xref: /linux/drivers/dma/amba-pl08x.c (revision 60e13231561b3a4c5269bfa1ef6c0569ad6f28ec)
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/delay.h>
83 #include <linux/dma-mapping.h>
84 #include <linux/dmapool.h>
85 #include <linux/dmaengine.h>
86 #include <linux/amba/bus.h>
87 #include <linux/amba/pl08x.h>
88 #include <linux/debugfs.h>
89 #include <linux/seq_file.h>
90 
91 #include <asm/hardware/pl080.h>
92 
93 #define DRIVER_NAME	"pl08xdmac"
94 
95 /**
96  * struct vendor_data - vendor-specific config parameters for PL08x derivatives
97  * @channels: the number of channels available in this variant
98  * @dualmaster: whether this version supports dual AHB masters or not.
99  */
100 struct vendor_data {
101 	u8 channels;
102 	bool dualmaster;
103 };
104 
105 /*
106  * PL08X private data structures
107  * An LLI struct - see PL08x TRM.  Note that next uses bit[0] as a bus bit,
108  * start & end do not - their bus bit info is in cctl.  Also note that these
109  * are fixed 32-bit quantities.
110  */
111 struct pl08x_lli {
112 	u32 src;
113 	u32 dst;
114 	u32 lli;
115 	u32 cctl;
116 };
117 
118 /**
119  * struct pl08x_driver_data - the local state holder for the PL08x
120  * @slave: slave engine for this instance
121  * @memcpy: memcpy engine for this instance
122  * @base: virtual memory base (remapped) for the PL08x
123  * @adev: the corresponding AMBA (PrimeCell) bus entry
124  * @vd: vendor data for this PL08x variant
125  * @pd: platform data passed in from the platform/machine
126  * @phy_chans: array of data for the physical channels
127  * @pool: a pool for the LLI descriptors
128  * @pool_ctr: counter of LLIs in the pool
129  * @lli_buses: bitmask to or in to LLI pointer selecting AHB port for LLI fetches
130  * @mem_buses: set to indicate memory transfers on AHB2.
131  * @lock: a spinlock for this struct
132  */
133 struct pl08x_driver_data {
134 	struct dma_device slave;
135 	struct dma_device memcpy;
136 	void __iomem *base;
137 	struct amba_device *adev;
138 	const struct vendor_data *vd;
139 	struct pl08x_platform_data *pd;
140 	struct pl08x_phy_chan *phy_chans;
141 	struct dma_pool *pool;
142 	int pool_ctr;
143 	u8 lli_buses;
144 	u8 mem_buses;
145 	spinlock_t lock;
146 };
147 
148 /*
149  * PL08X specific defines
150  */
151 
152 /*
153  * Memory boundaries: the manual for PL08x says that the controller
154  * cannot read past a 1KiB boundary, so these defines are used to
155  * create transfer LLIs that do not cross such boundaries.
156  */
157 #define PL08X_BOUNDARY_SHIFT		(10)	/* 1KB 0x400 */
158 #define PL08X_BOUNDARY_SIZE		(1 << PL08X_BOUNDARY_SHIFT)
159 
160 /* Size (bytes) of each LLI buffer allocated for one transfer */
161 # define PL08X_LLI_TSFR_SIZE	0x2000
162 
163 /* Maximum times we call dma_pool_alloc on this pool without freeing */
164 #define MAX_NUM_TSFR_LLIS	(PL08X_LLI_TSFR_SIZE/sizeof(struct pl08x_lli))
165 #define PL08X_ALIGN		8
166 
167 static inline struct pl08x_dma_chan *to_pl08x_chan(struct dma_chan *chan)
168 {
169 	return container_of(chan, struct pl08x_dma_chan, chan);
170 }
171 
172 static inline struct pl08x_txd *to_pl08x_txd(struct dma_async_tx_descriptor *tx)
173 {
174 	return container_of(tx, struct pl08x_txd, tx);
175 }
176 
177 /*
178  * Physical channel handling
179  */
180 
181 /* Whether a certain channel is busy or not */
182 static int pl08x_phy_channel_busy(struct pl08x_phy_chan *ch)
183 {
184 	unsigned int val;
185 
186 	val = readl(ch->base + PL080_CH_CONFIG);
187 	return val & PL080_CONFIG_ACTIVE;
188 }
189 
190 /*
191  * Set the initial DMA register values i.e. those for the first LLI
192  * The next LLI pointer and the configuration interrupt bit have
193  * been set when the LLIs were constructed.  Poke them into the hardware
194  * and start the transfer.
195  */
196 static void pl08x_start_txd(struct pl08x_dma_chan *plchan,
197 	struct pl08x_txd *txd)
198 {
199 	struct pl08x_driver_data *pl08x = plchan->host;
200 	struct pl08x_phy_chan *phychan = plchan->phychan;
201 	struct pl08x_lli *lli = &txd->llis_va[0];
202 	u32 val;
203 
204 	plchan->at = txd;
205 
206 	/* Wait for channel inactive */
207 	while (pl08x_phy_channel_busy(phychan))
208 		cpu_relax();
209 
210 	dev_vdbg(&pl08x->adev->dev,
211 		"WRITE channel %d: csrc=0x%08x, cdst=0x%08x, "
212 		"clli=0x%08x, cctl=0x%08x, ccfg=0x%08x\n",
213 		phychan->id, lli->src, lli->dst, lli->lli, lli->cctl,
214 		txd->ccfg);
215 
216 	writel(lli->src, phychan->base + PL080_CH_SRC_ADDR);
217 	writel(lli->dst, phychan->base + PL080_CH_DST_ADDR);
218 	writel(lli->lli, phychan->base + PL080_CH_LLI);
219 	writel(lli->cctl, phychan->base + PL080_CH_CONTROL);
220 	writel(txd->ccfg, phychan->base + PL080_CH_CONFIG);
221 
222 	/* Enable the DMA channel */
223 	/* Do not access config register until channel shows as disabled */
224 	while (readl(pl08x->base + PL080_EN_CHAN) & (1 << phychan->id))
225 		cpu_relax();
226 
227 	/* Do not access config register until channel shows as inactive */
228 	val = readl(phychan->base + PL080_CH_CONFIG);
229 	while ((val & PL080_CONFIG_ACTIVE) || (val & PL080_CONFIG_ENABLE))
230 		val = readl(phychan->base + PL080_CH_CONFIG);
231 
232 	writel(val | PL080_CONFIG_ENABLE, phychan->base + PL080_CH_CONFIG);
233 }
234 
235 /*
236  * Pause the channel by setting the HALT bit.
237  *
238  * For M->P transfers, pause the DMAC first and then stop the peripheral -
239  * the FIFO can only drain if the peripheral is still requesting data.
240  * (note: this can still timeout if the DMAC FIFO never drains of data.)
241  *
242  * For P->M transfers, disable the peripheral first to stop it filling
243  * the DMAC FIFO, and then pause the DMAC.
244  */
245 static void pl08x_pause_phy_chan(struct pl08x_phy_chan *ch)
246 {
247 	u32 val;
248 	int timeout;
249 
250 	/* Set the HALT bit and wait for the FIFO to drain */
251 	val = readl(ch->base + PL080_CH_CONFIG);
252 	val |= PL080_CONFIG_HALT;
253 	writel(val, ch->base + PL080_CH_CONFIG);
254 
255 	/* Wait for channel inactive */
256 	for (timeout = 1000; timeout; timeout--) {
257 		if (!pl08x_phy_channel_busy(ch))
258 			break;
259 		udelay(1);
260 	}
261 	if (pl08x_phy_channel_busy(ch))
262 		pr_err("pl08x: channel%u timeout waiting for pause\n", ch->id);
263 }
264 
265 static void pl08x_resume_phy_chan(struct pl08x_phy_chan *ch)
266 {
267 	u32 val;
268 
269 	/* Clear the HALT bit */
270 	val = readl(ch->base + PL080_CH_CONFIG);
271 	val &= ~PL080_CONFIG_HALT;
272 	writel(val, ch->base + PL080_CH_CONFIG);
273 }
274 
275 
276 /*
277  * pl08x_terminate_phy_chan() stops the channel, clears the FIFO and
278  * clears any pending interrupt status.  This should not be used for
279  * an on-going transfer, but as a method of shutting down a channel
280  * (eg, when it's no longer used) or terminating a transfer.
281  */
282 static void pl08x_terminate_phy_chan(struct pl08x_driver_data *pl08x,
283 	struct pl08x_phy_chan *ch)
284 {
285 	u32 val = readl(ch->base + PL080_CH_CONFIG);
286 
287 	val &= ~(PL080_CONFIG_ENABLE | PL080_CONFIG_ERR_IRQ_MASK |
288 	         PL080_CONFIG_TC_IRQ_MASK);
289 
290 	writel(val, ch->base + PL080_CH_CONFIG);
291 
292 	writel(1 << ch->id, pl08x->base + PL080_ERR_CLEAR);
293 	writel(1 << ch->id, pl08x->base + PL080_TC_CLEAR);
294 }
295 
296 static inline u32 get_bytes_in_cctl(u32 cctl)
297 {
298 	/* The source width defines the number of bytes */
299 	u32 bytes = cctl & PL080_CONTROL_TRANSFER_SIZE_MASK;
300 
301 	switch (cctl >> PL080_CONTROL_SWIDTH_SHIFT) {
302 	case PL080_WIDTH_8BIT:
303 		break;
304 	case PL080_WIDTH_16BIT:
305 		bytes *= 2;
306 		break;
307 	case PL080_WIDTH_32BIT:
308 		bytes *= 4;
309 		break;
310 	}
311 	return bytes;
312 }
313 
314 /* The channel should be paused when calling this */
315 static u32 pl08x_getbytes_chan(struct pl08x_dma_chan *plchan)
316 {
317 	struct pl08x_phy_chan *ch;
318 	struct pl08x_txd *txd;
319 	unsigned long flags;
320 	size_t bytes = 0;
321 
322 	spin_lock_irqsave(&plchan->lock, flags);
323 	ch = plchan->phychan;
324 	txd = plchan->at;
325 
326 	/*
327 	 * Follow the LLIs to get the number of remaining
328 	 * bytes in the currently active transaction.
