
1619 		spin_unlock_irqrestore(&plchan->lock, flags);
1620 
1621 		/* Callback to signal completion */
1622 		if (callback)
1623 			callback(callback_param);
1624 	}
1625 }
1626 
1627 static irqreturn_t pl08x_irq(int irq, void *dev)
1628 {
1629 	struct pl08x_driver_data *pl08x = dev;
1630 	u32 mask = 0;
1631 	u32 val;
1632 	int i;
1633 
1634 	val = readl(pl08x->base + PL080_ERR_STATUS);
1635 	if (val) {
1636 		/* An error interrupt (on one or more channels) */
1637 		dev_err(&pl08x->adev->dev,
1638 			"%s error interrupt, register value 0x%08x\n",
1639 				__func__, val);
1640 		/*
1641 		 * Simply clear ALL PL08X error interrupts,
1642 		 * regardless of channel and cause
1643 		 * FIXME: should be 0x00000003 on PL081 really.
1644 		 */
1645 		writel(0x000000FF, pl08x->base + PL080_ERR_CLEAR);
1646 	}
1647 	val = readl(pl08x->base + PL080_INT_STATUS);
1648 	for (i = 0; i < pl08x->vd->channels; i++) {
1649 		if ((1 << i) & val) {
1650 			/* Locate physical channel */
1651 			struct pl08x_phy_chan *phychan = &pl08x->phy_chans[i];
1652 			struct pl08x_dma_chan *plchan = phychan->serving;
1653 
1654 			/* Schedule tasklet on this channel */
1655 			tasklet_schedule(&plchan->tasklet);
1656 
1657 			mask |= (1 << i);
1658 		}
1659 	}
1660 	/* Clear only the terminal interrupts on channels we processed */
1661 	writel(mask, pl08x->base + PL080_TC_CLEAR);
1662 
1663 	return mask ? IRQ_HANDLED : IRQ_NONE;
1664 }
1665 
1666 /*
1667  * Initialise the DMAC memcpy/slave channels.
1668  * Make a local wrapper to hold required data
1669  */
1670 static int pl08x_dma_init_virtual_channels(struct pl08x_driver_data *pl08x,
1671 					   struct dma_device *dmadev,
1672 					   unsigned int channels,
1673 					   bool slave)
1674 {
1675 	struct pl08x_dma_chan *chan;
1676 	int i;
1677 
1678 	INIT_LIST_HEAD(&dmadev->channels);
1679 
1680 	/*
1681 	 * Register as many many memcpy as we have physical channels,
1682 	 * we won't always be able to use all but the code will have
1683 	 * to cope with that situation.
1684 	 */
1685 	for (i = 0; i < channels; i++) {
1686 		chan = kzalloc(sizeof(struct pl08x_dma_chan), GFP_KERNEL);
1687 		if (!chan) {
1688 			dev_err(&pl08x->adev->dev,
1689 				"%s no memory for channel\n", __func__);
1690 			return -ENOMEM;
1691 		}
1692 
1693 		chan->host = pl08x;
1694 		chan->state = PL08X_CHAN_IDLE;
1695 
1696 		if (slave) {
1697 			chan->slave = true;
1698 			chan->name = pl08x->pd->slave_channels[i].bus_id;
1699 			chan->cd = &pl08x->pd->slave_channels[i];
1700 		} else {
1701 			chan->cd = &pl08x->pd->memcpy_channel;
1702 			chan->name = kasprintf(GFP_KERNEL, "memcpy%d", i);
1703 			if (!chan->name) {
1704 				kfree(chan);
1705 				return -ENOMEM;
1706 			}
1707 		}
1708 		if (chan->cd->circular_buffer) {
1709 			dev_err(&pl08x->adev->dev,
1710 				"channel %s: circular buffers not supported\n",
1711 				chan->name);
1712 			kfree(chan);
1713 			continue;
1714 		}
1715 		dev_info(&pl08x->adev->dev,
1716 			 "initialize virtual channel \"%s\"\n",
1717 			 chan->name);
1718 
1719 		chan->chan.device = dmadev;
1720 		chan->chan.cookie = 0;
1721 		chan->lc = 0;
1722 
1723 		spin_lock_init(&chan->lock);
1724 		INIT_LIST_HEAD(&chan->pend_list);
1725 		tasklet_init(&chan->tasklet, pl08x_tasklet,
1726 			     (unsigned long) chan);
1727 
1728 		list_add_tail(&chan->chan.device_node, &dmadev->channels);
1729 	}
1730 	dev_info(&pl08x->adev->dev, "initialized %d virtual %s channels\n",
