xref: /openbmc/linux/drivers/dma/amba-pl08x.c (revision a06c488d)
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  * The full GNU General Public License is in this distribution in the file
19  * called COPYING.
20  *
21  * Documentation: ARM DDI 0196G == PL080
22  * Documentation: ARM DDI 0218E == PL081
23  * Documentation: S3C6410 User's Manual == PL080S
24  *
25  * PL080 & PL081 both have 16 sets of DMA signals that can be routed to any
26  * channel.
27  *
28  * The PL080 has 8 channels available for simultaneous use, and the PL081
29  * has only two channels. So on these DMA controllers the number of channels
30  * and the number of incoming DMA signals are two totally different things.
31  * It is usually not possible to theoretically handle all physical signals,
32  * so a multiplexing scheme with possible denial of use is necessary.
33  *
34  * The PL080 has a dual bus master, PL081 has a single master.
35  *
36  * PL080S is a version modified by Samsung and used in S3C64xx SoCs.
37  * It differs in following aspects:
38  * - CH_CONFIG register at different offset,
39  * - separate CH_CONTROL2 register for transfer size,
40  * - bigger maximum transfer size,
41  * - 8-word aligned LLI, instead of 4-word, due to extra CCTL2 word,
42  * - no support for peripheral flow control.
43  *
44  * Memory to peripheral transfer may be visualized as
45  *	Get data from memory to DMAC
46  *	Until no data left
47  *		On burst request from peripheral
48  *			Destination burst from DMAC to peripheral
49  *			Clear burst request
50  *	Raise terminal count interrupt
51  *
52  * For peripherals with a FIFO:
53  * Source      burst size == half the depth of the peripheral FIFO
54  * Destination burst size == the depth of the peripheral FIFO
55  *
56  * (Bursts are irrelevant for mem to mem transfers - there are no burst
57  * signals, the DMA controller will simply facilitate its AHB master.)
58  *
59  * ASSUMES default (little) endianness for DMA transfers
60  *
61  * The PL08x has two flow control settings:
62  *  - DMAC flow control: the transfer size defines the number of transfers
63  *    which occur for the current LLI entry, and the DMAC raises TC at the
64  *    end of every LLI entry.  Observed behaviour shows the DMAC listening
65  *    to both the BREQ and SREQ signals (contrary to documented),
66  *    transferring data if either is active.  The LBREQ and LSREQ signals
67  *    are ignored.
68  *
69  *  - Peripheral flow control: the transfer size is ignored (and should be
70  *    zero).  The data is transferred from the current LLI entry, until
71  *    after the final transfer signalled by LBREQ or LSREQ.  The DMAC
72  *    will then move to the next LLI entry. Unsupported by PL080S.
73  */
74 #include <linux/amba/bus.h>
75 #include <linux/amba/pl08x.h>
76 #include <linux/debugfs.h>
77 #include <linux/delay.h>
78 #include <linux/device.h>
79 #include <linux/dmaengine.h>
80 #include <linux/dmapool.h>
81 #include <linux/dma-mapping.h>
82 #include <linux/export.h>
83 #include <linux/init.h>
84 #include <linux/interrupt.h>
85 #include <linux/module.h>
86 #include <linux/of.h>
87 #include <linux/of_dma.h>
88 #include <linux/pm_runtime.h>
89 #include <linux/seq_file.h>
90 #include <linux/slab.h>
91 #include <linux/amba/pl080.h>
92 
93 #include "dmaengine.h"
94 #include "virt-dma.h"
95 
96 #define DRIVER_NAME	"pl08xdmac"
97 
98 #define PL80X_DMA_BUSWIDTHS \
99 	BIT(DMA_SLAVE_BUSWIDTH_UNDEFINED) | \
100 	BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
101 	BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
102 	BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
103 
104 static struct amba_driver pl08x_amba_driver;
105 struct pl08x_driver_data;
106 
107 /**
108  * struct vendor_data - vendor-specific config parameters for PL08x derivatives
109  * @channels: the number of channels available in this variant
110  * @dualmaster: whether this version supports dual AHB masters or not.
111  * @nomadik: whether the channels have Nomadik security extension bits
112  *	that need to be checked for permission before use and some registers are
113  *	missing
114  * @pl080s: whether this version is a PL080S, which has separate register and
115  *	LLI word for transfer size.
116  */
117 struct vendor_data {
118 	u8 config_offset;
119 	u8 channels;
120 	bool dualmaster;
121 	bool nomadik;
122 	bool pl080s;
123 	u32 max_transfer_size;
124 };
125 
126 /**
127  * struct pl08x_bus_data - information of source or destination
128  * busses for a transfer
129  * @addr: current address
130  * @maxwidth: the maximum width of a transfer on this bus
131  * @buswidth: the width of this bus in bytes: 1, 2 or 4
132  */
133 struct pl08x_bus_data {
134 	dma_addr_t addr;
135 	u8 maxwidth;
136 	u8 buswidth;
137 };
138 
139 #define IS_BUS_ALIGNED(bus) IS_ALIGNED((bus)->addr, (bus)->buswidth)
140 
141 /**
142  * struct pl08x_phy_chan - holder for the physical channels
143  * @id: physical index to this channel
144  * @lock: a lock to use when altering an instance of this struct
145  * @serving: the virtual channel currently being served by this physical
146  * channel
147  * @locked: channel unavailable for the system, e.g. dedicated to secure
148  * world
149  */
150 struct pl08x_phy_chan {
151 	unsigned int id;
152 	void __iomem *base;
153 	void __iomem *reg_config;
154 	spinlock_t lock;
155 	struct pl08x_dma_chan *serving;
156 	bool locked;
157 };
158 
159 /**
160  * struct pl08x_sg - structure containing data per sg
161  * @src_addr: src address of sg
162  * @dst_addr: dst address of sg
163  * @len: transfer len in bytes
164  * @node: node for txd's dsg_list
165  */
166 struct pl08x_sg {
167 	dma_addr_t src_addr;
168 	dma_addr_t dst_addr;
169 	size_t len;
170 	struct list_head node;
171 };
172 
173 /**
174  * struct pl08x_txd - wrapper for struct dma_async_tx_descriptor
175  * @vd: virtual DMA descriptor
176  * @dsg_list: list of children sg's
177  * @llis_bus: DMA memory address (physical) start for the LLIs
178  * @llis_va: virtual memory address start for the LLIs
179  * @cctl: control reg values for current txd
180  * @ccfg: config reg values for current txd
181  * @done: this marks completed descriptors, which should not have their
182  *   mux released.
183  * @cyclic: indicate cyclic transfers
184  */
185 struct pl08x_txd {
186 	struct virt_dma_desc vd;
187 	struct list_head dsg_list;
188 	dma_addr_t llis_bus;
189 	u32 *llis_va;
190 	/* Default cctl value for LLIs */
191 	u32 cctl;
192 	/*
193 	 * Settings to be put into the physical channel when we
194 	 * trigger this txd.  Other registers are in llis_va[0].
195 	 */
196 	u32 ccfg;
197 	bool done;
198 	bool cyclic;
199 };
200 
201 /**
202  * struct pl08x_dma_chan_state - holds the PL08x specific virtual channel
203  * states
204  * @PL08X_CHAN_IDLE: the channel is idle
205  * @PL08X_CHAN_RUNNING: the channel has allocated a physical transport
206  * channel and is running a transfer on it
207  * @PL08X_CHAN_PAUSED: the channel has allocated a physical transport
208  * channel, but the transfer is currently paused
209  * @PL08X_CHAN_WAITING: the channel is waiting for a physical transport
210  * channel to become available (only pertains to memcpy channels)
211  */
212 enum pl08x_dma_chan_state {
213 	PL08X_CHAN_IDLE,
214 	PL08X_CHAN_RUNNING,
215 	PL08X_CHAN_PAUSED,
216 	PL08X_CHAN_WAITING,
217 };
218 
219 /**
220  * struct pl08x_dma_chan - this structure wraps a DMA ENGINE channel
221  * @vc: wrappped virtual channel
222  * @phychan: the physical channel utilized by this channel, if there is one
223  * @name: name of channel
224  * @cd: channel platform data
225  * @runtime_addr: address for RX/TX according to the runtime config
226  * @at: active transaction on this channel
227  * @lock: a lock for this channel data
228  * @host: a pointer to the host (internal use)
229  * @state: whether the channel is idle, paused, running etc
230  * @slave: whether this channel is a device (slave) or for memcpy
231  * @signal: the physical DMA request signal which this channel is using
232  * @mux_use: count of descriptors using this DMA request signal setting
233  */
234 struct pl08x_dma_chan {
235 	struct virt_dma_chan vc;
236 	struct pl08x_phy_chan *phychan;
237 	const char *name;
238 	const struct pl08x_channel_data *cd;
239 	struct dma_slave_config cfg;
240 	struct pl08x_txd *at;
241 	struct pl08x_driver_data *host;
242 	enum pl08x_dma_chan_state state;
243 	bool slave;
244 	int signal;
245 	unsigned mux_use;
246 };
247 
248 /**
249  * struct pl08x_driver_data - the local state holder for the PL08x
250  * @slave: slave engine for this instance
251  * @memcpy: memcpy engine for this instance
252  * @base: virtual memory base (remapped) for the PL08x
253  * @adev: the corresponding AMBA (PrimeCell) bus entry
254  * @vd: vendor data for this PL08x variant
255  * @pd: platform data passed in from the platform/machine
256  * @phy_chans: array of data for the physical channels
257  * @pool: a pool for the LLI descriptors
258  * @lli_buses: bitmask to or in to LLI pointer selecting AHB port for LLI
259  * fetches
260  * @mem_buses: set to indicate memory transfers on AHB2.
261  * @lock: a spinlock for this struct
262  */
263 struct pl08x_driver_data {
264 	struct dma_device slave;
265 	struct dma_device memcpy;
266 	void __iomem *base;
267 	struct amba_device *adev;
268 	const struct vendor_data *vd;
269 	struct pl08x_platform_data *pd;
270 	struct pl08x_phy_chan *phy_chans;
271 	struct dma_pool *pool;
272 	u8 lli_buses;
273 	u8 mem_buses;
274 	u8 lli_words;
275 };
276 
277 /*
278  * PL08X specific defines
279  */
280 
281 /* The order of words in an LLI. */
282 #define PL080_LLI_SRC		0
283 #define PL080_LLI_DST		1
284 #define PL080_LLI_LLI		2
285 #define PL080_LLI_CCTL		3
286 #define PL080S_LLI_CCTL2	4
287 
288 /* Total words in an LLI. */
289 #define PL080_LLI_WORDS		4
290 #define PL080S_LLI_WORDS	8
291 
292 /*
293  * Number of LLIs in each LLI buffer allocated for one transfer
294  * (maximum times we call dma_pool_alloc on this pool without freeing)
295  */
296 #define MAX_NUM_TSFR_LLIS	512
297 #define PL08X_ALIGN		8
298 
299 static inline struct pl08x_dma_chan *to_pl08x_chan(struct dma_chan *chan)
300 {
301 	return container_of(chan, struct pl08x_dma_chan, vc.chan);
302 }
303 
304 static inline struct pl08x_txd *to_pl08x_txd(struct dma_async_tx_descriptor *tx)
305 {
306 	return container_of(tx, struct pl08x_txd, vd.tx);
307 }
308 
309 /*
310  * Mux handling.
311  *
312  * This gives us the DMA request input to the PL08x primecell which the
313  * peripheral described by the channel data will be routed to, possibly
314  * via a board/SoC specific external MUX.  One important point to note
315  * here is that this does not depend on the physical channel.