329 	 */
330 	if (ch && txd) {
331 		u32 clli = readl(ch->base + PL080_CH_LLI) & ~PL080_LLI_LM_AHB2;
332 
333 		/* First get the remaining bytes in the active transfer */
334 		bytes = get_bytes_in_cctl(readl(ch->base + PL080_CH_CONTROL));
335 
336 		if (clli) {
337 			struct pl08x_lli *llis_va = txd->llis_va;
338 			dma_addr_t llis_bus = txd->llis_bus;
339 			int index;
340 
341 			BUG_ON(clli < llis_bus || clli >= llis_bus +
342 				sizeof(struct pl08x_lli) * MAX_NUM_TSFR_LLIS);
343 
344 			/*
345 			 * Locate the next LLI - as this is an array,
346 			 * it's simple maths to find.
347 			 */
348 			index = (clli - llis_bus) / sizeof(struct pl08x_lli);
349 
350 			for (; index < MAX_NUM_TSFR_LLIS; index++) {
351 				bytes += get_bytes_in_cctl(llis_va[index].cctl);
352 
353 				/*
354 				 * A LLI pointer of 0 terminates the LLI list
355 				 */
356 				if (!llis_va[index].lli)
357 					break;
358 			}
359 		}
360 	}
361 
362 	/* Sum up all queued transactions */
363 	if (!list_empty(&plchan->pend_list)) {
364 		struct pl08x_txd *txdi;
365 		list_for_each_entry(txdi, &plchan->pend_list, node) {
366 			bytes += txdi->len;
367 		}
368 	}
369 
370 	spin_unlock_irqrestore(&plchan->lock, flags);
371 
372 	return bytes;
373 }
374 
375 /*
376  * Allocate a physical channel for a virtual channel
377  *
378  * Try to locate a physical channel to be used for this transfer. If all
379  * are taken return NULL and the requester will have to cope by using
380  * some fallback PIO mode or retrying later.
381  */
382 static struct pl08x_phy_chan *
383 pl08x_get_phy_channel(struct pl08x_driver_data *pl08x,
384 		      struct pl08x_dma_chan *virt_chan)
385 {
386 	struct pl08x_phy_chan *ch = NULL;
387 	unsigned long flags;
388 	int i;
389 
390 	for (i = 0; i < pl08x->vd->channels; i++) {
391 		ch = &pl08x->phy_chans[i];
392 
393 		spin_lock_irqsave(&ch->lock, flags);
394 
395 		if (!ch->serving) {
396 			ch->serving = virt_chan;
397 			ch->signal = -1;
398 			spin_unlock_irqrestore(&ch->lock, flags);
399 			break;
400 		}
401 
402 		spin_unlock_irqrestore(&ch->lock, flags);
403 	}
404 
405 	if (i == pl08x->vd->channels) {
406 		/* No physical channel available, cope with it */
407 		return NULL;
408 	}
409 
410 	return ch;
411 }
412 
413 static inline void pl08x_put_phy_channel(struct pl08x_driver_data *pl08x,
414 					 struct pl08x_phy_chan *ch)
415 {
416 	unsigned long flags;
417 
418 	spin_lock_irqsave(&ch->lock, flags);
419 
420 	/* Stop the channel and clear its interrupts */
421 	pl08x_terminate_phy_chan(pl08x, ch);
422 
423 	/* Mark it as free */
424 	ch->serving = NULL;
425 	spin_unlock_irqrestore(&ch->lock, flags);
426 }
427 
428 /*
429  * LLI handling
430  */
431 
432 static inline unsigned int pl08x_get_bytes_for_cctl(unsigned int coded)
433 {
434 	switch (coded) {
435 	case PL080_WIDTH_8BIT:
436 		return 1;
437 	case PL080_WIDTH_16BIT:
438 		return 2;
439 	case PL080_WIDTH_32BIT:
440 		return 4;
441 	default:
442 		break;
443 	}
444 	BUG();
445 	return 0;
446 }
447 
448 static inline u32 pl08x_cctl_bits(u32 cctl, u8 srcwidth, u8 dstwidth,
449 				  size_t tsize)
450 {
451 	u32 retbits = cctl;
452 
453 	/* Remove all src, dst and transfer size bits */
454 	retbits &= ~PL080_CONTROL_DWIDTH_MASK;
455 	retbits &= ~PL080_CONTROL_SWIDTH_MASK;
456 	retbits &= ~PL080_CONTROL_TRANSFER_SIZE_MASK;
457 
458 	/* Then set the bits according to the parameters */
459 	switch (srcwidth) {
460 	case 1:
461 		retbits |= PL080_WIDTH_8BIT << PL080_CONTROL_SWIDTH_SHIFT;
462 		break;
463 	case 2:
464 		retbits |= PL080_WIDTH_16BIT << PL080_CONTROL_SWIDTH_SHIFT;
465 		break;
466 	case 4:
467 		retbits |= PL080_WIDTH_32BIT << PL080_CONTROL_SWIDTH_SHIFT;
468 		break;
469 	default:
470 		BUG();
471 		break;
472 	}
473 
474 	switch (dstwidth) {
475 	case 1:
476 		retbits |= PL080_WIDTH_8BIT << PL080_CONTROL_DWIDTH_SHIFT;
477 		break;
478 	case 2:
479 		retbits |= PL080_WIDTH_16BIT << PL080_CONTROL_DWIDTH_SHIFT;
480 		break;
481 	case 4:
482 		retbits |= PL080_WIDTH_32BIT << PL080_CONTROL_DWIDTH_SHIFT;
483 		break;
484 	default:
485 		BUG();
486 		break;
487 	}
488 
489 	retbits |= tsize << PL080_CONTROL_TRANSFER_SIZE_SHIFT;
490 	return retbits;
491 }
492 
493 struct pl08x_lli_build_data {
494 	struct pl08x_txd *txd;
495 	struct pl08x_bus_data srcbus;
496 	struct pl08x_bus_data dstbus;
497 	size_t remainder;
498 	u32 lli_bus;
499 };
500 
501 /*
502  * Autoselect a master bus to use for the transfer this prefers the
503  * destination bus if both available if fixed address on one bus the
504  * other will be chosen
505  */
506 static void pl08x_choose_master_bus(struct pl08x_lli_build_data *bd,
507 	struct pl08x_bus_data **mbus, struct pl08x_bus_data **sbus, u32 cctl)
508 {
509 	if (!(cctl & PL080_CONTROL_DST_INCR)) {
510 		*mbus = &bd->srcbus;
511 		*sbus = &bd->dstbus;
512 	} else if (!(cctl & PL080_CONTROL_SRC_INCR)) {
513 		*mbus = &bd->dstbus;
514 		*sbus = &bd->srcbus;
515 	} else {
516 		if (bd->dstbus.buswidth == 4) {
517 			*mbus = &bd->dstbus;
518 			*sbus = &bd->srcbus;
519 		} else if (bd->srcbus.buswidth == 4) {
520 			*mbus = &bd->srcbus;
521 			*sbus = &bd->dstbus;
522 		} else if (bd->dstbus.buswidth == 2) {
523 			*mbus = &bd->dstbus;
524 			*sbus = &bd->srcbus;
525 		} else if (bd->srcbus.buswidth == 2) {
526 			*mbus = &bd->srcbus;
527 			*sbus = &bd->dstbus;
528 		} else {
529 			/* bd->srcbus.buswidth == 1 */
530 			*mbus = &bd->dstbus;
531 			*sbus = &bd->srcbus;
532 		}
533 	}
534 }
535 
536 /*
537  * Fills in one LLI for a certain transfer descriptor and advance the counter
538  */
539 static void pl08x_fill_lli_for_desc(struct pl08x_lli_build_data *bd,
540 	int num_llis, int len, u32 cctl)
541 {
542 	struct pl08x_lli *llis_va = bd->txd->llis_va;
543 	dma_addr_t llis_bus = bd->txd->llis_bus;
544 
545 	BUG_ON(num_llis >= MAX_NUM_TSFR_LLIS);
546 
547 	llis_va[num_llis].cctl = cctl;
548 	llis_va[num_llis].src = bd->srcbus.addr;
549 	llis_va[num_llis].dst = bd->dstbus.addr;
550 	llis_va[num_llis].lli = llis_bus + (num_llis + 1) * sizeof(struct pl08x_lli);
551 	llis_va[num_llis].lli |= bd->lli_bus;
552 
553 	if (cctl & PL080_CONTROL_SRC_INCR)
554 		bd->srcbus.addr += len;
555 	if (cctl & PL080_CONTROL_DST_INCR)
556 		bd->dstbus.addr += len;
557 
558 	BUG_ON(bd->remainder < len);
559 
560 	bd->remainder -= len;
561 }
562 
563 /*
564  * Return number of bytes to fill to boundary, or len.
565  * This calculation works for any value of addr.