1731 		 i, slave ? "slave" : "memcpy");
1732 	return i;
1733 }
1734 
1735 static void pl08x_free_virtual_channels(struct dma_device *dmadev)
1736 {
1737 	struct pl08x_dma_chan *chan = NULL;
1738 	struct pl08x_dma_chan *next;
1739 
1740 	list_for_each_entry_safe(chan,
1741 				 next, &dmadev->channels, chan.device_node) {
1742 		list_del(&chan->chan.device_node);
1743 		kfree(chan);
1744 	}
1745 }
1746 
1747 #ifdef CONFIG_DEBUG_FS
1748 static const char *pl08x_state_str(enum pl08x_dma_chan_state state)
1749 {
1750 	switch (state) {
1751 	case PL08X_CHAN_IDLE:
1752 		return "idle";
1753 	case PL08X_CHAN_RUNNING:
1754 		return "running";
1755 	case PL08X_CHAN_PAUSED:
1756 		return "paused";
1757 	case PL08X_CHAN_WAITING:
1758 		return "waiting";
1759 	default:
1760 		break;
1761 	}
1762 	return "UNKNOWN STATE";
1763 }
1764 
1765 static int pl08x_debugfs_show(struct seq_file *s, void *data)
1766 {
1767 	struct pl08x_driver_data *pl08x = s->private;
1768 	struct pl08x_dma_chan *chan;
1769 	struct pl08x_phy_chan *ch;
1770 	unsigned long flags;
1771 	int i;
1772 
1773 	seq_printf(s, "PL08x physical channels:\n");
1774 	seq_printf(s, "CHANNEL:\tUSER:\n");
1775 	seq_printf(s, "--------\t-----\n");
1776 	for (i = 0; i < pl08x->vd->channels; i++) {
1777 		struct pl08x_dma_chan *virt_chan;
1778 
1779 		ch = &pl08x->phy_chans[i];
1780 
1781 		spin_lock_irqsave(&ch->lock, flags);
1782 		virt_chan = ch->serving;
1783 
1784 		seq_printf(s, "%d\t\t%s\n",
1785 			   ch->id, virt_chan ? virt_chan->name : "(none)");
1786 
1787 		spin_unlock_irqrestore(&ch->lock, flags);
1788 	}
1789 
1790 	seq_printf(s, "\nPL08x virtual memcpy channels:\n");
1791 	seq_printf(s, "CHANNEL:\tSTATE:\n");
1792 	seq_printf(s, "--------\t------\n");
1793 	list_for_each_entry(chan, &pl08x->memcpy.channels, chan.device_node) {
1794 		seq_printf(s, "%s\t\t%s\n", chan->name,
1795 			   pl08x_state_str(chan->state));
1796 	}
1797 
1798 	seq_printf(s, "\nPL08x virtual slave channels:\n");
1799 	seq_printf(s, "CHANNEL:\tSTATE:\n");
1800 	seq_printf(s, "--------\t------\n");
1801 	list_for_each_entry(chan, &pl08x->slave.channels, chan.device_node) {
1802 		seq_printf(s, "%s\t\t%s\n", chan->name,
1803 			   pl08x_state_str(chan->state));
1804 	}
1805 
1806 	return 0;
1807 }
1808 
1809 static int pl08x_debugfs_open(struct inode *inode, struct file *file)
1810 {
1811 	return single_open(file, pl08x_debugfs_show, inode->i_private);
1812 }
1813 
1814 static const struct file_operations pl08x_debugfs_operations = {
1815 	.open		= pl08x_debugfs_open,
1816 	.read		= seq_read,
1817 	.llseek		= seq_lseek,
1818 	.release	= single_release,
1819 };
1820 
1821 static void init_pl08x_debugfs(struct pl08x_driver_data *pl08x)
1822 {
1823 	/* Expose a simple debugfs interface to view all clocks */
1824 	(void) debugfs_create_file(dev_name(&pl08x->adev->dev), S_IFREG | S_IRUGO,
1825 				   NULL, pl08x,
1826 				   &pl08x_debugfs_operations);
1827 }
1828 
1829 #else
1830 static inline void init_pl08x_debugfs(struct pl08x_driver_data *pl08x)
1831 {
1832 }
1833 #endif
1834 
1835 static int pl08x_probe(struct amba_device *adev, struct amba_id *id)
1836 {
1837 	struct pl08x_driver_data *pl08x;
1838 	const struct vendor_data *vd = id->data;
1839 	int ret = 0;
1840 	int i;
1841 
1842 	ret = amba_request_regions(adev, NULL);
1843 	if (ret)
1844 		return ret;
1845 
1846 	/* Create the driver state holder */
1847 	pl08x = kzalloc(sizeof(struct pl08x_driver_data), GFP_KERNEL);