316  */
317 static int pl08x_request_mux(struct pl08x_dma_chan *plchan)
318 {
319 	const struct pl08x_platform_data *pd = plchan->host->pd;
320 	int ret;
321 
322 	if (plchan->mux_use++ == 0 && pd->get_xfer_signal) {
323 		ret = pd->get_xfer_signal(plchan->cd);
324 		if (ret < 0) {
325 			plchan->mux_use = 0;
326 			return ret;
327 		}
328 
329 		plchan->signal = ret;
330 	}
331 	return 0;
332 }
333 
334 static void pl08x_release_mux(struct pl08x_dma_chan *plchan)
335 {
336 	const struct pl08x_platform_data *pd = plchan->host->pd;
337 
338 	if (plchan->signal >= 0) {
339 		WARN_ON(plchan->mux_use == 0);
340 
341 		if (--plchan->mux_use == 0 && pd->put_xfer_signal) {
342 			pd->put_xfer_signal(plchan->cd, plchan->signal);
343 			plchan->signal = -1;
344 		}
345 	}
346 }
347 
348 /*
349  * Physical channel handling
350  */
351 
352 /* Whether a certain channel is busy or not */
353 static int pl08x_phy_channel_busy(struct pl08x_phy_chan *ch)
354 {
355 	unsigned int val;
356 
357 	val = readl(ch->reg_config);
358 	return val & PL080_CONFIG_ACTIVE;
359 }
360 
361 static void pl08x_write_lli(struct pl08x_driver_data *pl08x,
362 		struct pl08x_phy_chan *phychan, const u32 *lli, u32 ccfg)
363 {
364 	if (pl08x->vd->pl080s)
365 		dev_vdbg(&pl08x->adev->dev,
366 			"WRITE channel %d: csrc=0x%08x, cdst=0x%08x, "
367 			"clli=0x%08x, cctl=0x%08x, cctl2=0x%08x, ccfg=0x%08x\n",
368 			phychan->id, lli[PL080_LLI_SRC], lli[PL080_LLI_DST],
369 			lli[PL080_LLI_LLI], lli[PL080_LLI_CCTL],
370 			lli[PL080S_LLI_CCTL2], ccfg);
371 	else
372 		dev_vdbg(&pl08x->adev->dev,
373 			"WRITE channel %d: csrc=0x%08x, cdst=0x%08x, "
374 			"clli=0x%08x, cctl=0x%08x, ccfg=0x%08x\n",
375 			phychan->id, lli[PL080_LLI_SRC], lli[PL080_LLI_DST],
376 			lli[PL080_LLI_LLI], lli[PL080_LLI_CCTL], ccfg);
377 
378 	writel_relaxed(lli[PL080_LLI_SRC], phychan->base + PL080_CH_SRC_ADDR);
379 	writel_relaxed(lli[PL080_LLI_DST], phychan->base + PL080_CH_DST_ADDR);
380 	writel_relaxed(lli[PL080_LLI_LLI], phychan->base + PL080_CH_LLI);
381 	writel_relaxed(lli[PL080_LLI_CCTL], phychan->base + PL080_CH_CONTROL);
382 
383 	if (pl08x->vd->pl080s)
384 		writel_relaxed(lli[PL080S_LLI_CCTL2],
385 				phychan->base + PL080S_CH_CONTROL2);
386 
387 	writel(ccfg, phychan->reg_config);
388 }
389 
390 /*
391  * Set the initial DMA register values i.e. those for the first LLI
392  * The next LLI pointer and the configuration interrupt bit have
393  * been set when the LLIs were constructed.  Poke them into the hardware
394  * and start the transfer.
395  */
396 static void pl08x_start_next_txd(struct pl08x_dma_chan *plchan)
397 {
398 	struct pl08x_driver_data *pl08x = plchan->host;
399 	struct pl08x_phy_chan *phychan = plchan->phychan;
400 	struct virt_dma_desc *vd = vchan_next_desc(&plchan->vc);
401 	struct pl08x_txd *txd = to_pl08x_txd(&vd->tx);
402 	u32 val;
403 
404 	list_del(&txd->vd.node);
405 
406 	plchan->at = txd;
407 
408 	/* Wait for channel inactive */
409 	while (pl08x_phy_channel_busy(phychan))
410 		cpu_relax();
411 
412 	pl08x_write_lli(pl08x, phychan, &txd->llis_va[0], txd->ccfg);
413 
414 	/* Enable the DMA channel */
415 	/* Do not access config register until channel shows as disabled */
416 	while (readl(pl08x->base + PL080_EN_CHAN) & (1 << phychan->id))
417 		cpu_relax();
418 
419 	/* Do not access config register until channel shows as inactive */
420 	val = readl(phychan->reg_config);
421 	while ((val & PL080_CONFIG_ACTIVE) || (val & PL080_CONFIG_ENABLE))
422 		val = readl(phychan->reg_config);
423 
424 	writel(val | PL080_CONFIG_ENABLE, phychan->reg_config);
425 }
426 
427 /*
428  * Pause the channel by setting the HALT bit.
429  *
430  * For M->P transfers, pause the DMAC first and then stop the peripheral -
431  * the FIFO can only drain if the peripheral is still requesting data.
432  * (note: this can still timeout if the DMAC FIFO never drains of data.)
433  *
434  * For P->M transfers, disable the peripheral first to stop it filling
435  * the DMAC FIFO, and then pause the DMAC.
436  */
437 static void pl08x_pause_phy_chan(struct pl08x_phy_chan *ch)
438 {
439 	u32 val;
440 	int timeout;
441 
442 	/* Set the HALT bit and wait for the FIFO to drain */
443 	val = readl(ch->reg_config);
444 	val |= PL080_CONFIG_HALT;
445 	writel(val, ch->reg_config);
446 
447 	/* Wait for channel inactive */
448 	for (timeout = 1000; timeout; timeout--) {
449 		if (!pl08x_phy_channel_busy(ch))
450 			break;
451 		udelay(1);
452 	}
453 	if (pl08x_phy_channel_busy(ch))
454 		pr_err("pl08x: channel%u timeout waiting for pause\n", ch->id);
455 }
456 
457 static void pl08x_resume_phy_chan(struct pl08x_phy_chan *ch)
458 {
459 	u32 val;
460 
461 	/* Clear the HALT bit */
462 	val = readl(ch->reg_config);
463 	val &= ~PL080_CONFIG_HALT;
464 	writel(val, ch->reg_config);
465 }
466 
467 /*
468  * pl08x_terminate_phy_chan() stops the channel, clears the FIFO and
469  * clears any pending interrupt status.  This should not be used for
470  * an on-going transfer, but as a method of shutting down a channel
471  * (eg, when it's no longer used) or terminating a transfer.
472  */
473 static void pl08x_terminate_phy_chan(struct pl08x_driver_data *pl08x,
474 	struct pl08x_phy_chan *ch)
475 {
476 	u32 val = readl(ch->reg_config);
477 
478 	val &= ~(PL080_CONFIG_ENABLE | PL080_CONFIG_ERR_IRQ_MASK |
479 		 PL080_CONFIG_TC_IRQ_MASK);
480 
481 	writel(val, ch->reg_config);
482 
483 	writel(1 << ch->id, pl08x->base + PL080_ERR_CLEAR);
484 	writel(1 << ch->id, pl08x->base + PL080_TC_CLEAR);
485 }
486 
487 static inline u32 get_bytes_in_cctl(u32 cctl)
488 {
489 	/* The source width defines the number of bytes */
490 	u32 bytes = cctl & PL080_CONTROL_TRANSFER_SIZE_MASK;
491 
492 	cctl &= PL080_CONTROL_SWIDTH_MASK;
493 
494 	switch (cctl >> PL080_CONTROL_SWIDTH_SHIFT) {
495 	case PL080_WIDTH_8BIT:
496 		break;
497 	case PL080_WIDTH_16BIT:
498 		bytes *= 2;
499 		break;
500 	case PL080_WIDTH_32BIT:
501 		bytes *= 4;
502 		break;
503 	}
504 	return bytes;
505 }
506 
507 static inline u32 get_bytes_in_cctl_pl080s(u32 cctl, u32 cctl1)
508 {
509 	/* The source width defines the number of bytes */
510 	u32 bytes = cctl1 & PL080S_CONTROL_TRANSFER_SIZE_MASK;
511 
512 	cctl &= PL080_CONTROL_SWIDTH_MASK;
513 
514 	switch (cctl >> PL080_CONTROL_SWIDTH_SHIFT) {
515 	case PL080_WIDTH_8BIT:
516 		break;
517 	case PL080_WIDTH_16BIT:
518 		bytes *= 2;
519 		break;
520 	case PL080_WIDTH_32BIT:
521 		bytes *= 4;
522 		break;
523 	}
524 	return bytes;
525 }
526 
527 /* The channel should be paused when calling this */
528 static u32 pl08x_getbytes_chan(struct pl08x_dma_chan *plchan)
529 {
530 	struct pl08x_driver_data *pl08x = plchan->host;
531 	const u32 *llis_va, *llis_va_limit;
532 	struct pl08x_phy_chan *ch;
533 	dma_addr_t llis_bus;
534 	struct pl08x_txd *txd;
535 	u32 llis_max_words;
536 	size_t bytes;
537 	u32 clli;
538 
539 	ch = plchan->phychan;
540 	txd = plchan->at;
541 
542 	if (!ch || !txd)
543 		return 0;
544 
545 	/*
546 	 * Follow the LLIs to get the number of remaining
547 	 * bytes in the currently active transaction.
548 	 */
549 	clli = readl(ch->base + PL080_CH_LLI) & ~PL080_LLI_LM_AHB2;
550 
551 	/* First get the remaining bytes in the active transfer */
552 	if (pl08x->vd->pl080s)
553 		bytes = get_bytes_in_cctl_pl080s(
554 				readl(ch->base + PL080_CH_CONTROL),
555 				readl(ch->base + PL080S_CH_CONTROL2));
556 	else
557 		bytes = get_bytes_in_cctl(readl(ch->base + PL080_CH_CONTROL));
558 
559 	if (!clli)
560 		return bytes;
561 
562 	llis_va = txd->llis_va;
563 	llis_bus = txd->llis_bus;
564 
565 	llis_max_words = pl08x->lli_words * MAX_NUM_TSFR_LLIS;
566 	BUG_ON(clli < llis_bus || clli >= llis_bus +
567 						sizeof(u32) * llis_max_words);
568 
569 	/*
570 	 * Locate the next LLI - as this is an array,
571 	 * it's simple maths to find.
572 	 */
573 	llis_va += (clli - llis_bus) / sizeof(u32);
574 
575 	llis_va_limit = llis_va + llis_max_words;
576 
577 	for (; llis_va < llis_va_limit; llis_va += pl08x->lli_words) {
578 		if (pl08x->vd->pl080s)
579 			bytes += get_bytes_in_cctl_pl080s(
580 						llis_va[PL080_LLI_CCTL],
581 						llis_va[PL080S_LLI_CCTL2]);
582 		else
583 			bytes += get_bytes_in_cctl(llis_va[PL080_LLI_CCTL]);
584 
585 		/*
586 		 * A LLI pointer going backward terminates the LLI list
587 		 */
588 		if (llis_va[PL080_LLI_LLI] <= clli)
589 			break;
590 	}
591 
592 	return bytes;
593 }
594 
595 /*
596  * Allocate a physical channel for a virtual channel
597  *
598  * Try to locate a physical channel to be used for this transfer. If all
599  * are taken return NULL and the requester will have to cope by using
600  * some fallback PIO mode or retrying later.
601  */
602 static struct pl08x_phy_chan *
603 pl08x_get_phy_channel(struct pl08x_driver_data *pl08x,
604 		      struct pl08x_dma_chan *virt_chan)
605 {
606 	struct pl08x_phy_chan *ch = NULL;
607 	unsigned long flags;
608 	int i;
609 
610 	for (i = 0; i < pl08x->vd->channels; i++) {
611 		ch = &pl08x->phy_chans[i];
612 
613 		spin_lock_irqsave(&ch->lock, flags);
614 
615 		if (!ch->locked && !ch->serving) {
616 			ch->serving = virt_chan;
617 			spin_unlock_irqrestore(&ch->lock, flags);
618 			break;
619 		}
620 
621 		spin_unlock_irqrestore(&ch->lock, flags);
622 	}
623 
624 	if (i == pl08x->vd->channels) {
625 		/* No physical channel available, cope with it */
626 		return NULL;
627 	}
628 
629 	return ch;
630 }
631 
632 /* Mark the physical channel as free.  Note, this write is atomic. */
633 static inline void pl08x_put_phy_channel(struct pl08x_driver_data *pl08x,
634 					 struct pl08x_phy_chan *ch)
635 {
636 	ch->serving = NULL;
637 }
638 
639 /*
640  * Try to allocate a physical channel.  When successful, assign it to
641  * this virtual channel, and initiate the next descriptor.  The
642  * virtual channel lock must be held at this point.