566  */
567 static inline size_t pl08x_pre_boundary(u32 addr, size_t len)
568 {
569 	size_t boundary_len = PL08X_BOUNDARY_SIZE -
570 			(addr & (PL08X_BOUNDARY_SIZE - 1));
571 
572 	return min(boundary_len, len);
573 }
574 
575 /*
576  * This fills in the table of LLIs for the transfer descriptor
577  * Note that we assume we never have to change the burst sizes
578  * Return 0 for error
579  */
580 static int pl08x_fill_llis_for_desc(struct pl08x_driver_data *pl08x,
581 			      struct pl08x_txd *txd)
582 {
583 	struct pl08x_bus_data *mbus, *sbus;
584 	struct pl08x_lli_build_data bd;
585 	int num_llis = 0;
586 	u32 cctl;
587 	size_t max_bytes_per_lli;
588 	size_t total_bytes = 0;
589 	struct pl08x_lli *llis_va;
590 
591 	txd->llis_va = dma_pool_alloc(pl08x->pool, GFP_NOWAIT,
592 				      &txd->llis_bus);
593 	if (!txd->llis_va) {
594 		dev_err(&pl08x->adev->dev, "%s no memory for llis\n", __func__);
595 		return 0;
596 	}
597 
598 	pl08x->pool_ctr++;
599 
600 	/* Get the default CCTL */
601 	cctl = txd->cctl;
602 
603 	bd.txd = txd;
604 	bd.srcbus.addr = txd->src_addr;
605 	bd.dstbus.addr = txd->dst_addr;
606 	bd.lli_bus = (pl08x->lli_buses & PL08X_AHB2) ? PL080_LLI_LM_AHB2 : 0;
607 
608 	/* Find maximum width of the source bus */
609 	bd.srcbus.maxwidth =
610 		pl08x_get_bytes_for_cctl((cctl & PL080_CONTROL_SWIDTH_MASK) >>
611 				       PL080_CONTROL_SWIDTH_SHIFT);
612 
613 	/* Find maximum width of the destination bus */
614 	bd.dstbus.maxwidth =
615 		pl08x_get_bytes_for_cctl((cctl & PL080_CONTROL_DWIDTH_MASK) >>
616 				       PL080_CONTROL_DWIDTH_SHIFT);
617 
618 	/* Set up the bus widths to the maximum */
619 	bd.srcbus.buswidth = bd.srcbus.maxwidth;
620 	bd.dstbus.buswidth = bd.dstbus.maxwidth;
621 
622 	/*
623 	 * Bytes transferred == tsize * MIN(buswidths), not max(buswidths)
624 	 */
625 	max_bytes_per_lli = min(bd.srcbus.buswidth, bd.dstbus.buswidth) *
626 		PL080_CONTROL_TRANSFER_SIZE_MASK;
627 
628 	/* We need to count this down to zero */
629 	bd.remainder = txd->len;
630 
631 	/*
632 	 * Choose bus to align to
633 	 * - prefers destination bus if both available
634 	 * - if fixed address on one bus chooses other
635 	 */
636 	pl08x_choose_master_bus(&bd, &mbus, &sbus, cctl);
637 
638 	dev_vdbg(&pl08x->adev->dev, "src=0x%08x%s/%u dst=0x%08x%s/%u len=%zu llimax=%zu\n",
639 		 bd.srcbus.addr, cctl & PL080_CONTROL_SRC_INCR ? "+" : "",
640 		 bd.srcbus.buswidth,
641 		 bd.dstbus.addr, cctl & PL080_CONTROL_DST_INCR ? "+" : "",
642 		 bd.dstbus.buswidth,
643 		 bd.remainder, max_bytes_per_lli);
644 	dev_vdbg(&pl08x->adev->dev, "mbus=%s sbus=%s\n",
645 		 mbus == &bd.srcbus ? "src" : "dst",
646 		 sbus == &bd.srcbus ? "src" : "dst");
647 
648 	if (txd->len < mbus->buswidth) {
649 		/* Less than a bus width available - send as single bytes */
650 		while (bd.remainder) {
651 			dev_vdbg(&pl08x->adev->dev,
652 				 "%s single byte LLIs for a transfer of "
653 				 "less than a bus width (remain 0x%08x)\n",
654 				 __func__, bd.remainder);
655 			cctl = pl08x_cctl_bits(cctl, 1, 1, 1);
656 			pl08x_fill_lli_for_desc(&bd, num_llis++, 1, cctl);
657 			total_bytes++;
658 		}
659 	} else {
660 		/* Make one byte LLIs until master bus is aligned */
661 		while ((mbus->addr) % (mbus->buswidth)) {
662 			dev_vdbg(&pl08x->adev->dev,
663 				"%s adjustment lli for less than bus width "
664 				 "(remain 0x%08x)\n",
665 				 __func__, bd.remainder);
666 			cctl = pl08x_cctl_bits(cctl, 1, 1, 1);
667 			pl08x_fill_lli_for_desc(&bd, num_llis++, 1, cctl);
668 			total_bytes++;
669 		}
670 
671 		/*
672 		 * Master now aligned
673 		 * - if slave is not then we must set its width down
674 		 */
675 		if (sbus->addr % sbus->buswidth) {
676 			dev_dbg(&pl08x->adev->dev,
677 				"%s set down bus width to one byte\n",
678 				 __func__);
679 
680 			sbus->buswidth = 1;
681 		}
682 
683 		/*
684 		 * Make largest possible LLIs until less than one bus
685 		 * width left
686 		 */
687 		while (bd.remainder > (mbus->buswidth - 1)) {
688 			size_t lli_len, target_len, tsize, odd_bytes;
689 
690 			/*
691 			 * If enough left try to send max possible,
692 			 * otherwise try to send the remainder
693 			 */
694 			target_len = min(bd.remainder, max_bytes_per_lli);
695 
696 			/*
697 			 * Set bus lengths for incrementing buses to the
698 			 * number of bytes which fill to next memory boundary,
699 			 * limiting on the target length calculated above.
700 			 */
701 			if (cctl & PL080_CONTROL_SRC_INCR)
702 				bd.srcbus.fill_bytes =
703 					pl08x_pre_boundary(bd.srcbus.addr,
704 						target_len);
705 			else
706 				bd.srcbus.fill_bytes = target_len;
707 
708 			if (cctl & PL080_CONTROL_DST_INCR)
709 				bd.dstbus.fill_bytes =
710 					pl08x_pre_boundary(bd.dstbus.addr,
711 						target_len);
712 			else
713 				bd.dstbus.fill_bytes = target_len;
714 
715 			/* Find the nearest */
716 			lli_len	= min(bd.srcbus.fill_bytes,
717 				      bd.dstbus.fill_bytes);
718 
719 			BUG_ON(lli_len > bd.remainder);
720 
721 			if (lli_len <= 0) {
722 				dev_err(&pl08x->adev->dev,
723 					"%s lli_len is %zu, <= 0\n",
724 						__func__, lli_len);
725 				return 0;
726 			}
727 
728 			if (lli_len == target_len) {
729 				/*
730 				 * Can send what we wanted.
731 				 * Maintain alignment
732 				 */
733 				lli_len	= (lli_len/mbus->buswidth) *
734 							mbus->buswidth;
735 				odd_bytes = 0;
736 			} else {
737 				/*
738 				 * So now we know how many bytes to transfer
739 				 * to get to the nearest boundary.  The next
740 				 * LLI will past the boundary.  However, we
741 				 * may be working to a boundary on the slave
742 				 * bus.  We need to ensure the master stays
743 				 * aligned, and that we are working in
744 				 * multiples of the bus widths.
745 				 */
746 				odd_bytes = lli_len % mbus->buswidth;
747 				lli_len -= odd_bytes;
748 
749 			}
750 
751 			if (lli_len) {
752 				/*
753 				 * Check against minimum bus alignment:
754 				 * Calculate actual transfer size in relation
755 				 * to bus width an get a maximum remainder of
756 				 * the smallest bus width - 1
757 				 */
758 				/* FIXME: use round_down()? */
759 				tsize = lli_len / min(mbus->buswidth,
760 						      sbus->buswidth);
761 				lli_len	= tsize * min(mbus->buswidth,
762 						      sbus->buswidth);
763 
764 				if (target_len != lli_len) {
765 					dev_vdbg(&pl08x->adev->dev,
766 					"%s can't send what we want. Desired 0x%08zx, lli of 0x%08zx bytes in txd of 0x%08zx\n",
767 					__func__, target_len, lli_len, txd->len);
768 				}
769 
770 				cctl = pl08x_cctl_bits(cctl,
771 						       bd.srcbus.buswidth,
772 						       bd.dstbus.buswidth,
773 						       tsize);
774 
775 				dev_vdbg(&pl08x->adev->dev,
776 					"%s fill lli with single lli chunk of size 0x%08zx (remainder 0x%08zx)\n",
777 					__func__, lli_len, bd.remainder);
778 				pl08x_fill_lli_for_desc(&bd, num_llis++,
779 					lli_len, cctl);
780 				total_bytes += lli_len;
781 			}
782 
783 
784 			if (odd_bytes) {
785 				/*
786 				 * Creep past the boundary, maintaining
787 				 * master alignment
788 				 */
789 				int j;
790 				for (j = 0; (j < mbus->buswidth)
791 						&& (bd.remainder); j++) {
792 					cctl = pl08x_cctl_bits(cctl, 1, 1, 1);
793 					dev_vdbg(&pl08x->adev->dev,
794 						"%s align with boundary, single byte (remain 0x%08zx)\n",
795 						__func__, bd.remainder);
796 					pl08x_fill_lli_for_desc(&bd,
797 						num_llis++, 1, cctl);
798 					total_bytes++;
799 				}
800 			}
801 		}
802 
803 		/*
804 		 * Send any odd bytes
805 		 */
806 		while (bd.remainder) {
807 			cctl = pl08x_cctl_bits(cctl, 1, 1, 1);
808 			dev_vdbg(&pl08x->adev->dev,
809 				"%s align with boundary, single odd byte (remain %zu)\n",
810 				__func__, bd.remainder);
811 			pl08x_fill_lli_for_desc(&bd, num_llis++, 1, cctl);
812 			total_bytes++;
813 		}
814 	}
815 	if (total_bytes != txd->len) {
816 		dev_err(&pl08x->adev->dev,
817 			"%s size of encoded lli:s don't match total txd, transferred 0x%08zx from size 0x%08zx\n",
818 			__func__, total_bytes, txd->len);
819 		return 0;
820 	}
821 
822 	if (num_llis >= MAX_NUM_TSFR_LLIS) {
823 		dev_err(&pl08x->adev->dev,
824 			"%s need to increase MAX_NUM_TSFR_LLIS from 0x%08x\n",
825 			__func__, (u32) MAX_NUM_TSFR_LLIS);
826 		return 0;
827 	}
828 
829 	llis_va = txd->llis_va;
830 	/* The final LLI terminates the LLI. */
831 	llis_va[num_llis - 1].lli = 0;
832 	/* The final LLI element shall also fire an interrupt. */
833 	llis_va[num_llis - 1].cctl |= PL080_CONTROL_TC_IRQ_EN;
834 
835 #ifdef VERBOSE_DEBUG
836 	{
837 		int i;
838 
839 		dev_vdbg(&pl08x->adev->dev,
840 			 "%-3s %-9s  %-10s %-10s %-10s %s\n",
841 			 "lli", "", "csrc", "cdst", "clli", "cctl");
842 		for (i = 0; i < num_llis; i++) {
843 			dev_vdbg(&pl08x->adev->dev,
844 				 "%3d @%p: 0x%08x 0x%08x 0x%08x 0x%08x\n",
845 				 i, &llis_va[i], llis_va[i].src,
846 				 llis_va[i].dst, llis_va[i].lli, llis_va[i].cctl
847 				);
848 		}
849 	}
850 #endif
851 
852 	return num_llis;
853 }
854 
855 /* You should call this with the struct pl08x lock held */
856 static void pl08x_free_txd(struct pl08x_driver_data *pl08x,
857 			   struct pl08x_txd *txd)
858 {
859 	/* Free the LLI */
860 	dma_pool_free(pl08x->pool, txd->llis_va, txd->llis_bus);
861 
862 	pl08x->pool_ctr--;
863 
864 	kfree(txd);
865 }
866 
867 static void pl08x_free_txd_list(struct pl08x_driver_data *pl08x,
868 				struct pl08x_dma_chan *plchan)
869 {
870 	struct pl08x_txd *txdi = NULL;
871 	struct pl08x_txd *next;
872 
873 	if (!list_empty(&plchan->pend_list)) {
874 		list_for_each_entry_safe(txdi,
875 					 next, &plchan->pend_list, node) {
876 			list_del(&txdi->node);
877 			pl08x_free_txd(pl08x, txdi);
878 		}
879 	}
880 }
881 
882 /*
883  * The DMA ENGINE API
884  */
885 static int pl08x_alloc_chan_resources(struct dma_chan *chan)
886 {
887 	return 0;
888 }
889 
890 static void pl08x_free_chan_resources(struct dma_chan *chan)
891 {
892 }
893 
894 /*
895  * This should be called with the channel plchan->lock held
896  */
897 static int prep_phy_channel(struct pl08x_dma_chan *plchan,
898 			    struct pl08x_txd *txd)
899 {
900 	struct pl08x_driver_data *pl08x = plchan->host;
901 	struct pl08x_phy_chan *ch;
902 	int ret;
903 
904 	/* Check if we already have a channel */
905 	if (plchan->phychan)
906 		return 0;
907 
908 	ch = pl08x_get_phy_channel(pl08x, plchan);
909 	if (!ch) {
910 		/* No physical channel available, cope with it */
911 		dev_dbg(&pl08x->adev->dev, "no physical channel available for xfer on %s\n", plchan->name);
912 		return -EBUSY;
913 	}
914 
915 	/*
916 	 * OK we have a physical channel: for memcpy() this is all we
917 	 * need, but for slaves the physical signals may be muxed!