1848 	if (!pl08x) {
1849 		ret = -ENOMEM;
1850 		goto out_no_pl08x;
1851 	}
1852 
1853 	/* Initialize memcpy engine */
1854 	dma_cap_set(DMA_MEMCPY, pl08x->memcpy.cap_mask);
1855 	pl08x->memcpy.dev = &adev->dev;
1856 	pl08x->memcpy.device_alloc_chan_resources = pl08x_alloc_chan_resources;
1857 	pl08x->memcpy.device_free_chan_resources = pl08x_free_chan_resources;
1858 	pl08x->memcpy.device_prep_dma_memcpy = pl08x_prep_dma_memcpy;
1859 	pl08x->memcpy.device_prep_dma_interrupt = pl08x_prep_dma_interrupt;
1860 	pl08x->memcpy.device_tx_status = pl08x_dma_tx_status;
1861 	pl08x->memcpy.device_issue_pending = pl08x_issue_pending;
1862 	pl08x->memcpy.device_control = pl08x_control;
1863 
1864 	/* Initialize slave engine */
1865 	dma_cap_set(DMA_SLAVE, pl08x->slave.cap_mask);
1866 	pl08x->slave.dev = &adev->dev;
1867 	pl08x->slave.device_alloc_chan_resources = pl08x_alloc_chan_resources;
1868 	pl08x->slave.device_free_chan_resources = pl08x_free_chan_resources;
1869 	pl08x->slave.device_prep_dma_interrupt = pl08x_prep_dma_interrupt;
1870 	pl08x->slave.device_tx_status = pl08x_dma_tx_status;
1871 	pl08x->slave.device_issue_pending = pl08x_issue_pending;
1872 	pl08x->slave.device_prep_slave_sg = pl08x_prep_slave_sg;
1873 	pl08x->slave.device_control = pl08x_control;
1874 
1875 	/* Get the platform data */
1876 	pl08x->pd = dev_get_platdata(&adev->dev);
1877 	if (!pl08x->pd) {
1878 		dev_err(&adev->dev, "no platform data supplied\n");
1879 		goto out_no_platdata;
1880 	}
1881 
1882 	/* Assign useful pointers to the driver state */
1883 	pl08x->adev = adev;
1884 	pl08x->vd = vd;
1885 
1886 	/* By default, AHB1 only.  If dualmaster, from platform */
1887 	pl08x->lli_buses = PL08X_AHB1;
1888 	pl08x->mem_buses = PL08X_AHB1;
1889 	if (pl08x->vd->dualmaster) {
1890 		pl08x->lli_buses = pl08x->pd->lli_buses;
1891 		pl08x->mem_buses = pl08x->pd->mem_buses;
1892 	}
1893 
1894 	/* A DMA memory pool for LLIs, align on 1-byte boundary */
1895 	pl08x->pool = dma_pool_create(DRIVER_NAME, &pl08x->adev->dev,
1896 			PL08X_LLI_TSFR_SIZE, PL08X_ALIGN, 0);
1897 	if (!pl08x->pool) {
1898 		ret = -ENOMEM;
1899 		goto out_no_lli_pool;
1900 	}
1901 
1902 	spin_lock_init(&pl08x->lock);
1903 
1904 	pl08x->base = ioremap(adev->res.start, resource_size(&adev->res));
1905 	if (!pl08x->base) {
1906 		ret = -ENOMEM;
1907 		goto out_no_ioremap;
1908 	}
1909 
1910 	/* Turn on the PL08x */
1911 	pl08x_ensure_on(pl08x);
1912 
1913 	/* Attach the interrupt handler */
1914 	writel(0x000000FF, pl08x->base + PL080_ERR_CLEAR);
1915 	writel(0x000000FF, pl08x->base + PL080_TC_CLEAR);
1916 
1917 	ret = request_irq(adev->irq[0], pl08x_irq, IRQF_DISABLED,
1918 			  DRIVER_NAME, pl08x);
1919 	if (ret) {
1920 		dev_err(&adev->dev, "%s failed to request interrupt %d\n",
1921 			__func__, adev->irq[0]);
1922 		goto out_no_irq;
1923 	}
1924 
1925 	/* Initialize physical channels */
1926 	pl08x->phy_chans = kmalloc((vd->channels * sizeof(struct pl08x_phy_chan)),
1927 			GFP_KERNEL);
1928 	if (!pl08x->phy_chans) {
1929 		dev_err(&adev->dev, "%s failed to allocate "
1930 			"physical channel holders\n",
1931 			__func__);
1932 		goto out_no_phychans;
1933 	}
1934 
1935 	for (i = 0; i < vd->channels; i++) {
1936 		struct pl08x_phy_chan *ch = &pl08x->phy_chans[i];
1937 
1938 		ch->id = i;
1939 		ch->base = pl08x->base + PL080_Cx_BASE(i);
1940 		spin_lock_init(&ch->lock);
1941 		ch->serving = NULL;
1942 		ch->signal = -1;
1943 		dev_info(&adev->dev,