643  */
644 static void pl08x_phy_alloc_and_start(struct pl08x_dma_chan *plchan)
645 {
646 	struct pl08x_driver_data *pl08x = plchan->host;
647 	struct pl08x_phy_chan *ch;
648 
649 	ch = pl08x_get_phy_channel(pl08x, plchan);
650 	if (!ch) {
651 		dev_dbg(&pl08x->adev->dev, "no physical channel available for xfer on %s\n", plchan->name);
652 		plchan->state = PL08X_CHAN_WAITING;
653 		return;
654 	}
655 
656 	dev_dbg(&pl08x->adev->dev, "allocated physical channel %d for xfer on %s\n",
657 		ch->id, plchan->name);
658 
659 	plchan->phychan = ch;
660 	plchan->state = PL08X_CHAN_RUNNING;
661 	pl08x_start_next_txd(plchan);
662 }
663 
664 static void pl08x_phy_reassign_start(struct pl08x_phy_chan *ch,
665 	struct pl08x_dma_chan *plchan)
666 {
667 	struct pl08x_driver_data *pl08x = plchan->host;
668 
669 	dev_dbg(&pl08x->adev->dev, "reassigned physical channel %d for xfer on %s\n",
670 		ch->id, plchan->name);
671 
672 	/*
673 	 * We do this without taking the lock; we're really only concerned
674 	 * about whether this pointer is NULL or not, and we're guaranteed
675 	 * that this will only be called when it _already_ is non-NULL.
676 	 */
677 	ch->serving = plchan;
678 	plchan->phychan = ch;
679 	plchan->state = PL08X_CHAN_RUNNING;
680 	pl08x_start_next_txd(plchan);
681 }
682 
683 /*
684  * Free a physical DMA channel, potentially reallocating it to another
685  * virtual channel if we have any pending.
686  */
687 static void pl08x_phy_free(struct pl08x_dma_chan *plchan)
688 {
689 	struct pl08x_driver_data *pl08x = plchan->host;
690 	struct pl08x_dma_chan *p, *next;
691 
692  retry:
693 	next = NULL;
694 
695 	/* Find a waiting virtual channel for the next transfer. */
696 	list_for_each_entry(p, &pl08x->memcpy.channels, vc.chan.device_node)
697 		if (p->state == PL08X_CHAN_WAITING) {
698 			next = p;
699 			break;
700 		}
701 
702 	if (!next) {
703 		list_for_each_entry(p, &pl08x->slave.channels, vc.chan.device_node)
704 			if (p->state == PL08X_CHAN_WAITING) {
705 				next = p;
706 				break;
707 			}
708 	}
709 
710 	/* Ensure that the physical channel is stopped */
711 	pl08x_terminate_phy_chan(pl08x, plchan->phychan);
712 
713 	if (next) {
714 		bool success;
715 
716 		/*
717 		 * Eww.  We know this isn't going to deadlock
718 		 * but lockdep probably doesn't.
719 		 */
720 		spin_lock(&next->vc.lock);
721 		/* Re-check the state now that we have the lock */
722 		success = next->state == PL08X_CHAN_WAITING;
723 		if (success)
724 			pl08x_phy_reassign_start(plchan->phychan, next);
725 		spin_unlock(&next->vc.lock);
726 
727 		/* If the state changed, try to find another channel */
728 		if (!success)
729 			goto retry;
730 	} else {
731 		/* No more jobs, so free up the physical channel */
732 		pl08x_put_phy_channel(pl08x, plchan->phychan);
733 	}
734 
735 	plchan->phychan = NULL;
736 	plchan->state = PL08X_CHAN_IDLE;
737 }
738 
739 /*
740  * LLI handling
741  */
742 
743 static inline unsigned int pl08x_get_bytes_for_cctl(unsigned int coded)
744 {
745 	switch (coded) {
746 	case PL080_WIDTH_8BIT:
747 		return 1;
748 	case PL080_WIDTH_16BIT:
749 		return 2;
750 	case PL080_WIDTH_32BIT:
751 		return 4;
752 	default:
753 		break;
754 	}
755 	BUG();
756 	return 0;
757 }
758 
759 static inline u32 pl08x_cctl_bits(u32 cctl, u8 srcwidth, u8 dstwidth,
760 				  size_t tsize)
761 {
762 	u32 retbits = cctl;
763 
764 	/* Remove all src, dst and transfer size bits */
765 	retbits &= ~PL080_CONTROL_DWIDTH_MASK;
766 	retbits &= ~PL080_CONTROL_SWIDTH_MASK;
767 	retbits &= ~PL080_CONTROL_TRANSFER_SIZE_MASK;
768 
769 	/* Then set the bits according to the parameters */
770 	switch (srcwidth) {
771 	case 1:
772 		retbits |= PL080_WIDTH_8BIT << PL080_CONTROL_SWIDTH_SHIFT;
773 		break;
774 	case 2:
775 		retbits |= PL080_WIDTH_16BIT << PL080_CONTROL_SWIDTH_SHIFT;
776 		break;
777 	case 4:
778 		retbits |= PL080_WIDTH_32BIT << PL080_CONTROL_SWIDTH_SHIFT;
779 		break;
780 	default:
781 		BUG();
782 		break;
783 	}
784 
785 	switch (dstwidth) {
786 	case 1:
787 		retbits |= PL080_WIDTH_8BIT << PL080_CONTROL_DWIDTH_SHIFT;
788 		break;
789 	case 2:
790 		retbits |= PL080_WIDTH_16BIT << PL080_CONTROL_DWIDTH_SHIFT;
791 		break;
792 	case 4:
793 		retbits |= PL080_WIDTH_32BIT << PL080_CONTROL_DWIDTH_SHIFT;
794 		break;
795 	default:
796 		BUG();
797 		break;
798 	}
799 
800 	tsize &= PL080_CONTROL_TRANSFER_SIZE_MASK;
801 	retbits |= tsize << PL080_CONTROL_TRANSFER_SIZE_SHIFT;
802 	return retbits;
803 }
804 
805 struct pl08x_lli_build_data {
806 	struct pl08x_txd *txd;
807 	struct pl08x_bus_data srcbus;
808 	struct pl08x_bus_data dstbus;
809 	size_t remainder;
810 	u32 lli_bus;
811 };
812 
813 /*
814  * Autoselect a master bus to use for the transfer. Slave will be the chosen as
815  * victim in case src & dest are not similarly aligned. i.e. If after aligning
816  * masters address with width requirements of transfer (by sending few byte by
817  * byte data), slave is still not aligned, then its width will be reduced to
818  * BYTE.
819  * - prefers the destination bus if both available
820  * - prefers bus with fixed address (i.e. peripheral)
821  */
822 static void pl08x_choose_master_bus(struct pl08x_lli_build_data *bd,
823 	struct pl08x_bus_data **mbus, struct pl08x_bus_data **sbus, u32 cctl)
824 {
825 	if (!(cctl & PL080_CONTROL_DST_INCR)) {
826 		*mbus = &bd->dstbus;
827 		*sbus = &bd->srcbus;
828 	} else if (!(cctl & PL080_CONTROL_SRC_INCR)) {
829 		*mbus = &bd->srcbus;
830 		*sbus = &bd->dstbus;
831 	} else {
832 		if (bd->dstbus.buswidth >= bd->srcbus.buswidth) {
833 			*mbus = &bd->dstbus;
834 			*sbus = &bd->srcbus;
835 		} else {
836 			*mbus = &bd->srcbus;
837 			*sbus = &bd->dstbus;
838 		}
839 	}
840 }
841 
842 /*
843  * Fills in one LLI for a certain transfer descriptor and advance the counter
844  */
845 static void pl08x_fill_lli_for_desc(struct pl08x_driver_data *pl08x,
846 				    struct pl08x_lli_build_data *bd,
847 				    int num_llis, int len, u32 cctl, u32 cctl2)
848 {
849 	u32 offset = num_llis * pl08x->lli_words;
850 	u32 *llis_va = bd->txd->llis_va + offset;
851 	dma_addr_t llis_bus = bd->txd->llis_bus;
852 
853 	BUG_ON(num_llis >= MAX_NUM_TSFR_LLIS);
854 
855 	/* Advance the offset to next LLI. */
856 	offset += pl08x->lli_words;
857 
858 	llis_va[PL080_LLI_SRC] = bd->srcbus.addr;
859 	llis_va[PL080_LLI_DST] = bd->dstbus.addr;
860 	llis_va[PL080_LLI_LLI] = (llis_bus + sizeof(u32) * offset);
861 	llis_va[PL080_LLI_LLI] |= bd->lli_bus;
862 	llis_va[PL080_LLI_CCTL] = cctl;
863 	if (pl08x->vd->pl080s)
864 		llis_va[PL080S_LLI_CCTL2] = cctl2;
865 
866 	if (cctl & PL080_CONTROL_SRC_INCR)
867 		bd->srcbus.addr += len;
868 	if (cctl & PL080_CONTROL_DST_INCR)
869 		bd->dstbus.addr += len;
870 
871 	BUG_ON(bd->remainder < len);
872 
873 	bd->remainder -= len;
874 }
875 
876 static inline void prep_byte_width_lli(struct pl08x_driver_data *pl08x,
877 			struct pl08x_lli_build_data *bd, u32 *cctl, u32 len,
878 			int num_llis, size_t *total_bytes)
879 {
880 	*cctl = pl08x_cctl_bits(*cctl, 1, 1, len);
881 	pl08x_fill_lli_for_desc(pl08x, bd, num_llis, len, *cctl, len);
882 	(*total_bytes) += len;
883 }
884 
885 #ifdef VERBOSE_DEBUG
886 static void pl08x_dump_lli(struct pl08x_driver_data *pl08x,
887 			   const u32 *llis_va, int num_llis)
888 {
889 	int i;
890 
891 	if (pl08x->vd->pl080s) {
892 		dev_vdbg(&pl08x->adev->dev,
893 			"%-3s %-9s  %-10s %-10s %-10s %-10s %s\n",
894 			"lli", "", "csrc", "cdst", "clli", "cctl", "cctl2");
895 		for (i = 0; i < num_llis; i++) {
896 			dev_vdbg(&pl08x->adev->dev,
897 				"%3d @%p: 0x%08x 0x%08x 0x%08x 0x%08x 0x%08x\n",
898 				i, llis_va, llis_va[PL080_LLI_SRC],
899 				llis_va[PL080_LLI_DST], llis_va[PL080_LLI_LLI],
900 				llis_va[PL080_LLI_CCTL],
901 				llis_va[PL080S_LLI_CCTL2]);
902 			llis_va += pl08x->lli_words;
903 		}
904 	} else {
905 		dev_vdbg(&pl08x->adev->dev,
906 			"%-3s %-9s  %-10s %-10s %-10s %s\n",
907 			"lli", "", "csrc", "cdst", "clli", "cctl");
908 		for (i = 0; i < num_llis; i++) {
909 			dev_vdbg(&pl08x->adev->dev,
910 				"%3d @%p: 0x%08x 0x%08x 0x%08x 0x%08x\n",
911 				i, llis_va, llis_va[PL080_LLI_SRC],
912 				llis_va[PL080_LLI_DST], llis_va[PL080_LLI_LLI],
913 				llis_va[PL080_LLI_CCTL]);
914 			llis_va += pl08x->lli_words;
915 		}
916 	}
917 }
918 #else
919 static inline void pl08x_dump_lli(struct pl08x_driver_data *pl08x,
920 				  const u32 *llis_va, int num_llis) {}
921 #endif
922 
923 /*
924  * This fills in the table of LLIs for the transfer descriptor
925  * Note that we assume we never have to change the burst sizes
926  * Return 0 for error
927  */
928 static int pl08x_fill_llis_for_desc(struct pl08x_driver_data *pl08x,
929 			      struct pl08x_txd *txd)
930 {
931 	struct pl08x_bus_data *mbus, *sbus;
932 	struct pl08x_lli_build_data bd;
933 	int num_llis = 0;
934 	u32 cctl, early_bytes = 0;
935 	size_t max_bytes_per_lli, total_bytes;
936 	u32 *llis_va, *last_lli;
937 	struct pl08x_sg *dsg;
938 
939 	txd->llis_va = dma_pool_alloc(pl08x->pool, GFP_NOWAIT, &txd->llis_bus);
940 	if (!txd->llis_va) {
941 		dev_err(&pl08x->adev->dev, "%s no memory for llis\n", __func__);
942 		return 0;
943 	}
944 
945 	bd.txd = txd;
946 	bd.lli_bus = (pl08x->lli_buses & PL08X_AHB2) ? PL080_LLI_LM_AHB2 : 0;
947 	cctl = txd->cctl;
948 
949 	/* Find maximum width of the source bus */
950 	bd.srcbus.maxwidth =
951 		pl08x_get_bytes_for_cctl((cctl & PL080_CONTROL_SWIDTH_MASK) >>
952 				       PL080_CONTROL_SWIDTH_SHIFT);
953 
954 	/* Find maximum width of the destination bus */
955 	bd.dstbus.maxwidth =
956 		pl08x_get_bytes_for_cctl((cctl & PL080_CONTROL_DWIDTH_MASK) >>
957 				       PL080_CONTROL_DWIDTH_SHIFT);
958 
959 	list_for_each_entry(dsg, &txd->dsg_list, node) {
960 		total_bytes = 0;
961 		cctl = txd->cctl;
962 
963 		bd.srcbus.addr = dsg->src_addr;
964 		bd.dstbus.addr = dsg->dst_addr;
965 		bd.remainder = dsg->len;
966 		bd.srcbus.buswidth = bd.srcbus.maxwidth;
967 		bd.dstbus.buswidth = bd.dstbus.maxwidth;
968 
969 		pl08x_choose_master_bus(&bd, &mbus, &sbus, cctl);
970 
971 		dev_vdbg(&pl08x->adev->dev,
972 			"src=0x%08llx%s/%u dst=0x%08llx%s/%u len=%zu\n",
973 			(u64)bd.srcbus.addr,
974 			cctl & PL080_CONTROL_SRC_INCR ? "+" : "",
975 			bd.srcbus.buswidth,
976 			(u64)bd.dstbus.addr,
977 			cctl & PL080_CONTROL_DST_INCR ? "+" : "",
978 			bd.dstbus.buswidth,
979 			bd.remainder);
980 		dev_vdbg(&pl08x->adev->dev, "mbus=%s sbus=%s\n",
981 			mbus == &bd.srcbus ? "src" : "dst",
982 			sbus == &bd.srcbus ? "src" : "dst");
983 
984 		/*
985 		 * Zero length is only allowed if all these requirements are
986 		 * met:
987 		 * - flow controller is peripheral.