918 	 * Can the platform allow us to use this channel?
919 	 */
920 	if (plchan->slave &&
921 	    ch->signal < 0 &&
922 	    pl08x->pd->get_signal) {
923 		ret = pl08x->pd->get_signal(plchan);
924 		if (ret < 0) {
925 			dev_dbg(&pl08x->adev->dev,
926 				"unable to use physical channel %d for transfer on %s due to platform restrictions\n",
927 				ch->id, plchan->name);
928 			/* Release physical channel & return */
929 			pl08x_put_phy_channel(pl08x, ch);
930 			return -EBUSY;
931 		}
932 		ch->signal = ret;
933 
934 		/* Assign the flow control signal to this channel */
935 		if (txd->direction == DMA_TO_DEVICE)
936 			txd->ccfg |= ch->signal << PL080_CONFIG_DST_SEL_SHIFT;
937 		else if (txd->direction == DMA_FROM_DEVICE)
938 			txd->ccfg |= ch->signal << PL080_CONFIG_SRC_SEL_SHIFT;
939 	}
940 
941 	dev_dbg(&pl08x->adev->dev, "allocated physical channel %d and signal %d for xfer on %s\n",
942 		 ch->id,
943 		 ch->signal,
944 		 plchan->name);
945 
946 	plchan->phychan_hold++;
947 	plchan->phychan = ch;
948 
949 	return 0;
950 }
951 
952 static void release_phy_channel(struct pl08x_dma_chan *plchan)
953 {
954 	struct pl08x_driver_data *pl08x = plchan->host;
955 
956 	if ((plchan->phychan->signal >= 0) && pl08x->pd->put_signal) {
957 		pl08x->pd->put_signal(plchan);
958 		plchan->phychan->signal = -1;
959 	}
960 	pl08x_put_phy_channel(pl08x, plchan->phychan);
961 	plchan->phychan = NULL;
962 }
963 
964 static dma_cookie_t pl08x_tx_submit(struct dma_async_tx_descriptor *tx)
965 {
966 	struct pl08x_dma_chan *plchan = to_pl08x_chan(tx->chan);
967 	struct pl08x_txd *txd = to_pl08x_txd(tx);
968 	unsigned long flags;
969 
970 	spin_lock_irqsave(&plchan->lock, flags);
971 
972 	plchan->chan.cookie += 1;
973 	if (plchan->chan.cookie < 0)
974 		plchan->chan.cookie = 1;
975 	tx->cookie = plchan->chan.cookie;
976 
977 	/* Put this onto the pending list */
978 	list_add_tail(&txd->node, &plchan->pend_list);
979 
980 	/*
981 	 * If there was no physical channel available for this memcpy,
982 	 * stack the request up and indicate that the channel is waiting
983 	 * for a free physical channel.
984 	 */
985 	if (!plchan->slave && !plchan->phychan) {
986 		/* Do this memcpy whenever there is a channel ready */
987 		plchan->state = PL08X_CHAN_WAITING;
988 		plchan->waiting = txd;
989 	} else {
990 		plchan->phychan_hold--;
991 	}
992 
993 	spin_unlock_irqrestore(&plchan->lock, flags);
994 
995 	return tx->cookie;
996 }
997 
998 static struct dma_async_tx_descriptor *pl08x_prep_dma_interrupt(
999 		struct dma_chan *chan, unsigned long flags)
1000 {
1001 	struct dma_async_tx_descriptor *retval = NULL;
1002 
1003 	return retval;
1004 }
1005 
1006 /*
1007  * Code accessing dma_async_is_complete() in a tight loop may give problems.
1008  * If slaves are relying on interrupts to signal completion this function
1009  * must not be called with interrupts disabled.
1010  */
1011 static enum dma_status
1012 pl08x_dma_tx_status(struct dma_chan *chan,
1013 		    dma_cookie_t cookie,
1014 		    struct dma_tx_state *txstate)
1015 {
1016 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1017 	dma_cookie_t last_used;
1018 	dma_cookie_t last_complete;
1019 	enum dma_status ret;
1020 	u32 bytesleft = 0;
1021 
1022 	last_used = plchan->chan.cookie;
1023 	last_complete = plchan->lc;
1024 
1025 	ret = dma_async_is_complete(cookie, last_complete, last_used);
1026 	if (ret == DMA_SUCCESS) {
1027 		dma_set_tx_state(txstate, last_complete, last_used, 0);
1028 		return ret;
1029 	}
1030 
1031 	/*
1032 	 * This cookie not complete yet
1033 	 */
1034 	last_used = plchan->chan.cookie;
1035 	last_complete = plchan->lc;
1036 
1037 	/* Get number of bytes left in the active transactions and queue */
1038 	bytesleft = pl08x_getbytes_chan(plchan);
1039 
1040 	dma_set_tx_state(txstate, last_complete, last_used,
1041 			 bytesleft);
1042 
1043 	if (plchan->state == PL08X_CHAN_PAUSED)
1044 		return DMA_PAUSED;
1045 
1046 	/* Whether waiting or running, we're in progress */
1047 	return DMA_IN_PROGRESS;
1048 }
1049 
1050 /* PrimeCell DMA extension */
1051 struct burst_table {
1052 	u32 burstwords;
1053 	u32 reg;
1054 };
1055 
1056 static const struct burst_table burst_sizes[] = {
1057 	{
1058 		.burstwords = 256,
1059 		.reg = PL080_BSIZE_256,
1060 	},
1061 	{
1062 		.burstwords = 128,
1063 		.reg = PL080_BSIZE_128,
1064 	},
1065 	{
1066 		.burstwords = 64,
1067 		.reg = PL080_BSIZE_64,
1068 	},
1069 	{
1070 		.burstwords = 32,
1071 		.reg = PL080_BSIZE_32,
1072 	},
1073 	{
1074 		.burstwords = 16,
1075 		.reg = PL080_BSIZE_16,
1076 	},
1077 	{
1078 		.burstwords = 8,
1079 		.reg = PL080_BSIZE_8,
1080 	},
1081 	{
1082 		.burstwords = 4,
1083 		.reg = PL080_BSIZE_4,
1084 	},
1085 	{
1086 		.burstwords = 0,
1087 		.reg = PL080_BSIZE_1,
1088 	},
1089 };
1090 
1091 /*
1092  * Given the source and destination available bus masks, select which
1093  * will be routed to each port.  We try to have source and destination
1094  * on separate ports, but always respect the allowable settings.