1944 			 "physical channel %d is %s\n", i,
1945 			 pl08x_phy_channel_busy(ch) ? "BUSY" : "FREE");
1946 	}
1947 
1948 	/* Register as many memcpy channels as there are physical channels */
1949 	ret = pl08x_dma_init_virtual_channels(pl08x, &pl08x->memcpy,
1950 					      pl08x->vd->channels, false);
1951 	if (ret <= 0) {
1952 		dev_warn(&pl08x->adev->dev,
1953 			 "%s failed to enumerate memcpy channels - %d\n",
1954 			 __func__, ret);
1955 		goto out_no_memcpy;
1956 	}
1957 	pl08x->memcpy.chancnt = ret;
1958 
1959 	/* Register slave channels */
1960 	ret = pl08x_dma_init_virtual_channels(pl08x, &pl08x->slave,
1961 					      pl08x->pd->num_slave_channels,
1962 					      true);
1963 	if (ret <= 0) {
1964 		dev_warn(&pl08x->adev->dev,
1965 			"%s failed to enumerate slave channels - %d\n",
1966 				__func__, ret);
1967 		goto out_no_slave;
1968 	}
1969 	pl08x->slave.chancnt = ret;
1970 
1971 	ret = dma_async_device_register(&pl08x->memcpy);
1972 	if (ret) {
1973 		dev_warn(&pl08x->adev->dev,
1974 			"%s failed to register memcpy as an async device - %d\n",
1975 			__func__, ret);
1976 		goto out_no_memcpy_reg;
1977 	}
1978 
1979 	ret = dma_async_device_register(&pl08x->slave);
1980 	if (ret) {
1981 		dev_warn(&pl08x->adev->dev,
1982 			"%s failed to register slave as an async device - %d\n",
1983 			__func__, ret);
1984 		goto out_no_slave_reg;
1985 	}
1986 
1987 	amba_set_drvdata(adev, pl08x);
1988 	init_pl08x_debugfs(pl08x);
1989 	dev_info(&pl08x->adev->dev, "DMA: PL%03x rev%u at 0x%08llx irq %d\n",
1990 		 amba_part(adev), amba_rev(adev),
1991 		 (unsigned long long)adev->res.start, adev->irq[0]);
1992 	return 0;
1993 
1994 out_no_slave_reg:
1995 	dma_async_device_unregister(&pl08x->memcpy);
1996 out_no_memcpy_reg:
1997 	pl08x_free_virtual_channels(&pl08x->slave);
1998 out_no_slave:
1999 	pl08x_free_virtual_channels(&pl08x->memcpy);
2000 out_no_memcpy:
2001 	kfree(pl08x->phy_chans);
2002 out_no_phychans:
2003 	free_irq(adev->irq[0], pl08x);
2004 out_no_irq:
2005 	iounmap(pl08x->base);
2006 out_no_ioremap:
2007 	dma_pool_destroy(pl08x->pool);
2008 out_no_lli_pool:
2009 out_no_platdata:
2010 	kfree(pl08x);
2011 out_no_pl08x:
2012 	amba_release_regions(adev);
2013 	return ret;
2014 }
2015 
2016 /* PL080 has 8 channels and the PL080 have just 2 */
2017 static struct vendor_data vendor_pl080 = {
2018 	.channels = 8,
2019 	.dualmaster = true,
2020 };
2021 
2022 static struct vendor_data vendor_pl081 = {
2023 	.channels = 2,
2024 	.dualmaster = false,
2025 };
2026 
2027 static struct amba_id pl08x_ids[] = {
2028 	/* PL080 */
2029 	{
2030 		.id	= 0x00041080,
2031 		.mask	= 0x000fffff,
2032 		.data	= &vendor_pl080,
2033 	},
2034 	/* PL081 */
2035 	{
2036 		.id	= 0x00041081,
2037 		.mask	= 0x000fffff,
2038 		.data	= &vendor_pl081,
2039 	},
2040 	/* Nomadik 8815 PL080 variant */
2041 	{
2042 		.id	= 0x00280880,
2043 		.mask	= 0x00ffffff,
2044 		.data	= &vendor_pl080,
2045 	},
2046 	{ 0, 0 },
2047 };
2048 
2049 static struct amba_driver pl08x_amba_driver = {
2050 	.drv.name	= DRIVER_NAME,
2051 	.id_table	= pl08x_ids,
2052 	.probe		= pl08x_probe,
2053 };
2054 
2055 static int __init pl08x_init(void)
2056 {
2057 	int retval;
2058 	retval = amba_driver_register(&pl08x_amba_driver);
2059 	if (retval)
2060 		printk(KERN_WARNING DRIVER_NAME
2061 		       "failed to register as an AMBA device (%d)\n",
2062 		       retval);
2063 	return retval;
2064 }
2065 subsys_initcall(pl08x_init);
2066