988 		 * - src.addr is aligned to src.width
989 		 * - dst.addr is aligned to dst.width
990 		 *
991 		 * sg_len == 1 should be true, as there can be two cases here:
992 		 *
993 		 * - Memory addresses are contiguous and are not scattered.
994 		 *   Here, Only one sg will be passed by user driver, with
995 		 *   memory address and zero length. We pass this to controller
996 		 *   and after the transfer it will receive the last burst
997 		 *   request from peripheral and so transfer finishes.
998 		 *
999 		 * - Memory addresses are scattered and are not contiguous.
1000 		 *   Here, Obviously as DMA controller doesn't know when a lli's
1001 		 *   transfer gets over, it can't load next lli. So in this
1002 		 *   case, there has to be an assumption that only one lli is
1003 		 *   supported. Thus, we can't have scattered addresses.
1004 		 */
1005 		if (!bd.remainder) {
1006 			u32 fc = (txd->ccfg & PL080_CONFIG_FLOW_CONTROL_MASK) >>
1007 				PL080_CONFIG_FLOW_CONTROL_SHIFT;
1008 			if (!((fc >= PL080_FLOW_SRC2DST_DST) &&
1009 					(fc <= PL080_FLOW_SRC2DST_SRC))) {
1010 				dev_err(&pl08x->adev->dev, "%s sg len can't be zero",
1011 					__func__);
1012 				return 0;
1013 			}
1014 
1015 			if (!IS_BUS_ALIGNED(&bd.srcbus) ||
1016 				!IS_BUS_ALIGNED(&bd.dstbus)) {
1017 				dev_err(&pl08x->adev->dev,
1018 					"%s src & dst address must be aligned to src"
1019 					" & dst width if peripheral is flow controller",
1020 					__func__);
1021 				return 0;
1022 			}
1023 
1024 			cctl = pl08x_cctl_bits(cctl, bd.srcbus.buswidth,
1025 					bd.dstbus.buswidth, 0);
1026 			pl08x_fill_lli_for_desc(pl08x, &bd, num_llis++,
1027 					0, cctl, 0);
1028 			break;
1029 		}
1030 
1031 		/*
1032 		 * Send byte by byte for following cases
1033 		 * - Less than a bus width available
1034 		 * - until master bus is aligned
1035 		 */
1036 		if (bd.remainder < mbus->buswidth)
1037 			early_bytes = bd.remainder;
1038 		else if (!IS_BUS_ALIGNED(mbus)) {
1039 			early_bytes = mbus->buswidth -
1040 				(mbus->addr & (mbus->buswidth - 1));
1041 			if ((bd.remainder - early_bytes) < mbus->buswidth)
1042 				early_bytes = bd.remainder;
1043 		}
1044 
1045 		if (early_bytes) {
1046 			dev_vdbg(&pl08x->adev->dev,
1047 				"%s byte width LLIs (remain 0x%08zx)\n",
1048 				__func__, bd.remainder);
1049 			prep_byte_width_lli(pl08x, &bd, &cctl, early_bytes,
1050 				num_llis++, &total_bytes);
1051 		}
1052 
1053 		if (bd.remainder) {
1054 			/*
1055 			 * Master now aligned
1056 			 * - if slave is not then we must set its width down
1057 			 */
1058 			if (!IS_BUS_ALIGNED(sbus)) {
1059 				dev_dbg(&pl08x->adev->dev,
1060 					"%s set down bus width to one byte\n",
1061 					__func__);
1062 
1063 				sbus->buswidth = 1;
1064 			}
1065 
1066 			/*
1067 			 * Bytes transferred = tsize * src width, not
1068 			 * MIN(buswidths)
1069 			 */
1070 			max_bytes_per_lli = bd.srcbus.buswidth *
1071 						pl08x->vd->max_transfer_size;
1072 			dev_vdbg(&pl08x->adev->dev,
1073 				"%s max bytes per lli = %zu\n",
1074 				__func__, max_bytes_per_lli);
1075 
1076 			/*
1077 			 * Make largest possible LLIs until less than one bus
1078 			 * width left
1079 			 */
1080 			while (bd.remainder > (mbus->buswidth - 1)) {
1081 				size_t lli_len, tsize, width;
1082 
1083 				/*
1084 				 * If enough left try to send max possible,
1085 				 * otherwise try to send the remainder
1086 				 */
1087 				lli_len = min(bd.remainder, max_bytes_per_lli);
1088 
1089 				/*
1090 				 * Check against maximum bus alignment:
1091 				 * Calculate actual transfer size in relation to
1092 				 * bus width an get a maximum remainder of the
1093 				 * highest bus width - 1
1094 				 */
1095 				width = max(mbus->buswidth, sbus->buswidth);
1096 				lli_len = (lli_len / width) * width;
1097 				tsize = lli_len / bd.srcbus.buswidth;
1098 
1099 				dev_vdbg(&pl08x->adev->dev,
1100 					"%s fill lli with single lli chunk of "
1101 					"size 0x%08zx (remainder 0x%08zx)\n",
1102 					__func__, lli_len, bd.remainder);
1103 
1104 				cctl = pl08x_cctl_bits(cctl, bd.srcbus.buswidth,
1105 					bd.dstbus.buswidth, tsize);
1106 				pl08x_fill_lli_for_desc(pl08x, &bd, num_llis++,
1107 						lli_len, cctl, tsize);
1108 				total_bytes += lli_len;
1109 			}
1110 
1111 			/*
1112 			 * Send any odd bytes
1113 			 */
1114 			if (bd.remainder) {
1115 				dev_vdbg(&pl08x->adev->dev,
1116 					"%s align with boundary, send odd bytes (remain %zu)\n",
1117 					__func__, bd.remainder);
1118 				prep_byte_width_lli(pl08x, &bd, &cctl,
1119 					bd.remainder, num_llis++, &total_bytes);
1120 			}
1121 		}
1122 
1123 		if (total_bytes != dsg->len) {
1124 			dev_err(&pl08x->adev->dev,
1125 				"%s size of encoded lli:s don't match total txd, transferred 0x%08zx from size 0x%08zx\n",
1126 				__func__, total_bytes, dsg->len);
1127 			return 0;
1128 		}
1129 
1130 		if (num_llis >= MAX_NUM_TSFR_LLIS) {
1131 			dev_err(&pl08x->adev->dev,
1132 				"%s need to increase MAX_NUM_TSFR_LLIS from 0x%08x\n",
1133 				__func__, MAX_NUM_TSFR_LLIS);
1134 			return 0;
1135 		}
1136 	}
1137 
1138 	llis_va = txd->llis_va;
1139 	last_lli = llis_va + (num_llis - 1) * pl08x->lli_words;
1140 
1141 	if (txd->cyclic) {
1142 		/* Link back to the first LLI. */
1143 		last_lli[PL080_LLI_LLI] = txd->llis_bus | bd.lli_bus;
1144 	} else {
1145 		/* The final LLI terminates the LLI. */
1146 		last_lli[PL080_LLI_LLI] = 0;
1147 		/* The final LLI element shall also fire an interrupt. */
1148 		last_lli[PL080_LLI_CCTL] |= PL080_CONTROL_TC_IRQ_EN;
1149 	}
1150 
1151 	pl08x_dump_lli(pl08x, llis_va, num_llis);
1152 
1153 	return num_llis;
1154 }
1155 
1156 static void pl08x_free_txd(struct pl08x_driver_data *pl08x,
1157 			   struct pl08x_txd *txd)
1158 {
1159 	struct pl08x_sg *dsg, *_dsg;
1160 
1161 	if (txd->llis_va)
1162 		dma_pool_free(pl08x->pool, txd->llis_va, txd->llis_bus);
1163 
1164 	list_for_each_entry_safe(dsg, _dsg, &txd->dsg_list, node) {
1165 		list_del(&dsg->node);
1166 		kfree(dsg);
1167 	}
1168 
1169 	kfree(txd);
1170 }
1171 
1172 static void pl08x_desc_free(struct virt_dma_desc *vd)
1173 {
1174 	struct pl08x_txd *txd = to_pl08x_txd(&vd->tx);
1175 	struct pl08x_dma_chan *plchan = to_pl08x_chan(vd->tx.chan);
1176 
1177 	dma_descriptor_unmap(&vd->tx);
1178 	if (!txd->done)
1179 		pl08x_release_mux(plchan);
1180 
1181 	pl08x_free_txd(plchan->host, txd);
1182 }
1183 
1184 static void pl08x_free_txd_list(struct pl08x_driver_data *pl08x,
1185 				struct pl08x_dma_chan *plchan)
1186 {
1187 	LIST_HEAD(head);
1188 
1189 	vchan_get_all_descriptors(&plchan->vc, &head);
1190 	vchan_dma_desc_free_list(&plchan->vc, &head);
1191 }
1192 
1193 /*
1194  * The DMA ENGINE API
1195  */
1196 static void pl08x_free_chan_resources(struct dma_chan *chan)
1197 {
1198 	/* Ensure all queued descriptors are freed */
1199 	vchan_free_chan_resources(to_virt_chan(chan));
1200 }
1201 
1202 static struct dma_async_tx_descriptor *pl08x_prep_dma_interrupt(
1203 		struct dma_chan *chan, unsigned long flags)
1204 {
1205 	struct dma_async_tx_descriptor *retval = NULL;
1206 
1207 	return retval;
1208 }
1209 
1210 /*
1211  * Code accessing dma_async_is_complete() in a tight loop may give problems.
1212  * If slaves are relying on interrupts to signal completion this function
1213  * must not be called with interrupts disabled.
1214  */
1215 static enum dma_status pl08x_dma_tx_status(struct dma_chan *chan,
1216 		dma_cookie_t cookie, struct dma_tx_state *txstate)
1217 {
1218 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1219 	struct virt_dma_desc *vd;
1220 	unsigned long flags;
1221 	enum dma_status ret;
1222 	size_t bytes = 0;
1223 
1224 	ret = dma_cookie_status(chan, cookie, txstate);
1225 	if (ret == DMA_COMPLETE)
1226 		return ret;
1227 
1228 	/*
1229 	 * There's no point calculating the residue if there's
1230 	 * no txstate to store the value.