1095  */
1096 static u32 pl08x_select_bus(u8 src, u8 dst)
1097 {
1098 	u32 cctl = 0;
1099 
1100 	if (!(dst & PL08X_AHB1) || ((dst & PL08X_AHB2) && (src & PL08X_AHB1)))
1101 		cctl |= PL080_CONTROL_DST_AHB2;
1102 	if (!(src & PL08X_AHB1) || ((src & PL08X_AHB2) && !(dst & PL08X_AHB2)))
1103 		cctl |= PL080_CONTROL_SRC_AHB2;
1104 
1105 	return cctl;
1106 }
1107 
1108 static u32 pl08x_cctl(u32 cctl)
1109 {
1110 	cctl &= ~(PL080_CONTROL_SRC_AHB2 | PL080_CONTROL_DST_AHB2 |
1111 		  PL080_CONTROL_SRC_INCR | PL080_CONTROL_DST_INCR |
1112 		  PL080_CONTROL_PROT_MASK);
1113 
1114 	/* Access the cell in privileged mode, non-bufferable, non-cacheable */
1115 	return cctl | PL080_CONTROL_PROT_SYS;
1116 }
1117 
1118 static u32 pl08x_width(enum dma_slave_buswidth width)
1119 {
1120 	switch (width) {
1121 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
1122 		return PL080_WIDTH_8BIT;
1123 	case DMA_SLAVE_BUSWIDTH_2_BYTES:
1124 		return PL080_WIDTH_16BIT;
1125 	case DMA_SLAVE_BUSWIDTH_4_BYTES:
1126 		return PL080_WIDTH_32BIT;
1127 	default:
1128 		return ~0;
1129 	}
1130 }
1131 
1132 static u32 pl08x_burst(u32 maxburst)
1133 {
1134 	int i;
1135 
1136 	for (i = 0; i < ARRAY_SIZE(burst_sizes); i++)
1137 		if (burst_sizes[i].burstwords <= maxburst)
1138 			break;
1139 
1140 	return burst_sizes[i].reg;
1141 }
1142 
1143 static int dma_set_runtime_config(struct dma_chan *chan,
1144 				  struct dma_slave_config *config)
1145 {
1146 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1147 	struct pl08x_driver_data *pl08x = plchan->host;
1148 	enum dma_slave_buswidth addr_width;
1149 	u32 width, burst, maxburst;
1150 	u32 cctl = 0;
1151 
1152 	if (!plchan->slave)
1153 		return -EINVAL;
1154 
1155 	/* Transfer direction */
1156 	plchan->runtime_direction = config->direction;
1157 	if (config->direction == DMA_TO_DEVICE) {
1158 		addr_width = config->dst_addr_width;
1159 		maxburst = config->dst_maxburst;
1160 	} else if (config->direction == DMA_FROM_DEVICE) {
1161 		addr_width = config->src_addr_width;
1162 		maxburst = config->src_maxburst;
1163 	} else {
1164 		dev_err(&pl08x->adev->dev,
1165 			"bad runtime_config: alien transfer direction\n");
1166 		return -EINVAL;
1167 	}
1168 
1169 	width = pl08x_width(addr_width);
1170 	if (width == ~0) {
1171 		dev_err(&pl08x->adev->dev,
1172 			"bad runtime_config: alien address width\n");
1173 		return -EINVAL;
1174 	}
1175 
1176 	cctl |= width << PL080_CONTROL_SWIDTH_SHIFT;
1177 	cctl |= width << PL080_CONTROL_DWIDTH_SHIFT;
1178 
1179 	/*
1180 	 * If this channel will only request single transfers, set this
1181 	 * down to ONE element.  Also select one element if no maxburst
1182 	 * is specified.
1183 	 */
1184 	if (plchan->cd->single)
1185 		maxburst = 1;
1186 
1187 	burst = pl08x_burst(maxburst);
1188 	cctl |= burst << PL080_CONTROL_SB_SIZE_SHIFT;
1189 	cctl |= burst << PL080_CONTROL_DB_SIZE_SHIFT;
1190 
1191 	if (plchan->runtime_direction == DMA_FROM_DEVICE) {
1192 		plchan->src_addr = config->src_addr;
1193 		plchan->src_cctl = pl08x_cctl(cctl) | PL080_CONTROL_DST_INCR |
1194 			pl08x_select_bus(plchan->cd->periph_buses,
1195 					 pl08x->mem_buses);
1196 	} else {
1197 		plchan->dst_addr = config->dst_addr;
1198 		plchan->dst_cctl = pl08x_cctl(cctl) | PL080_CONTROL_SRC_INCR |
1199 			pl08x_select_bus(pl08x->mem_buses,
1200 					 plchan->cd->periph_buses);
1201 	}
1202 
1203 	dev_dbg(&pl08x->adev->dev,
1204 		"configured channel %s (%s) for %s, data width %d, "
1205 		"maxburst %d words, LE, CCTL=0x%08x\n",
1206 		dma_chan_name(chan), plchan->name,
1207 		(config->direction == DMA_FROM_DEVICE) ? "RX" : "TX",
1208 		addr_width,
1209 		maxburst,
1210 		cctl);
1211 
1212 	return 0;
1213 }
1214 
1215 /*
1216  * Slave transactions callback to the slave device to allow
1217  * synchronization of slave DMA signals with the DMAC enable
1218  */
1219 static void pl08x_issue_pending(struct dma_chan *chan)
1220 {
1221 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1222 	unsigned long flags;
1223 
1224 	spin_lock_irqsave(&plchan->lock, flags);
1225 	/* Something is already active, or we're waiting for a channel... */
1226 	if (plchan->at || plchan->state == PL08X_CHAN_WAITING) {
1227 		spin_unlock_irqrestore(&plchan->lock, flags);
1228 		return;
1229 	}
1230 
1231 	/* Take the first element in the queue and execute it */
1232 	if (!list_empty(&plchan->pend_list)) {
1233 		struct pl08x_txd *next;
1234 
1235 		next = list_first_entry(&plchan->pend_list,
1236 					struct pl08x_txd,
1237 					node);
1238 		list_del(&next->node);
1239 		plchan->state = PL08X_CHAN_RUNNING;
1240 
1241 		pl08x_start_txd(plchan, next);
1242 	}
1243 
1244 	spin_unlock_irqrestore(&plchan->lock, flags);
1245 }
1246 
1247 static int pl08x_prep_channel_resources(struct pl08x_dma_chan *plchan,
1248 					struct pl08x_txd *txd)
1249 {
1250 	struct pl08x_driver_data *pl08x = plchan->host;
1251 	unsigned long flags;
1252 	int num_llis, ret;
1253 
1254 	num_llis = pl08x_fill_llis_for_desc(pl08x, txd);
1255 	if (!num_llis) {
1256 		kfree(txd);
1257 		return -EINVAL;
1258 	}
1259 
1260 	spin_lock_irqsave(&plchan->lock, flags);
1261 
1262 	/*
1263 	 * See if we already have a physical channel allocated,
1264 	 * else this is the time to try to get one.
1265 	 */
1266 	ret = prep_phy_channel(plchan, txd);
1267 	if (ret) {
1268 		/*
1269 		 * No physical channel was available.
1270 		 *
1271 		 * memcpy transfers can be sorted out at submission time.
1272 		 *
1273 		 * Slave transfers may have been denied due to platform
1274 		 * channel muxing restrictions.  Since there is no guarantee
1275 		 * that this will ever be resolved, and the signal must be
1276 		 * acquired AFTER acquiring the physical channel, we will let
1277 		 * them be NACK:ed with -EBUSY here. The drivers can retry
1278 		 * the prep() call if they are eager on doing this using DMA.
1279 		 */
1280 		if (plchan->slave) {
1281 			pl08x_free_txd_list(pl08x, plchan);
1282 			pl08x_free_txd(pl08x, txd);
1283 			spin_unlock_irqrestore(&plchan->lock, flags);
1284 			return -EBUSY;
1285 		}
1286 	} else
1287 		/*
1288 		 * Else we're all set, paused and ready to roll, status
1289 		 * will switch to PL08X_CHAN_RUNNING when we call
1290 		 * issue_pending(). If there is something running on the
1291 		 * channel already we don't change its state.
1292 		 */
1293 		if (plchan->state == PL08X_CHAN_IDLE)
1294 			plchan->state = PL08X_CHAN_PAUSED;
1295 
1296 	spin_unlock_irqrestore(&plchan->lock, flags);
1297 
1298 	return 0;
1299 }
1300 
1301 static struct pl08x_txd *pl08x_get_txd(struct pl08x_dma_chan *plchan,
1302 	unsigned long flags)
1303 {
1304 	struct pl08x_txd *txd = kzalloc(sizeof(struct pl08x_txd), GFP_NOWAIT);
1305 
1306 	if (txd) {
1307 		dma_async_tx_descriptor_init(&txd->tx, &plchan->chan);
1308 		txd->tx.flags = flags;
1309 		txd->tx.tx_submit = pl08x_tx_submit;
1310 		INIT_LIST_HEAD(&txd->node);
1311 
1312 		/* Always enable error and terminal interrupts */
1313 		txd->ccfg = PL080_CONFIG_ERR_IRQ_MASK |
1314 			    PL080_CONFIG_TC_IRQ_MASK;
1315 	}
1316 	return txd;
1317 }
1318 
1319 /*
1320  * Initialize a descriptor to be used by memcpy submit
1321  */
1322 static struct dma_async_tx_descriptor *pl08x_prep_dma_memcpy(
1323 		struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
1324 		size_t len, unsigned long flags)
1325 {
1326 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1327 	struct pl08x_driver_data *pl08x = plchan->host;
1328 	struct pl08x_txd *txd;
1329 	int ret;
1330 
1331 	txd = pl08x_get_txd(plchan, flags);
1332 	if (!txd) {
1333 		dev_err(&pl08x->adev->dev,
1334 			"%s no memory for descriptor\n", __func__);
1335 		return NULL;
1336 	}
1337 
1338 	txd->direction = DMA_NONE;
1339 	txd->src_addr = src;
1340 	txd->dst_addr = dest;
1341 	txd->len = len;
1342 
1343 	/* Set platform data for m2m */
1344 	txd->ccfg |= PL080_FLOW_MEM2MEM << PL080_CONFIG_FLOW_CONTROL_SHIFT;
1345 	txd->cctl = pl08x->pd->memcpy_channel.cctl &
1346 			~(PL080_CONTROL_DST_AHB2 | PL080_CONTROL_SRC_AHB2);
1347 
1348 	/* Both to be incremented or the code will break */
1349 	txd->cctl |= PL080_CONTROL_SRC_INCR | PL080_CONTROL_DST_INCR;
1350 
1351 	if (pl08x->vd->dualmaster)
1352 		txd->cctl |= pl08x_select_bus(pl08x->mem_buses,
1353 					      pl08x->mem_buses);
1354 
1355 	ret = pl08x_prep_channel_resources(plchan, txd);
1356 	if (ret)
1357 		return NULL;
1358 
1359 	return &txd->tx;
1360 }
1361 
1362 static struct dma_async_tx_descriptor *pl08x_prep_slave_sg(
1363 		struct dma_chan *chan, struct scatterlist *sgl,
1364 		unsigned int sg_len, enum dma_data_direction direction,
1365 		unsigned long flags)
1366 {
1367 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1368 	struct pl08x_driver_data *pl08x = plchan->host;
1369 	struct pl08x_txd *txd;
1370 	int ret;
1371 
1372 	/*
1373 	 * Current implementation ASSUMES only one sg
1374 	 */
1375 	if (sg_len != 1) {
1376 		dev_err(&pl08x->adev->dev, "%s prepared too long sglist\n",
1377 			__func__);
1378 		BUG();
1379 	}
1380 
1381 	dev_dbg(&pl08x->adev->dev, "%s prepare transaction of %d bytes from %s\n",
1382 		__func__, sgl->length, plchan->name);
1383 
1384 	txd = pl08x_get_txd(plchan, flags);
1385 	if (!txd) {
1386 		dev_err(&pl08x->adev->dev, "%s no txd\n", __func__);
1387 		return NULL;
1388 	}
1389 
1390 	if (direction != plchan->runtime_direction)
1391 		dev_err(&pl08x->adev->dev, "%s DMA setup does not match "
1392 			"the direction configured for the PrimeCell\n",
1393 			__func__);
1394 
1395 	/*
1396 	 * Set up addresses, the PrimeCell configured address
1397 	 * will take precedence since this may configure the
1398 	 * channel target address dynamically at runtime.