1231 	 */
1232 	if (!txstate) {
1233 		if (plchan->state == PL08X_CHAN_PAUSED)
1234 			ret = DMA_PAUSED;
1235 		return ret;
1236 	}
1237 
1238 	spin_lock_irqsave(&plchan->vc.lock, flags);
1239 	ret = dma_cookie_status(chan, cookie, txstate);
1240 	if (ret != DMA_COMPLETE) {
1241 		vd = vchan_find_desc(&plchan->vc, cookie);
1242 		if (vd) {
1243 			/* On the issued list, so hasn't been processed yet */
1244 			struct pl08x_txd *txd = to_pl08x_txd(&vd->tx);
1245 			struct pl08x_sg *dsg;
1246 
1247 			list_for_each_entry(dsg, &txd->dsg_list, node)
1248 				bytes += dsg->len;
1249 		} else {
1250 			bytes = pl08x_getbytes_chan(plchan);
1251 		}
1252 	}
1253 	spin_unlock_irqrestore(&plchan->vc.lock, flags);
1254 
1255 	/*
1256 	 * This cookie not complete yet
1257 	 * Get number of bytes left in the active transactions and queue
1258 	 */
1259 	dma_set_residue(txstate, bytes);
1260 
1261 	if (plchan->state == PL08X_CHAN_PAUSED && ret == DMA_IN_PROGRESS)
1262 		ret = DMA_PAUSED;
1263 
1264 	/* Whether waiting or running, we're in progress */
1265 	return ret;
1266 }
1267 
1268 /* PrimeCell DMA extension */
1269 struct burst_table {
1270 	u32 burstwords;
1271 	u32 reg;
1272 };
1273 
1274 static const struct burst_table burst_sizes[] = {
1275 	{
1276 		.burstwords = 256,
1277 		.reg = PL080_BSIZE_256,
1278 	},
1279 	{
1280 		.burstwords = 128,
1281 		.reg = PL080_BSIZE_128,
1282 	},
1283 	{
1284 		.burstwords = 64,
1285 		.reg = PL080_BSIZE_64,
1286 	},
1287 	{
1288 		.burstwords = 32,
1289 		.reg = PL080_BSIZE_32,
1290 	},
1291 	{
1292 		.burstwords = 16,
1293 		.reg = PL080_BSIZE_16,
1294 	},
1295 	{
1296 		.burstwords = 8,
1297 		.reg = PL080_BSIZE_8,
1298 	},
1299 	{
1300 		.burstwords = 4,
1301 		.reg = PL080_BSIZE_4,
1302 	},
1303 	{
1304 		.burstwords = 0,
1305 		.reg = PL080_BSIZE_1,
1306 	},
1307 };
1308 
1309 /*
1310  * Given the source and destination available bus masks, select which
1311  * will be routed to each port.  We try to have source and destination
1312  * on separate ports, but always respect the allowable settings.
1313  */
1314 static u32 pl08x_select_bus(u8 src, u8 dst)
1315 {
1316 	u32 cctl = 0;
1317 
1318 	if (!(dst & PL08X_AHB1) || ((dst & PL08X_AHB2) && (src & PL08X_AHB1)))
1319 		cctl |= PL080_CONTROL_DST_AHB2;
1320 	if (!(src & PL08X_AHB1) || ((src & PL08X_AHB2) && !(dst & PL08X_AHB2)))
1321 		cctl |= PL080_CONTROL_SRC_AHB2;
1322 
1323 	return cctl;
1324 }
1325 
1326 static u32 pl08x_cctl(u32 cctl)
1327 {
1328 	cctl &= ~(PL080_CONTROL_SRC_AHB2 | PL080_CONTROL_DST_AHB2 |
1329 		  PL080_CONTROL_SRC_INCR | PL080_CONTROL_DST_INCR |
1330 		  PL080_CONTROL_PROT_MASK);
1331 
1332 	/* Access the cell in privileged mode, non-bufferable, non-cacheable */
1333 	return cctl | PL080_CONTROL_PROT_SYS;
1334 }
1335 
1336 static u32 pl08x_width(enum dma_slave_buswidth width)
1337 {
1338 	switch (width) {
1339 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
1340 		return PL080_WIDTH_8BIT;
1341 	case DMA_SLAVE_BUSWIDTH_2_BYTES:
1342 		return PL080_WIDTH_16BIT;
1343 	case DMA_SLAVE_BUSWIDTH_4_BYTES:
1344 		return PL080_WIDTH_32BIT;
1345 	default:
1346 		return ~0;
1347 	}
1348 }
1349 
1350 static u32 pl08x_burst(u32 maxburst)
1351 {
1352 	int i;
1353 
1354 	for (i = 0; i < ARRAY_SIZE(burst_sizes); i++)
1355 		if (burst_sizes[i].burstwords <= maxburst)
1356 			break;
1357 
1358 	return burst_sizes[i].reg;
1359 }
1360 
1361 static u32 pl08x_get_cctl(struct pl08x_dma_chan *plchan,
1362 	enum dma_slave_buswidth addr_width, u32 maxburst)
1363 {
1364 	u32 width, burst, cctl = 0;
1365 
1366 	width = pl08x_width(addr_width);
1367 	if (width == ~0)
1368 		return ~0;
1369 
1370 	cctl |= width << PL080_CONTROL_SWIDTH_SHIFT;
1371 	cctl |= width << PL080_CONTROL_DWIDTH_SHIFT;
1372 
1373 	/*
1374 	 * If this channel will only request single transfers, set this
1375 	 * down to ONE element.  Also select one element if no maxburst
1376 	 * is specified.
1377 	 */
1378 	if (plchan->cd->single)
1379 		maxburst = 1;
1380 
1381 	burst = pl08x_burst(maxburst);
1382 	cctl |= burst << PL080_CONTROL_SB_SIZE_SHIFT;
1383 	cctl |= burst << PL080_CONTROL_DB_SIZE_SHIFT;
1384 
1385 	return pl08x_cctl(cctl);
1386 }
1387 
1388 /*
1389  * Slave transactions callback to the slave device to allow
1390  * synchronization of slave DMA signals with the DMAC enable
1391  */
1392 static void pl08x_issue_pending(struct dma_chan *chan)
1393 {
1394 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1395 	unsigned long flags;
1396 
1397 	spin_lock_irqsave(&plchan->vc.lock, flags);
1398 	if (vchan_issue_pending(&plchan->vc)) {
1399 		if (!plchan->phychan && plchan->state != PL08X_CHAN_WAITING)
1400 			pl08x_phy_alloc_and_start(plchan);
1401 	}
1402 	spin_unlock_irqrestore(&plchan->vc.lock, flags);
1403 }
1404 
1405 static struct pl08x_txd *pl08x_get_txd(struct pl08x_dma_chan *plchan)
1406 {
1407 	struct pl08x_txd *txd = kzalloc(sizeof(*txd), GFP_NOWAIT);
1408 
1409 	if (txd) {
1410 		INIT_LIST_HEAD(&txd->dsg_list);
1411 
1412 		/* Always enable error and terminal interrupts */
1413 		txd->ccfg = PL080_CONFIG_ERR_IRQ_MASK |
1414 			    PL080_CONFIG_TC_IRQ_MASK;
1415 	}
1416 	return txd;
1417 }
1418 
1419 /*
1420  * Initialize a descriptor to be used by memcpy submit
1421  */
1422 static struct dma_async_tx_descriptor *pl08x_prep_dma_memcpy(
1423 		struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
1424 		size_t len, unsigned long flags)
1425 {
1426 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1427 	struct pl08x_driver_data *pl08x = plchan->host;
1428 	struct pl08x_txd *txd;
1429 	struct pl08x_sg *dsg;
1430 	int ret;
1431 
1432 	txd = pl08x_get_txd(plchan);
1433 	if (!txd) {
1434 		dev_err(&pl08x->adev->dev,
1435 			"%s no memory for descriptor\n", __func__);
1436 		return NULL;
1437 	}
1438 
1439 	dsg = kzalloc(sizeof(struct pl08x_sg), GFP_NOWAIT);
1440 	if (!dsg) {
1441 		pl08x_free_txd(pl08x, txd);
1442 		dev_err(&pl08x->adev->dev, "%s no memory for pl080 sg\n",
1443 				__func__);
1444 		return NULL;
1445 	}
1446 	list_add_tail(&dsg->node, &txd->dsg_list);
1447 
1448 	dsg->src_addr = src;
1449 	dsg->dst_addr = dest;
1450 	dsg->len = len;
1451 
1452 	/* Set platform data for m2m */
1453 	txd->ccfg |= PL080_FLOW_MEM2MEM << PL080_CONFIG_FLOW_CONTROL_SHIFT;
1454 	txd->cctl = pl08x->pd->memcpy_channel.cctl_memcpy &
1455 			~(PL080_CONTROL_DST_AHB2 | PL080_CONTROL_SRC_AHB2);
1456 
1457 	/* Both to be incremented or the code will break */
1458 	txd->cctl |= PL080_CONTROL_SRC_INCR | PL080_CONTROL_DST_INCR;
1459 
1460 	if (pl08x->vd->dualmaster)
1461 		txd->cctl |= pl08x_select_bus(pl08x->mem_buses,
1462 					      pl08x->mem_buses);
1463 
1464 	ret = pl08x_fill_llis_for_desc(plchan->host, txd);
1465 	if (!ret) {
1466 		pl08x_free_txd(pl08x, txd);
1467 		return NULL;
1468 	}
1469 
1470 	return vchan_tx_prep(&plchan->vc, &txd->vd, flags);
1471 }
1472 
1473 static struct pl08x_txd *pl08x_init_txd(
1474 		struct dma_chan *chan,
1475 		enum dma_transfer_direction direction,
1476 		dma_addr_t *slave_addr)
1477 {
1478 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1479 	struct pl08x_driver_data *pl08x = plchan->host;
1480 	struct pl08x_txd *txd;
1481 	enum dma_slave_buswidth addr_width;
1482 	int ret, tmp;
1483 	u8 src_buses, dst_buses;
1484 	u32 maxburst, cctl;
1485 
1486 	txd = pl08x_get_txd(plchan);
1487 	if (!txd) {
1488 		dev_err(&pl08x->adev->dev, "%s no txd\n", __func__);
1489 		return NULL;
1490 	}
1491 
1492 	/*
1493 	 * Set up addresses, the PrimeCell configured address
1494 	 * will take precedence since this may configure the
1495 	 * channel target address dynamically at runtime.