1399 	 */
1400 	txd->direction = direction;
1401 	txd->len = sgl->length;
1402 
1403 	if (direction == DMA_TO_DEVICE) {
1404 		txd->ccfg |= PL080_FLOW_MEM2PER << PL080_CONFIG_FLOW_CONTROL_SHIFT;
1405 		txd->cctl = plchan->dst_cctl;
1406 		txd->src_addr = sgl->dma_address;
1407 		txd->dst_addr = plchan->dst_addr;
1408 	} else if (direction == DMA_FROM_DEVICE) {
1409 		txd->ccfg |= PL080_FLOW_PER2MEM << PL080_CONFIG_FLOW_CONTROL_SHIFT;
1410 		txd->cctl = plchan->src_cctl;
1411 		txd->src_addr = plchan->src_addr;
1412 		txd->dst_addr = sgl->dma_address;
1413 	} else {
1414 		dev_err(&pl08x->adev->dev,
1415 			"%s direction unsupported\n", __func__);
1416 		return NULL;
1417 	}
1418 
1419 	ret = pl08x_prep_channel_resources(plchan, txd);
1420 	if (ret)
1421 		return NULL;
1422 
1423 	return &txd->tx;
1424 }
1425 
1426 static int pl08x_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
1427 			 unsigned long arg)
1428 {
1429 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1430 	struct pl08x_driver_data *pl08x = plchan->host;
1431 	unsigned long flags;
1432 	int ret = 0;
1433 
1434 	/* Controls applicable to inactive channels */
1435 	if (cmd == DMA_SLAVE_CONFIG) {
1436 		return dma_set_runtime_config(chan,
1437 					      (struct dma_slave_config *)arg);
1438 	}
1439 
1440 	/*
1441 	 * Anything succeeds on channels with no physical allocation and
1442 	 * no queued transfers.
1443 	 */
1444 	spin_lock_irqsave(&plchan->lock, flags);
1445 	if (!plchan->phychan && !plchan->at) {
1446 		spin_unlock_irqrestore(&plchan->lock, flags);
1447 		return 0;
1448 	}
1449 
1450 	switch (cmd) {
1451 	case DMA_TERMINATE_ALL:
1452 		plchan->state = PL08X_CHAN_IDLE;
1453 
1454 		if (plchan->phychan) {
1455 			pl08x_terminate_phy_chan(pl08x, plchan->phychan);
1456 
1457 			/*
1458 			 * Mark physical channel as free and free any slave
1459 			 * signal
1460 			 */
1461 			release_phy_channel(plchan);
1462 		}
1463 		/* Dequeue jobs and free LLIs */
1464 		if (plchan->at) {
1465 			pl08x_free_txd(pl08x, plchan->at);
1466 			plchan->at = NULL;
1467 		}
1468 		/* Dequeue jobs not yet fired as well */
1469 		pl08x_free_txd_list(pl08x, plchan);
1470 		break;
1471 	case DMA_PAUSE:
1472 		pl08x_pause_phy_chan(plchan->phychan);
1473 		plchan->state = PL08X_CHAN_PAUSED;
1474 		break;
1475 	case DMA_RESUME:
1476 		pl08x_resume_phy_chan(plchan->phychan);
1477 		plchan->state = PL08X_CHAN_RUNNING;
1478 		break;
1479 	default:
1480 		/* Unknown command */
1481 		ret = -ENXIO;
1482 		break;
1483 	}
1484 
1485 	spin_unlock_irqrestore(&plchan->lock, flags);
1486 
1487 	return ret;
1488 }
1489 
1490 bool pl08x_filter_id(struct dma_chan *chan, void *chan_id)
1491 {
1492 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1493 	char *name = chan_id;
1494 
1495 	/* Check that the channel is not taken! */
1496 	if (!strcmp(plchan->name, name))
1497 		return true;
1498 
1499 	return false;
1500 }
1501 
1502 /*
1503  * Just check that the device is there and active
1504  * TODO: turn this bit on/off depending on the number of physical channels
1505  * actually used, if it is zero... well shut it off. That will save some
1506  * power. Cut the clock at the same time.
1507  */
1508 static void pl08x_ensure_on(struct pl08x_driver_data *pl08x)
1509 {
1510 	u32 val;
1511 
1512 	val = readl(pl08x->base + PL080_CONFIG);
1513 	val &= ~(PL080_CONFIG_M2_BE | PL080_CONFIG_M1_BE | PL080_CONFIG_ENABLE);
1514 	/* We implicitly clear bit 1 and that means little-endian mode */
1515 	val |= PL080_CONFIG_ENABLE;
1516 	writel(val, pl08x->base + PL080_CONFIG);
1517 }
1518 
1519 static void pl08x_unmap_buffers(struct pl08x_txd *txd)
1520 {
1521 	struct device *dev = txd->tx.chan->device->dev;
1522 
1523 	if (!(txd->tx.flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
1524 		if (txd->tx.flags & DMA_COMPL_SRC_UNMAP_SINGLE)
1525 			dma_unmap_single(dev, txd->src_addr, txd->len,
1526 				DMA_TO_DEVICE);
1527 		else
1528 			dma_unmap_page(dev, txd->src_addr, txd->len,
1529 				DMA_TO_DEVICE);
1530 	}
1531 	if (!(txd->tx.flags & DMA_COMPL_SKIP_DEST_UNMAP)) {
1532 		if (txd->tx.flags & DMA_COMPL_DEST_UNMAP_SINGLE)
1533 			dma_unmap_single(dev, txd->dst_addr, txd->len,
1534 				DMA_FROM_DEVICE);
1535 		else
1536 			dma_unmap_page(dev, txd->dst_addr, txd->len,
1537 				DMA_FROM_DEVICE);
1538 	}
1539 }
1540 
1541 static void pl08x_tasklet(unsigned long data)
1542 {
1543 	struct pl08x_dma_chan *plchan = (struct pl08x_dma_chan *) data;
1544 	struct pl08x_driver_data *pl08x = plchan->host;
1545 	struct pl08x_txd *txd;
1546 	unsigned long flags;
1547 
1548 	spin_lock_irqsave(&plchan->lock, flags);
1549 
1550 	txd = plchan->at;
1551 	plchan->at = NULL;
1552 
1553 	if (txd) {
1554 		/* Update last completed */
1555 		plchan->lc = txd->tx.cookie;
1556 	}
1557 
1558 	/* If a new descriptor is queued, set it up plchan->at is NULL here */
1559 	if (!list_empty(&plchan->pend_list)) {
1560 		struct pl08x_txd *next;
1561 
1562 		next = list_first_entry(&plchan->pend_list,
1563 					struct pl08x_txd,
1564 					node);
1565 		list_del(&next->node);
1566 
1567 		pl08x_start_txd(plchan, next);
1568 	} else if (plchan->phychan_hold) {
1569 		/*
1570 		 * This channel is still in use - we have a new txd being
1571 		 * prepared and will soon be queued.  Don't give up the
1572 		 * physical channel.
1573 		 */
1574 	} else {
1575 		struct pl08x_dma_chan *waiting = NULL;
1576 
1577 		/*
1578 		 * No more jobs, so free up the physical channel
1579 		 * Free any allocated signal on slave transfers too
1580 		 */
1581 		release_phy_channel(plchan);
1582 		plchan->state = PL08X_CHAN_IDLE;
1583 
1584 		/*
1585 		 * And NOW before anyone else can grab that free:d up
1586 		 * physical channel, see if there is some memcpy pending
1587 		 * that seriously needs to start because of being stacked
1588 		 * up while we were choking the physical channels with data.
1589 		 */
1590 		list_for_each_entry(waiting, &pl08x->memcpy.channels,
1591 				    chan.device_node) {
1592 		  if (waiting->state == PL08X_CHAN_WAITING &&
1593 			    waiting->waiting != NULL) {
1594 				int ret;
1595 
1596 				/* This should REALLY not fail now */
1597 				ret = prep_phy_channel(waiting,
1598 						       waiting->waiting);
1599 				BUG_ON(ret);
1600 				waiting->phychan_hold--;
1601 				waiting->state = PL08X_CHAN_RUNNING;
1602 				waiting->waiting = NULL;
1603 				pl08x_issue_pending(&waiting->chan);
1604 				break;
1605 			}
1606 		}
1607 	}
1608 
1609 	spin_unlock_irqrestore(&plchan->lock, flags);
1610 
1611 	if (txd) {
1612 		dma_async_tx_callback callback = txd->tx.callback;
1613 		void *callback_param = txd->tx.callback_param;
1614 
1615 		/* Don't try to unmap buffers on slave channels */
1616 		if (!plchan->slave)
1617 			pl08x_unmap_buffers(txd);
1618 
1619 		/* Free the descriptor */
1620 		spin_lock_irqsave(&plchan->lock, flags);
1621 		pl08x_free_txd(pl08x, txd);
1622 		spin_unlock_irqrestore(&plchan->lock, flags);
1623 
1624 		/* Callback to signal completion */
1625 		if (callback)
1626 			callback(callback_param);
1627 	}
1628 }
1629 
1630 static irqreturn_t pl08x_irq(int irq, void *dev)
1631 {
1632 	struct pl08x_driver_data *pl08x = dev;
1633 	u32 mask = 0;
1634 	u32 val;
1635 	int i;
1636 
1637 	val = readl(pl08x->base + PL080_ERR_STATUS);
1638 	if (val) {
1639 		/* An error interrupt (on one or more channels) */
1640 		dev_err(&pl08x->adev->dev,
1641 			"%s error interrupt, register value 0x%08x\n",
1642 				__func__, val);
1643 		/*
1644 		 * Simply clear ALL PL08X error interrupts,
1645 		 * regardless of channel and cause
1646 		 * FIXME: should be 0x00000003 on PL081 really.
1647 		 */
1648 		writel(0x000000FF, pl08x->base + PL080_ERR_CLEAR);
1649 	}
1650 	val = readl(pl08x->base + PL080_INT_STATUS);
1651 	for (i = 0; i < pl08x->vd->channels; i++) {
1652 		if ((1 << i) & val) {
1653 			/* Locate physical channel */
1654 			struct pl08x_phy_chan *phychan = &pl08x->phy_chans[i];
1655 			struct pl08x_dma_chan *plchan = phychan->serving;
1656 
1657 			/* Schedule tasklet on this channel */
1658 			tasklet_schedule(&plchan->tasklet);
1659 
1660 			mask |= (1 << i);
1661 		}
1662 	}
1663 	/* Clear only the terminal interrupts on channels we processed */
1664 	writel(mask, pl08x->base + PL080_TC_CLEAR);
1665 
1666 	return mask ? IRQ_HANDLED : IRQ_NONE;
1667 }
1668 
1669 static void pl08x_dma_slave_init(struct pl08x_dma_chan *chan)
1670 {
1671 	u32 cctl = pl08x_cctl(chan->cd->cctl);
1672 
1673 	chan->slave = true;
1674 	chan->name = chan->cd->bus_id;
1675 	chan->src_addr = chan->cd->addr;
1676 	chan->dst_addr = chan->cd->addr;
1677 	chan->src_cctl = cctl | PL080_CONTROL_DST_INCR |
1678 		pl08x_select_bus(chan->cd->periph_buses, chan->host->mem_buses);
1679 	chan->dst_cctl = cctl | PL080_CONTROL_SRC_INCR |
1680 		pl08x_select_bus(chan->host->mem_buses, chan->cd->periph_buses);
1681 }
1682 
1683 /*
1684  * Initialise the DMAC memcpy/slave channels.
1685  * Make a local wrapper to hold required data
1686  */
1687 static int pl08x_dma_init_virtual_channels(struct pl08x_driver_data *pl08x,
1688 					   struct dma_device *dmadev,
1689 					   unsigned int channels,
1690 					   bool slave)
1691 {
1692 	struct pl08x_dma_chan *chan;
1693 	int i;
1694 
1695 	INIT_LIST_HEAD(&dmadev->channels);
1696 
1697 	/*
1698 	 * Register as many many memcpy as we have physical channels,
1699 	 * we won't always be able to use all but the code will have
1700 	 * to cope with that situation.
1701 	 */
1702 	for (i = 0; i < channels; i++) {
1703 		chan = kzalloc(sizeof(struct pl08x_dma_chan), GFP_KERNEL);
1704 		if (!chan) {
1705 			dev_err(&pl08x->adev->dev,
1706 				"%s no memory for channel\n", __func__);
1707 			return -ENOMEM;
1708 		}
1709 
1710 		chan->host = pl08x;
1711 		chan->state = PL08X_CHAN_IDLE;
1712 
1713 		if (slave) {
1714 			chan->cd = &pl08x->pd->slave_channels[i];
1715 			pl08x_dma_slave_init(chan);
1716 		} else {
1717 			chan->cd = &pl08x->pd->memcpy_channel;
1718 			chan->name = kasprintf(GFP_KERNEL, "memcpy%d", i);
1719 			if (!chan->name) {
1720 				kfree(chan);
1721 				return -ENOMEM;
1722 			}
1723 		}
1724 		if (chan->cd->circular_buffer) {
1725 			dev_err(&pl08x->adev->dev,
1726 				"channel %s: circular buffers not supported\n",
1727 				chan->name);
1728 			kfree(chan);
1729 			continue;
1730 		}
1731 		dev_info(&pl08x->adev->dev,
1732 			 "initialize virtual channel \"%s\"\n",
1733 			 chan->name);
1734 
1735 		chan->chan.device = dmadev;
1736 		chan->chan.cookie = 0;
1737 		chan->lc = 0;
1738 
1739 		spin_lock_init(&chan->lock);
1740 		INIT_LIST_HEAD(&chan->pend_list);
1741 		tasklet_init(&chan->tasklet, pl08x_tasklet,
1742 			     (unsigned long) chan);
1743 
1744 		list_add_tail(&chan->chan.device_node, &dmadev->channels);
1745 	}
1746 	dev_info(&pl08x->adev->dev, "initialized %d virtual %s channels\n",
1747 		 i, slave ? "slave" : "memcpy");
1748 	return i;
1749 }
1750 
1751 static void pl08x_free_virtual_channels(struct dma_device *dmadev)
1752 {
1753 	struct pl08x_dma_chan *chan = NULL;
1754 	struct pl08x_dma_chan *next;
1755 
1756 	list_for_each_entry_safe(chan,
1757 				 next, &dmadev->channels, chan.device_node) {
1758 		list_del(&chan->chan.device_node);
1759 		kfree(chan);
1760 	}
1761 }
1762 
1763 #ifdef CONFIG_DEBUG_FS
1764 static const char *pl08x_state_str(enum pl08x_dma_chan_state state)
1765 {
1766 	switch (state) {
1767 	case PL08X_CHAN_IDLE:
1768 		return "idle";
1769 	case PL08X_CHAN_RUNNING:
1770 		return "running";
1771 	case PL08X_CHAN_PAUSED:
1772 		return "paused";
1773 	case PL08X_CHAN_WAITING:
1774 		return "waiting";
1775 	default:
1776 		break;
1777 	}
1778 	return "UNKNOWN STATE";
1779 }
1780 
1781 static int pl08x_debugfs_show(struct seq_file *s, void *data)
1782 {
1783 	struct pl08x_driver_data *pl08x = s->private;
1784 	struct pl08x_dma_chan *chan;
1785 	struct pl08x_phy_chan *ch;
1786 	unsigned long flags;
1787 	int i;
1788 
1789 	seq_printf(s, "PL08x physical channels:\n");
1790 	seq_printf(s, "CHANNEL:\tUSER:\n");
1791 	seq_printf(s, "--------\t-----\n");
1792 	for (i = 0; i < pl08x->vd->channels; i++) {
1793 		struct pl08x_dma_chan *virt_chan;
1794 
1795 		ch = &pl08x->phy_chans[i];
1796 
1797 		spin_lock_irqsave(&ch->lock, flags);
1798 		virt_chan = ch->serving;
1799 
1800 		seq_printf(s, "%d\t\t%s\n",
1801 			   ch->id, virt_chan ? virt_chan->name : "(none)");
1802 
1803 		spin_unlock_irqrestore(&ch->lock, flags);
1804 	}
1805 
1806 	seq_printf(s, "\nPL08x virtual memcpy channels:\n");
1807 	seq_printf(s, "CHANNEL:\tSTATE:\n");
1808 	seq_printf(s, "--------\t------\n");
1809 	list_for_each_entry(chan, &pl08x->memcpy.channels, chan.device_node) {
1810 		seq_printf(s, "%s\t\t%s\n", chan->name,
1811 			   pl08x_state_str(chan->state));
1812 	}
1813 
1814 	seq_printf(s, "\nPL08x virtual slave channels:\n");
1815 	seq_printf(s, "CHANNEL:\tSTATE:\n");
1816 	seq_printf(s, "--------\t------\n");
1817 	list_for_each_entry(chan, &pl08x->slave.channels, chan.device_node) {
1818 		seq_printf(s, "%s\t\t%s\n", chan->name,
1819 			   pl08x_state_str(chan->state));
1820 	}
1821 
1822 	return 0;
1823 }
1824 
1825 static int pl08x_debugfs_open(struct inode *inode, struct file *file)
1826 {
1827 	return single_open(file, pl08x_debugfs_show, inode->i_private);
1828 }
1829 
1830 static const struct file_operations pl08x_debugfs_operations = {
1831 	.open		= pl08x_debugfs_open,
1832 	.read		= seq_read,
1833 	.llseek		= seq_lseek,
1834 	.release	= single_release,
1835 };
1836 
1837 static void init_pl08x_debugfs(struct pl08x_driver_data *pl08x)