1496 	 */
1497 	if (direction == DMA_MEM_TO_DEV) {
1498 		cctl = PL080_CONTROL_SRC_INCR;
1499 		*slave_addr = plchan->cfg.dst_addr;
1500 		addr_width = plchan->cfg.dst_addr_width;
1501 		maxburst = plchan->cfg.dst_maxburst;
1502 		src_buses = pl08x->mem_buses;
1503 		dst_buses = plchan->cd->periph_buses;
1504 	} else if (direction == DMA_DEV_TO_MEM) {
1505 		cctl = PL080_CONTROL_DST_INCR;
1506 		*slave_addr = plchan->cfg.src_addr;
1507 		addr_width = plchan->cfg.src_addr_width;
1508 		maxburst = plchan->cfg.src_maxburst;
1509 		src_buses = plchan->cd->periph_buses;
1510 		dst_buses = pl08x->mem_buses;
1511 	} else {
1512 		pl08x_free_txd(pl08x, txd);
1513 		dev_err(&pl08x->adev->dev,
1514 			"%s direction unsupported\n", __func__);
1515 		return NULL;
1516 	}
1517 
1518 	cctl |= pl08x_get_cctl(plchan, addr_width, maxburst);
1519 	if (cctl == ~0) {
1520 		pl08x_free_txd(pl08x, txd);
1521 		dev_err(&pl08x->adev->dev,
1522 			"DMA slave configuration botched?\n");
1523 		return NULL;
1524 	}
1525 
1526 	txd->cctl = cctl | pl08x_select_bus(src_buses, dst_buses);
1527 
1528 	if (plchan->cfg.device_fc)
1529 		tmp = (direction == DMA_MEM_TO_DEV) ? PL080_FLOW_MEM2PER_PER :
1530 			PL080_FLOW_PER2MEM_PER;
1531 	else
1532 		tmp = (direction == DMA_MEM_TO_DEV) ? PL080_FLOW_MEM2PER :
1533 			PL080_FLOW_PER2MEM;
1534 
1535 	txd->ccfg |= tmp << PL080_CONFIG_FLOW_CONTROL_SHIFT;
1536 
1537 	ret = pl08x_request_mux(plchan);
1538 	if (ret < 0) {
1539 		pl08x_free_txd(pl08x, txd);
1540 		dev_dbg(&pl08x->adev->dev,
1541 			"unable to mux for transfer on %s due to platform restrictions\n",
1542 			plchan->name);
1543 		return NULL;
1544 	}
1545 
1546 	dev_dbg(&pl08x->adev->dev, "allocated DMA request signal %d for xfer on %s\n",
1547 		 plchan->signal, plchan->name);
1548 
1549 	/* Assign the flow control signal to this channel */
1550 	if (direction == DMA_MEM_TO_DEV)
1551 		txd->ccfg |= plchan->signal << PL080_CONFIG_DST_SEL_SHIFT;
1552 	else
1553 		txd->ccfg |= plchan->signal << PL080_CONFIG_SRC_SEL_SHIFT;
1554 
1555 	return txd;
1556 }
1557 
1558 static int pl08x_tx_add_sg(struct pl08x_txd *txd,
1559 			   enum dma_transfer_direction direction,
1560 			   dma_addr_t slave_addr,
1561 			   dma_addr_t buf_addr,
1562 			   unsigned int len)
1563 {
1564 	struct pl08x_sg *dsg;
1565 
1566 	dsg = kzalloc(sizeof(struct pl08x_sg), GFP_NOWAIT);
1567 	if (!dsg)
1568 		return -ENOMEM;
1569 
1570 	list_add_tail(&dsg->node, &txd->dsg_list);
1571 
1572 	dsg->len = len;
1573 	if (direction == DMA_MEM_TO_DEV) {
1574 		dsg->src_addr = buf_addr;
1575 		dsg->dst_addr = slave_addr;
1576 	} else {
1577 		dsg->src_addr = slave_addr;
1578 		dsg->dst_addr = buf_addr;
1579 	}
1580 
1581 	return 0;
1582 }
1583 
1584 static struct dma_async_tx_descriptor *pl08x_prep_slave_sg(
1585 		struct dma_chan *chan, struct scatterlist *sgl,
1586 		unsigned int sg_len, enum dma_transfer_direction direction,
1587 		unsigned long flags, void *context)
1588 {
1589 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1590 	struct pl08x_driver_data *pl08x = plchan->host;
1591 	struct pl08x_txd *txd;
1592 	struct scatterlist *sg;
1593 	int ret, tmp;
1594 	dma_addr_t slave_addr;
1595 
1596 	dev_dbg(&pl08x->adev->dev, "%s prepare transaction of %d bytes from %s\n",
1597 			__func__, sg_dma_len(sgl), plchan->name);
1598 
1599 	txd = pl08x_init_txd(chan, direction, &slave_addr);
1600 	if (!txd)
1601 		return NULL;
1602 
1603 	for_each_sg(sgl, sg, sg_len, tmp) {
1604 		ret = pl08x_tx_add_sg(txd, direction, slave_addr,
1605 				      sg_dma_address(sg),
1606 				      sg_dma_len(sg));
1607 		if (ret) {
1608 			pl08x_release_mux(plchan);
1609 			pl08x_free_txd(pl08x, txd);
1610 			dev_err(&pl08x->adev->dev, "%s no mem for pl080 sg\n",
1611 					__func__);
1612 			return NULL;
1613 		}
1614 	}
1615 
1616 	ret = pl08x_fill_llis_for_desc(plchan->host, txd);
1617 	if (!ret) {
1618 		pl08x_release_mux(plchan);
1619 		pl08x_free_txd(pl08x, txd);
1620 		return NULL;
1621 	}
1622 
1623 	return vchan_tx_prep(&plchan->vc, &txd->vd, flags);
1624 }
1625 
1626 static struct dma_async_tx_descriptor *pl08x_prep_dma_cyclic(
1627 		struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len,
1628 		size_t period_len, enum dma_transfer_direction direction,
1629 		unsigned long flags)
1630 {
1631 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1632 	struct pl08x_driver_data *pl08x = plchan->host;
1633 	struct pl08x_txd *txd;
1634 	int ret, tmp;
1635 	dma_addr_t slave_addr;
1636 
1637 	dev_dbg(&pl08x->adev->dev,
1638 		"%s prepare cyclic transaction of %zd/%zd bytes %s %s\n",
1639 		__func__, period_len, buf_len,
1640 		direction == DMA_MEM_TO_DEV ? "to" : "from",
1641 		plchan->name);
1642 
1643 	txd = pl08x_init_txd(chan, direction, &slave_addr);
1644 	if (!txd)
1645 		return NULL;
1646 
1647 	txd->cyclic = true;
1648 	txd->cctl |= PL080_CONTROL_TC_IRQ_EN;
1649 	for (tmp = 0; tmp < buf_len; tmp += period_len) {
1650 		ret = pl08x_tx_add_sg(txd, direction, slave_addr,
1651 				      buf_addr + tmp, period_len);
1652 		if (ret) {
1653 			pl08x_release_mux(plchan);
1654 			pl08x_free_txd(pl08x, txd);
1655 			return NULL;
1656 		}
1657 	}
1658 
1659 	ret = pl08x_fill_llis_for_desc(plchan->host, txd);
1660 	if (!ret) {
1661 		pl08x_release_mux(plchan);
1662 		pl08x_free_txd(pl08x, txd);
1663 		return NULL;
1664 	}
1665 
1666 	return vchan_tx_prep(&plchan->vc, &txd->vd, flags);
1667 }
1668 
1669 static int pl08x_config(struct dma_chan *chan,
1670 			struct dma_slave_config *config)
1671 {
1672 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1673 	struct pl08x_driver_data *pl08x = plchan->host;
1674 
1675 	if (!plchan->slave)
1676 		return -EINVAL;
1677 
1678 	/* Reject definitely invalid configurations */
1679 	if (config->src_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES ||
1680 	    config->dst_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES)
1681 		return -EINVAL;
1682 
1683 	if (config->device_fc && pl08x->vd->pl080s) {
1684 		dev_err(&pl08x->adev->dev,
1685 			"%s: PL080S does not support peripheral flow control\n",
1686 			__func__);
1687 		return -EINVAL;
1688 	}
1689 
1690 	plchan->cfg = *config;
1691 
1692 	return 0;
1693 }
1694 
1695 static int pl08x_terminate_all(struct dma_chan *chan)
1696 {
1697 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1698 	struct pl08x_driver_data *pl08x = plchan->host;
1699 	unsigned long flags;
1700 
1701 	spin_lock_irqsave(&plchan->vc.lock, flags);
1702 	if (!plchan->phychan && !plchan->at) {
1703 		spin_unlock_irqrestore(&plchan->vc.lock, flags);
1704 		return 0;
1705 	}
1706 
1707 	plchan->state = PL08X_CHAN_IDLE;
1708 
1709 	if (plchan->phychan) {
1710 		/*
1711 		 * Mark physical channel as free and free any slave
1712 		 * signal
1713 		 */
1714 		pl08x_phy_free(plchan);
1715 	}
1716 	/* Dequeue jobs and free LLIs */
1717 	if (plchan->at) {
1718 		pl08x_desc_free(&plchan->at->vd);
1719 		plchan->at = NULL;
1720 	}
1721 	/* Dequeue jobs not yet fired as well */
1722 	pl08x_free_txd_list(pl08x, plchan);
1723 
1724 	spin_unlock_irqrestore(&plchan->vc.lock, flags);
1725 
1726 	return 0;
1727 }
1728 
1729 static int pl08x_pause(struct dma_chan *chan)
1730 {
1731 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1732 	unsigned long flags;
1733 
1734 	/*
1735 	 * Anything succeeds on channels with no physical allocation and
1736 	 * no queued transfers.
1737 	 */
1738 	spin_lock_irqsave(&plchan->vc.lock, flags);
1739 	if (!plchan->phychan && !plchan->at) {
1740 		spin_unlock_irqrestore(&plchan->vc.lock, flags);
1741 		return 0;
1742 	}
1743 
1744 	pl08x_pause_phy_chan(plchan->phychan);
1745 	plchan->state = PL08X_CHAN_PAUSED;
1746 
1747 	spin_unlock_irqrestore(&plchan->vc.lock, flags);
1748 
1749 	return 0;
1750 }
1751 
1752 static int pl08x_resume(struct dma_chan *chan)
1753 {
1754 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1755 	unsigned long flags;
1756 
1757 	/*
1758 	 * Anything succeeds on channels with no physical allocation and
1759 	 * no queued transfers.
1760 	 */
1761 	spin_lock_irqsave(&plchan->vc.lock, flags);
1762 	if (!plchan->phychan && !plchan->at) {
1763 		spin_unlock_irqrestore(&plchan->vc.lock, flags);
1764 		return 0;
1765 	}
1766 
1767 	pl08x_resume_phy_chan(plchan->phychan);
1768 	plchan->state = PL08X_CHAN_RUNNING;
1769 
1770 	spin_unlock_irqrestore(&plchan->vc.lock, flags);
1771 
1772 	return 0;
1773 }
1774 
1775 bool pl08x_filter_id(struct dma_chan *chan, void *chan_id)
1776 {
1777 	struct pl08x_dma_chan *plchan;
1778 	char *name = chan_id;
1779 
1780 	/* Reject channels for devices not bound to this driver */
1781 	if (chan->device->dev->driver != &pl08x_amba_driver.drv)
1782 		return false;
1783 
1784 	plchan = to_pl08x_chan(chan);
1785 
1786 	/* Check that the channel is not taken! */
1787 	if (!strcmp(plchan->name, name))
1788 		return true;
1789 
1790 	return false;
1791 }
1792 EXPORT_SYMBOL_GPL(pl08x_filter_id);
1793 
1794 /*
1795  * Just check that the device is there and active
1796  * TODO: turn this bit on/off depending on the number of physical channels
1797  * actually used, if it is zero... well shut it off. That will save some
1798  * power. Cut the clock at the same time.
1799  */
1800 static void pl08x_ensure_on(struct pl08x_driver_data *pl08x)
1801 {
1802 	/* The Nomadik variant does not have the config register */
1803 	if (pl08x->vd->nomadik)
1804 		return;
1805 	writel(PL080_CONFIG_ENABLE, pl08x->base + PL080_CONFIG);
1806 }
1807 
1808 static irqreturn_t pl08x_irq(int irq, void *dev)
1809 {
1810 	struct pl08x_driver_data *pl08x = dev;
1811 	u32 mask = 0, err, tc, i;
1812 
1813 	/* check & clear - ERR & TC interrupts */
1814 	err = readl(pl08x->base + PL080_ERR_STATUS);
1815 	if (err) {
1816 		dev_err(&pl08x->adev->dev, "%s error interrupt, register value 0x%08x\n",
1817 			__func__, err);
1818 		writel(err, pl08x->base + PL080_ERR_CLEAR);
1819 	}
1820 	tc = readl(pl08x->base + PL080_TC_STATUS);
1821 	if (tc)
1822 		writel(tc, pl08x->base + PL080_TC_CLEAR);
1823 
1824 	if (!err && !tc)
1825 		return IRQ_NONE;
1826 
1827 	for (i = 0; i < pl08x->vd->channels; i++) {
1828 		if (((1 << i) & err) || ((1 << i) & tc)) {
1829 			/* Locate physical channel */
1830 			struct pl08x_phy_chan *phychan = &pl08x->phy_chans[i];
1831 			struct pl08x_dma_chan *plchan = phychan->serving;
1832 			struct pl08x_txd *tx;
1833 
1834 			if (!plchan) {
1835 				dev_err(&pl08x->adev->dev,
1836 					"%s Error TC interrupt on unused channel: 0x%08x\n",
1837 					__func__, i);
1838 				continue;
1839 			}
1840 
1841 			spin_lock(&plchan->vc.lock);
1842 			tx = plchan->at;
1843 			if (tx && tx->cyclic) {
1844 				vchan_cyclic_callback(&tx->vd);
1845 			} else if (tx) {
1846 				plchan->at = NULL;
1847 				/*
1848 				 * This descriptor is done, release its mux
1849 				 * reservation.
1850 				 */
1851 				pl08x_release_mux(plchan);
1852 				tx->done = true;
1853 				vchan_cookie_complete(&tx->vd);
1854 
1855 				/*
1856 				 * And start the next descriptor (if any),
1857 				 * otherwise free this channel.
1858 				 */
1859 				if (vchan_next_desc(&plchan->vc))
1860 					pl08x_start_next_txd(plchan);
1861 				else
1862 					pl08x_phy_free(plchan);
1863 			}
1864 			spin_unlock(&plchan->vc.lock);
1865 
1866 			mask |= (1 << i);
1867 		}
1868 	}
1869 
1870 	return mask ? IRQ_HANDLED : IRQ_NONE;
1871 }
1872 
1873 static void pl08x_dma_slave_init(struct pl08x_dma_chan *chan)
1874 {
1875 	chan->slave = true;
1876 	chan->name = chan->cd->bus_id;
1877 	chan->cfg.src_addr = chan->cd->addr;
1878 	chan->cfg.dst_addr = chan->cd->addr;
1879 }
1880 
1881 /*
1882  * Initialise the DMAC memcpy/slave channels.
1883  * Make a local wrapper to hold required data
1884  */
1885 static int pl08x_dma_init_virtual_channels(struct pl08x_driver_data *pl08x,
1886 		struct dma_device *dmadev, unsigned int channels, bool slave)
1887 {
1888 	struct pl08x_dma_chan *chan;
1889 	int i;
1890 
1891 	INIT_LIST_HEAD(&dmadev->channels);
1892 
1893 	/*
1894 	 * Register as many many memcpy as we have physical channels,
1895 	 * we won't always be able to use all but the code will have
1896 	 * to cope with that situation.
1897 	 */
1898 	for (i = 0; i < channels; i++) {
1899 		chan = kzalloc(sizeof(*chan), GFP_KERNEL);
1900 		if (!chan) {
1901 			dev_err(&pl08x->adev->dev,
1902 				"%s no memory for channel\n", __func__);
1903 			return -ENOMEM;
1904 		}
1905 
1906 		chan->host = pl08x;
1907 		chan->state = PL08X_CHAN_IDLE;
1908 		chan->signal = -1;
1909 
1910 		if (slave) {
1911 			chan->cd = &pl08x->pd->slave_channels[i];
1912 			pl08x_dma_slave_init(chan);
1913 		} else {
1914 			chan->cd = &pl08x->pd->memcpy_channel;
1915 			chan->name = kasprintf(GFP_KERNEL, "memcpy%d", i);
1916 			if (!chan->name) {
1917 				kfree(chan);
1918 				return -ENOMEM;
1919 			}
1920 		}
1921 		dev_dbg(&pl08x->adev->dev,
1922 			 "initialize virtual channel \"%s\"\n",
1923 			 chan->name);
1924 
1925 		chan->vc.desc_free = pl08x_desc_free;
1926 		vchan_init(&chan->vc, dmadev);
1927 	}
1928 	dev_info(&pl08x->adev->dev, "initialized %d virtual %s channels\n",
1929 		 i, slave ? "slave" : "memcpy");
1930 	return i;
1931 }
1932 
1933 static void pl08x_free_virtual_channels(struct dma_device *dmadev)
1934 {
1935 	struct pl08x_dma_chan *chan = NULL;
1936 	struct pl08x_dma_chan *next;
1937 
1938 	list_for_each_entry_safe(chan,
1939 				 next, &dmadev->channels, vc.chan.device_node) {
1940 		list_del(&chan->vc.chan.device_node);
1941 		kfree(chan);
1942 	}
1943 }
1944 
1945 #ifdef CONFIG_DEBUG_FS
1946 static const char *pl08x_state_str(enum pl08x_dma_chan_state state)
1947 {
1948 	switch (state) {
1949 	case PL08X_CHAN_IDLE:
1950 		return "idle";
1951 	case PL08X_CHAN_RUNNING:
1952 		return "running";
1953 	case PL08X_CHAN_PAUSED:
1954 		return "paused";
1955 	case PL08X_CHAN_WAITING:
1956 		return "waiting";
1957 	default:
1958 		break;
1959 	}
1960 	return "UNKNOWN STATE";
1961 }
1962 
1963 static int pl08x_debugfs_show(struct seq_file *s, void *data)
1964 {
1965 	struct pl08x_driver_data *pl08x = s->private;
1966 	struct pl08x_dma_chan *chan;
1967 	struct pl08x_phy_chan *ch;
1968 	unsigned long flags;
1969 	int i;
1970 
1971 	seq_printf(s, "PL08x physical channels:\n");
1972 	seq_printf(s, "CHANNEL:\tUSER:\n");
1973 	seq_printf(s, "--------\t-----\n");
1974 	for (i = 0; i < pl08x->vd->channels; i++) {
1975 		struct pl08x_dma_chan *virt_chan;
1976 
1977 		ch = &pl08x->phy_chans[i];
1978 
1979 		spin_lock_irqsave(&ch->lock, flags);
1980 		virt_chan = ch->serving;
1981 
1982 		seq_printf(s, "%d\t\t%s%s\n",
1983 			   ch->id,
1984 			   virt_chan ? virt_chan->name : "(none)",
1985 			   ch->locked ? " LOCKED" : "");
1986 
1987 		spin_unlock_irqrestore(&ch->lock, flags);
1988 	}
1989 
1990 	seq_printf(s, "\nPL08x virtual memcpy channels:\n");
1991 	seq_printf(s, "CHANNEL:\tSTATE:\n");
1992 	seq_printf(s, "--------\t------\n");
1993 	list_for_each_entry(chan, &pl08x->memcpy.channels, vc.chan.device_node) {
1994 		seq_printf(s, "%s\t\t%s\n", chan->name,
1995 			   pl08x_state_str(chan->state));
1996 	}
1997 
1998 	seq_printf(s, "\nPL08x virtual slave channels:\n");
1999 	seq_printf(s, "CHANNEL:\tSTATE:\n");
2000 	seq_printf(s, "--------\t------\n");
2001 	list_for_each_entry(chan, &pl08x->slave.channels, vc.chan.device_node) {
2002 		seq_printf(s, "%s\t\t%s\n", chan->name,
2003 			   pl08x_state_str(chan->state));
2004 	}
2005 
2006 	return 0;
2007 }
2008 
2009 static int pl08x_debugfs_open(struct inode *inode, struct file *file)
2010 {
2011 	return single_open(file, pl08x_debugfs_show, inode->i_private);
2012 }
2013 
2014 static const struct file_operations pl08x_debugfs_operations = {
2015 	.open		= pl08x_debugfs_open,
2016 	.read		= seq_read,
2017 	.llseek		= seq_lseek,
2018 	.release	= single_release,
2019 };
2020 
2021 static void init_pl08x_debugfs(struct pl08x_driver_data *pl08x)
2022 {
2023 	/* Expose a simple debugfs interface to view all clocks */
2024 	(void) debugfs_create_file(dev_name(&pl08x->adev->dev),
2025 			S_IFREG | S_IRUGO, NULL, pl08x,
2026 			&pl08x_debugfs_operations);
2027 }
2028 
2029 #else
2030 static inline void init_pl08x_debugfs(struct pl08x_driver_data *pl08x)
2031 {
2032 }
2033 #endif
2034 
2035 #ifdef CONFIG_OF
2036 static struct dma_chan *pl08x_find_chan_id(struct pl08x_driver_data *pl08x,
2037 					 u32 id)
2038 {
2039 	struct pl08x_dma_chan *chan;
2040 
2041 	list_for_each_entry(chan, &pl08x->slave.channels, vc.chan.device_node) {
2042 		if (chan->signal == id)
2043 			return &chan->vc.chan;
2044 	}
2045 
2046 	return NULL;
2047 }
2048 
2049 static struct dma_chan *pl08x_of_xlate(struct of_phandle_args *dma_spec,
2050 				       struct of_dma *ofdma)
2051 {
2052 	struct pl08x_driver_data *pl08x = ofdma->of_dma_data;
2053 	struct pl08x_channel_data *data;
2054 	struct pl08x_dma_chan *chan;
2055 	struct dma_chan *dma_chan;
2056 
2057 	if (!pl08x)
2058 		return NULL;
2059 
2060 	if (dma_spec->args_count != 2)
2061 		return NULL;
2062 
2063 	dma_chan = pl08x_find_chan_id(pl08x, dma_spec->args[0]);
2064 	if (dma_chan)
2065 		return dma_get_slave_channel(dma_chan);
2066 
2067 	chan = devm_kzalloc(pl08x->slave.dev, sizeof(*chan) + sizeof(*data),
2068 			    GFP_KERNEL);
2069 	if (!chan)
2070 		return NULL;
2071 
2072 	data = (void *)&chan[1];
2073 	data->bus_id = "(none)";
2074 	data->periph_buses = dma_spec->args[1];
2075 
2076 	chan->cd = data;
2077 	chan->host = pl08x;
2078 	chan->slave = true;
2079 	chan->name = data->bus_id;
2080 	chan->state = PL08X_CHAN_IDLE;
2081 	chan->signal = dma_spec->args[0];
2082 	chan->vc.desc_free = pl08x_desc_free;
2083 
2084 	vchan_init(&chan->vc, &pl08x->slave);
2085 
2086 	return dma_get_slave_channel(&chan->vc.chan);
2087 }
2088 
2089 static int pl08x_of_probe(struct amba_device *adev,
2090 			  struct pl08x_driver_data *pl08x,
2091 			  struct device_node *np)
2092 {
2093 	struct pl08x_platform_data *pd;
2094 	u32 cctl_memcpy = 0;
2095 	u32 val;
2096 	int ret;
2097 
2098 	pd = devm_kzalloc(&adev->dev, sizeof(*pd), GFP_KERNEL);
2099 	if (!pd)
2100 		return -ENOMEM;
2101 
2102 	/* Eligible bus masters for fetching LLIs */
2103 	if (of_property_read_bool(np, "lli-bus-interface-ahb1"))
2104 		pd->lli_buses |= PL08X_AHB1;
2105 	if (of_property_read_bool(np, "lli-bus-interface-ahb2"))
2106 		pd->lli_buses |= PL08X_AHB2;
2107 	if (!pd->lli_buses) {
2108 		dev_info(&adev->dev, "no bus masters for LLIs stated, assume all\n");
2109 		pd->lli_buses |= PL08X_AHB1 | PL08X_AHB2;
2110 	}
2111 
2112 	/* Eligible bus masters for memory access */
2113 	if (of_property_read_bool(np, "mem-bus-interface-ahb1"))
2114 		pd->mem_buses |= PL08X_AHB1;
2115 	if (of_property_read_bool(np, "mem-bus-interface-ahb2"))
2116 		pd->mem_buses |= PL08X_AHB2;
2117 	if (!pd->mem_buses) {
2118 		dev_info(&adev->dev, "no bus masters for memory stated, assume all\n");
2119 		pd->mem_buses |= PL08X_AHB1 | PL08X_AHB2;
2120 	}
2121 
2122 	/* Parse the memcpy channel properties */
2123 	ret = of_property_read_u32(np, "memcpy-burst-size", &val);
2124 	if (ret) {
2125 		dev_info(&adev->dev, "no memcpy burst size specified, using 1 byte\n");
2126 		val = 1;
2127 	}
2128 	switch (val) {
2129 	default:
2130 		dev_err(&adev->dev, "illegal burst size for memcpy, set to 1\n");
2131 		/* Fall through */
2132 	case 1:
2133 		cctl_memcpy |= PL080_BSIZE_1 << PL080_CONTROL_SB_SIZE_SHIFT |
2134 			       PL080_BSIZE_1 << PL080_CONTROL_DB_SIZE_SHIFT;
2135 		break;
2136 	case 4:
2137 		cctl_memcpy |= PL080_BSIZE_4 << PL080_CONTROL_SB_SIZE_SHIFT |
2138 			       PL080_BSIZE_4 << PL080_CONTROL_DB_SIZE_SHIFT;
2139 		break;
2140 	case 8:
2141 		cctl_memcpy |= PL080_BSIZE_8 << PL080_CONTROL_SB_SIZE_SHIFT |
2142 			       PL080_BSIZE_8 << PL080_CONTROL_DB_SIZE_SHIFT;
2143 		break;
2144 	case 16:
2145 		cctl_memcpy |= PL080_BSIZE_16 << PL080_CONTROL_SB_SIZE_SHIFT |
2146 			       PL080_BSIZE_16 << PL080_CONTROL_DB_SIZE_SHIFT;
2147 		break;