1838 {
1839 	/* Expose a simple debugfs interface to view all clocks */
1840 	(void) debugfs_create_file(dev_name(&pl08x->adev->dev), S_IFREG | S_IRUGO,
1841 				   NULL, pl08x,
1842 				   &pl08x_debugfs_operations);
1843 }
1844 
1845 #else
1846 static inline void init_pl08x_debugfs(struct pl08x_driver_data *pl08x)
1847 {
1848 }
1849 #endif
1850 
1851 static int pl08x_probe(struct amba_device *adev, const struct amba_id *id)
1852 {
1853 	struct pl08x_driver_data *pl08x;
1854 	const struct vendor_data *vd = id->data;
1855 	int ret = 0;
1856 	int i;
1857 
1858 	ret = amba_request_regions(adev, NULL);
1859 	if (ret)
1860 		return ret;
1861 
1862 	/* Create the driver state holder */
1863 	pl08x = kzalloc(sizeof(struct pl08x_driver_data), GFP_KERNEL);
1864 	if (!pl08x) {
1865 		ret = -ENOMEM;
1866 		goto out_no_pl08x;
1867 	}
1868 
1869 	/* Initialize memcpy engine */
1870 	dma_cap_set(DMA_MEMCPY, pl08x->memcpy.cap_mask);
1871 	pl08x->memcpy.dev = &adev->dev;
1872 	pl08x->memcpy.device_alloc_chan_resources = pl08x_alloc_chan_resources;
1873 	pl08x->memcpy.device_free_chan_resources = pl08x_free_chan_resources;
1874 	pl08x->memcpy.device_prep_dma_memcpy = pl08x_prep_dma_memcpy;
1875 	pl08x->memcpy.device_prep_dma_interrupt = pl08x_prep_dma_interrupt;
1876 	pl08x->memcpy.device_tx_status = pl08x_dma_tx_status;
1877 	pl08x->memcpy.device_issue_pending = pl08x_issue_pending;
1878 	pl08x->memcpy.device_control = pl08x_control;
1879 
1880 	/* Initialize slave engine */
1881 	dma_cap_set(DMA_SLAVE, pl08x->slave.cap_mask);
1882 	pl08x->slave.dev = &adev->dev;
1883 	pl08x->slave.device_alloc_chan_resources = pl08x_alloc_chan_resources;
1884 	pl08x->slave.device_free_chan_resources = pl08x_free_chan_resources;
1885 	pl08x->slave.device_prep_dma_interrupt = pl08x_prep_dma_interrupt;
1886 	pl08x->slave.device_tx_status = pl08x_dma_tx_status;
1887 	pl08x->slave.device_issue_pending = pl08x_issue_pending;
1888 	pl08x->slave.device_prep_slave_sg = pl08x_prep_slave_sg;
1889 	pl08x->slave.device_control = pl08x_control;
1890 
1891 	/* Get the platform data */
1892 	pl08x->pd = dev_get_platdata(&adev->dev);
1893 	if (!pl08x->pd) {
1894 		dev_err(&adev->dev, "no platform data supplied\n");
1895 		goto out_no_platdata;
1896 	}
1897 
1898 	/* Assign useful pointers to the driver state */
1899 	pl08x->adev = adev;
1900 	pl08x->vd = vd;
1901 
1902 	/* By default, AHB1 only.  If dualmaster, from platform */
1903 	pl08x->lli_buses = PL08X_AHB1;
1904 	pl08x->mem_buses = PL08X_AHB1;
1905 	if (pl08x->vd->dualmaster) {
1906 		pl08x->lli_buses = pl08x->pd->lli_buses;
1907 		pl08x->mem_buses = pl08x->pd->mem_buses;
1908 	}
1909 
1910 	/* A DMA memory pool for LLIs, align on 1-byte boundary */
1911 	pl08x->pool = dma_pool_create(DRIVER_NAME, &pl08x->adev->dev,
1912 			PL08X_LLI_TSFR_SIZE, PL08X_ALIGN, 0);
1913 	if (!pl08x->pool) {
1914 		ret = -ENOMEM;
1915 		goto out_no_lli_pool;
1916 	}
1917 
1918 	spin_lock_init(&pl08x->lock);
1919 
1920 	pl08x->base = ioremap(adev->res.start, resource_size(&adev->res));
1921 	if (!pl08x->base) {
1922 		ret = -ENOMEM;
1923 		goto out_no_ioremap;
1924 	}
1925 
1926 	/* Turn on the PL08x */
1927 	pl08x_ensure_on(pl08x);
1928 
1929 	/* Attach the interrupt handler */
1930 	writel(0x000000FF, pl08x->base + PL080_ERR_CLEAR);
1931 	writel(0x000000FF, pl08x->base + PL080_TC_CLEAR);
1932 
1933 	ret = request_irq(adev->irq[0], pl08x_irq, IRQF_DISABLED,
1934 			  DRIVER_NAME, pl08x);
1935 	if (ret) {
1936 		dev_err(&adev->dev, "%s failed to request interrupt %d\n",
1937 			__func__, adev->irq[0]);
1938 		goto out_no_irq;
1939 	}
1940 
1941 	/* Initialize physical channels */
1942 	pl08x->phy_chans = kmalloc((vd->channels * sizeof(struct pl08x_phy_chan)),
1943 			GFP_KERNEL);
1944 	if (!pl08x->phy_chans) {
1945 		dev_err(&adev->dev, "%s failed to allocate "
1946 			"physical channel holders\n",
1947 			__func__);
1948 		goto out_no_phychans;
1949 	}
1950 
1951 	for (i = 0; i < vd->channels; i++) {
1952 		struct pl08x_phy_chan *ch = &pl08x->phy_chans[i];
1953 
1954 		ch->id = i;
1955 		ch->base = pl08x->base + PL080_Cx_BASE(i);
1956 		spin_lock_init(&ch->lock);
1957 		ch->serving = NULL;
1958 		ch->signal = -1;
1959 		dev_info(&adev->dev,
1960 			 "physical channel %d is %s\n", i,
1961 			 pl08x_phy_channel_busy(ch) ? "BUSY" : "FREE");
1962 	}
1963 
1964 	/* Register as many memcpy channels as there are physical channels */
1965 	ret = pl08x_dma_init_virtual_channels(pl08x, &pl08x->memcpy,
1966 					      pl08x->vd->channels, false);
1967 	if (ret <= 0) {
1968 		dev_warn(&pl08x->adev->dev,
1969 			 "%s failed to enumerate memcpy channels - %d\n",
1970 			 __func__, ret);
1971 		goto out_no_memcpy;
1972 	}
1973 	pl08x->memcpy.chancnt = ret;
1974 
1975 	/* Register slave channels */
1976 	ret = pl08x_dma_init_virtual_channels(pl08x, &pl08x->slave,
1977 					      pl08x->pd->num_slave_channels,
1978 					      true);
1979 	if (ret <= 0) {
1980 		dev_warn(&pl08x->adev->dev,
1981 			"%s failed to enumerate slave channels - %d\n",
1982 				__func__, ret);
1983 		goto out_no_slave;
1984 	}
1985 	pl08x->slave.chancnt = ret;
1986 
1987 	ret = dma_async_device_register(&pl08x->memcpy);
1988 	if (ret) {
1989 		dev_warn(&pl08x->adev->dev,
1990 			"%s failed to register memcpy as an async device - %d\n",
1991 			__func__, ret);
1992 		goto out_no_memcpy_reg;
1993 	}
1994 
1995 	ret = dma_async_device_register(&pl08x->slave);
1996 	if (ret) {
1997 		dev_warn(&pl08x->adev->dev,
1998 			"%s failed to register slave as an async device - %d\n",
1999 			__func__, ret);
2000 		goto out_no_slave_reg;
2001 	}
2002 
2003 	amba_set_drvdata(adev, pl08x);
2004 	init_pl08x_debugfs(pl08x);
2005 	dev_info(&pl08x->adev->dev, "DMA: PL%03x rev%u at 0x%08llx irq %d\n",
2006 		 amba_part(adev), amba_rev(adev),
2007 		 (unsigned long long)adev->res.start, adev->irq[0]);
2008 	return 0;
2009 
2010 out_no_slave_reg:
2011 	dma_async_device_unregister(&pl08x->memcpy);
2012 out_no_memcpy_reg:
2013 	pl08x_free_virtual_channels(&pl08x->slave);
2014 out_no_slave:
2015 	pl08x_free_virtual_channels(&pl08x->memcpy);
2016 out_no_memcpy:
2017 	kfree(pl08x->phy_chans);
2018 out_no_phychans:
2019 	free_irq(adev->irq[0], pl08x);
2020 out_no_irq:
2021 	iounmap(pl08x->base);
2022 out_no_ioremap:
2023 	dma_pool_destroy(pl08x->pool);
2024 out_no_lli_pool:
2025 out_no_platdata:
2026 	kfree(pl08x);
2027 out_no_pl08x:
2028 	amba_release_regions(adev);
2029 	return ret;
2030 }
2031 
2032 /* PL080 has 8 channels and the PL080 have just 2 */
2033 static struct vendor_data vendor_pl080 = {
2034 	.channels = 8,
2035 	.dualmaster = true,
2036 };
2037 
2038 static struct vendor_data vendor_pl081 = {
2039 	.channels = 2,
2040 	.dualmaster = false,
2041 };
2042 
2043 static struct amba_id pl08x_ids[] = {
2044 	/* PL080 */
2045 	{
2046 		.id	= 0x00041080,
2047 		.mask	= 0x000fffff,
2048 		.data	= &vendor_pl080,
2049 	},
2050 	/* PL081 */
2051 	{
2052 		.id	= 0x00041081,
2053 		.mask	= 0x000fffff,
2054 		.data	= &vendor_pl081,
2055 	},
2056 	/* Nomadik 8815 PL080 variant */
2057 	{
2058 		.id	= 0x00280880,
2059 		.mask	= 0x00ffffff,
2060 		.data	= &vendor_pl080,
2061 	},
2062 	{ 0, 0 },
2063 };
2064 
2065 static struct amba_driver pl08x_amba_driver = {
2066 	.drv.name	= DRIVER_NAME,
2067 	.id_table	= pl08x_ids,
2068 	.probe		= pl08x_probe,
2069 };
2070 
2071 static int __init pl08x_init(void)
2072 {
2073 	int retval;
2074 	retval = amba_driver_register(&pl08x_amba_driver);
2075 	if (retval)
2076 		printk(KERN_WARNING DRIVER_NAME
2077 		       "failed to register as an AMBA device (%d)\n",
2078 		       retval);
2079 	return retval;
2080 }
2081 subsys_initcall(pl08x_init);
2082