2148 	case 32:
2149 		cctl_memcpy |= PL080_BSIZE_32 << PL080_CONTROL_SB_SIZE_SHIFT |
2150 			       PL080_BSIZE_32 << PL080_CONTROL_DB_SIZE_SHIFT;
2151 		break;
2152 	case 64:
2153 		cctl_memcpy |= PL080_BSIZE_64 << PL080_CONTROL_SB_SIZE_SHIFT |
2154 			       PL080_BSIZE_64 << PL080_CONTROL_DB_SIZE_SHIFT;
2155 		break;
2156 	case 128:
2157 		cctl_memcpy |= PL080_BSIZE_128 << PL080_CONTROL_SB_SIZE_SHIFT |
2158 			       PL080_BSIZE_128 << PL080_CONTROL_DB_SIZE_SHIFT;
2159 		break;
2160 	case 256:
2161 		cctl_memcpy |= PL080_BSIZE_256 << PL080_CONTROL_SB_SIZE_SHIFT |
2162 			       PL080_BSIZE_256 << PL080_CONTROL_DB_SIZE_SHIFT;
2163 		break;
2164 	}
2165 
2166 	ret = of_property_read_u32(np, "memcpy-bus-width", &val);
2167 	if (ret) {
2168 		dev_info(&adev->dev, "no memcpy bus width specified, using 8 bits\n");
2169 		val = 8;
2170 	}
2171 	switch (val) {
2172 	default:
2173 		dev_err(&adev->dev, "illegal bus width for memcpy, set to 8 bits\n");
2174 		/* Fall through */
2175 	case 8:
2176 		cctl_memcpy |= PL080_WIDTH_8BIT << PL080_CONTROL_SWIDTH_SHIFT |
2177 			       PL080_WIDTH_8BIT << PL080_CONTROL_DWIDTH_SHIFT;
2178 		break;
2179 	case 16:
2180 		cctl_memcpy |= PL080_WIDTH_16BIT << PL080_CONTROL_SWIDTH_SHIFT |
2181 			       PL080_WIDTH_16BIT << PL080_CONTROL_DWIDTH_SHIFT;
2182 		break;
2183 	case 32:
2184 		cctl_memcpy |= PL080_WIDTH_32BIT << PL080_CONTROL_SWIDTH_SHIFT |
2185 			       PL080_WIDTH_32BIT << PL080_CONTROL_DWIDTH_SHIFT;
2186 		break;
2187 	}
2188 
2189 	/* This is currently the only thing making sense */
2190 	cctl_memcpy |= PL080_CONTROL_PROT_SYS;
2191 
2192 	/* Set up memcpy channel */
2193 	pd->memcpy_channel.bus_id = "memcpy";
2194 	pd->memcpy_channel.cctl_memcpy = cctl_memcpy;
2195 	/* Use the buses that can access memory, obviously */
2196 	pd->memcpy_channel.periph_buses = pd->mem_buses;
2197 
2198 	pl08x->pd = pd;
2199 
2200 	return of_dma_controller_register(adev->dev.of_node, pl08x_of_xlate,
2201 					  pl08x);
2202 }
2203 #else
2204 static inline int pl08x_of_probe(struct amba_device *adev,
2205 				 struct pl08x_driver_data *pl08x,
2206 				 struct device_node *np)
2207 {
2208 	return -EINVAL;
2209 }
2210 #endif
2211 
2212 static int pl08x_probe(struct amba_device *adev, const struct amba_id *id)
2213 {
2214 	struct pl08x_driver_data *pl08x;
2215 	const struct vendor_data *vd = id->data;
2216 	struct device_node *np = adev->dev.of_node;
2217 	u32 tsfr_size;
2218 	int ret = 0;
2219 	int i;
2220 
2221 	ret = amba_request_regions(adev, NULL);
2222 	if (ret)
2223 		return ret;
2224 
2225 	/* Ensure that we can do DMA */
2226 	ret = dma_set_mask_and_coherent(&adev->dev, DMA_BIT_MASK(32));
2227 	if (ret)
2228 		goto out_no_pl08x;
2229 
2230 	/* Create the driver state holder */
2231 	pl08x = kzalloc(sizeof(*pl08x), GFP_KERNEL);
2232 	if (!pl08x) {
2233 		ret = -ENOMEM;
2234 		goto out_no_pl08x;
2235 	}
2236 
2237 	/* Initialize memcpy engine */
2238 	dma_cap_set(DMA_MEMCPY, pl08x->memcpy.cap_mask);
2239 	pl08x->memcpy.dev = &adev->dev;
2240 	pl08x->memcpy.device_free_chan_resources = pl08x_free_chan_resources;
2241 	pl08x->memcpy.device_prep_dma_memcpy = pl08x_prep_dma_memcpy;
2242 	pl08x->memcpy.device_prep_dma_interrupt = pl08x_prep_dma_interrupt;
2243 	pl08x->memcpy.device_tx_status = pl08x_dma_tx_status;
2244 	pl08x->memcpy.device_issue_pending = pl08x_issue_pending;
2245 	pl08x->memcpy.device_config = pl08x_config;
2246 	pl08x->memcpy.device_pause = pl08x_pause;
2247 	pl08x->memcpy.device_resume = pl08x_resume;
2248 	pl08x->memcpy.device_terminate_all = pl08x_terminate_all;
2249 	pl08x->memcpy.src_addr_widths = PL80X_DMA_BUSWIDTHS;
2250 	pl08x->memcpy.dst_addr_widths = PL80X_DMA_BUSWIDTHS;
2251 	pl08x->memcpy.directions = BIT(DMA_MEM_TO_MEM);
2252 	pl08x->memcpy.residue_granularity = DMA_RESIDUE_GRANULARITY_SEGMENT;
2253 
2254 	/* Initialize slave engine */
2255 	dma_cap_set(DMA_SLAVE, pl08x->slave.cap_mask);
2256 	dma_cap_set(DMA_CYCLIC, pl08x->slave.cap_mask);
2257 	pl08x->slave.dev = &adev->dev;
2258 	pl08x->slave.device_free_chan_resources = pl08x_free_chan_resources;
2259 	pl08x->slave.device_prep_dma_interrupt = pl08x_prep_dma_interrupt;
2260 	pl08x->slave.device_tx_status = pl08x_dma_tx_status;
2261 	pl08x->slave.device_issue_pending = pl08x_issue_pending;
2262 	pl08x->slave.device_prep_slave_sg = pl08x_prep_slave_sg;
2263 	pl08x->slave.device_prep_dma_cyclic = pl08x_prep_dma_cyclic;
2264 	pl08x->slave.device_config = pl08x_config;
2265 	pl08x->slave.device_pause = pl08x_pause;
2266 	pl08x->slave.device_resume = pl08x_resume;
2267 	pl08x->slave.device_terminate_all = pl08x_terminate_all;
2268 	pl08x->slave.src_addr_widths = PL80X_DMA_BUSWIDTHS;
2269 	pl08x->slave.dst_addr_widths = PL80X_DMA_BUSWIDTHS;
2270 	pl08x->slave.directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
2271 	pl08x->slave.residue_granularity = DMA_RESIDUE_GRANULARITY_SEGMENT;
2272 
2273 	/* Get the platform data */
2274 	pl08x->pd = dev_get_platdata(&adev->dev);
2275 	if (!pl08x->pd) {
2276 		if (np) {
2277 			ret = pl08x_of_probe(adev, pl08x, np);
2278 			if (ret)
2279 				goto out_no_platdata;
2280 		} else {
2281 			dev_err(&adev->dev, "no platform data supplied\n");
2282 			ret = -EINVAL;
2283 			goto out_no_platdata;
2284 		}
2285 	}
2286 
2287 	/* Assign useful pointers to the driver state */
2288 	pl08x->adev = adev;
2289 	pl08x->vd = vd;
2290 
2291 	/* By default, AHB1 only.  If dualmaster, from platform */
2292 	pl08x->lli_buses = PL08X_AHB1;
2293 	pl08x->mem_buses = PL08X_AHB1;
2294 	if (pl08x->vd->dualmaster) {
2295 		pl08x->lli_buses = pl08x->pd->lli_buses;
2296 		pl08x->mem_buses = pl08x->pd->mem_buses;
2297 	}
2298 
2299 	if (vd->pl080s)
2300 		pl08x->lli_words = PL080S_LLI_WORDS;
2301 	else
2302 		pl08x->lli_words = PL080_LLI_WORDS;
2303 	tsfr_size = MAX_NUM_TSFR_LLIS * pl08x->lli_words * sizeof(u32);
2304 
2305 	/* A DMA memory pool for LLIs, align on 1-byte boundary */
2306 	pl08x->pool = dma_pool_create(DRIVER_NAME, &pl08x->adev->dev,
2307 						tsfr_size, PL08X_ALIGN, 0);
2308 	if (!pl08x->pool) {
2309 		ret = -ENOMEM;
2310 		goto out_no_lli_pool;
2311 	}
2312 
2313 	pl08x->base = ioremap(adev->res.start, resource_size(&adev->res));
2314 	if (!pl08x->base) {
2315 		ret = -ENOMEM;
2316 		goto out_no_ioremap;
2317 	}
2318 
2319 	/* Turn on the PL08x */
2320 	pl08x_ensure_on(pl08x);
2321 
2322 	/* Attach the interrupt handler */
2323 	writel(0x000000FF, pl08x->base + PL080_ERR_CLEAR);
2324 	writel(0x000000FF, pl08x->base + PL080_TC_CLEAR);
2325 
2326 	ret = request_irq(adev->irq[0], pl08x_irq, 0, DRIVER_NAME, pl08x);
2327 	if (ret) {
2328 		dev_err(&adev->dev, "%s failed to request interrupt %d\n",
2329 			__func__, adev->irq[0]);
2330 		goto out_no_irq;
2331 	}
2332 
2333 	/* Initialize physical channels */
2334 	pl08x->phy_chans = kzalloc((vd->channels * sizeof(*pl08x->phy_chans)),
2335 			GFP_KERNEL);
2336 	if (!pl08x->phy_chans) {
2337 		dev_err(&adev->dev, "%s failed to allocate "
2338 			"physical channel holders\n",
2339 			__func__);
2340 		ret = -ENOMEM;
2341 		goto out_no_phychans;
2342 	}
2343 
2344 	for (i = 0; i < vd->channels; i++) {
2345 		struct pl08x_phy_chan *ch = &pl08x->phy_chans[i];
2346 
2347 		ch->id = i;
2348 		ch->base = pl08x->base + PL080_Cx_BASE(i);
2349 		ch->reg_config = ch->base + vd->config_offset;
2350 		spin_lock_init(&ch->lock);
2351 
2352 		/*
2353 		 * Nomadik variants can have channels that are locked
2354 		 * down for the secure world only. Lock up these channels
2355 		 * by perpetually serving a dummy virtual channel.
2356 		 */
2357 		if (vd->nomadik) {
2358 			u32 val;
2359 
2360 			val = readl(ch->reg_config);
2361 			if (val & (PL080N_CONFIG_ITPROT | PL080N_CONFIG_SECPROT)) {
2362 				dev_info(&adev->dev, "physical channel %d reserved for secure access only\n", i);
2363 				ch->locked = true;
2364 			}
2365 		}
2366 
2367 		dev_dbg(&adev->dev, "physical channel %d is %s\n",
2368 			i, pl08x_phy_channel_busy(ch) ? "BUSY" : "FREE");
2369 	}
2370 
2371 	/* Register as many memcpy channels as there are physical channels */
2372 	ret = pl08x_dma_init_virtual_channels(pl08x, &pl08x->memcpy,
2373 					      pl08x->vd->channels, false);
2374 	if (ret <= 0) {
2375 		dev_warn(&pl08x->adev->dev,
2376 			 "%s failed to enumerate memcpy channels - %d\n",
2377 			 __func__, ret);
2378 		goto out_no_memcpy;
2379 	}
2380 
2381 	/* Register slave channels */
2382 	ret = pl08x_dma_init_virtual_channels(pl08x, &pl08x->slave,
2383 			pl08x->pd->num_slave_channels, true);
2384 	if (ret < 0) {
2385 		dev_warn(&pl08x->adev->dev,
2386 			"%s failed to enumerate slave channels - %d\n",
2387 				__func__, ret);
2388 		goto out_no_slave;
2389 	}
2390 
2391 	ret = dma_async_device_register(&pl08x->memcpy);
2392 	if (ret) {
2393 		dev_warn(&pl08x->adev->dev,
2394 			"%s failed to register memcpy as an async device - %d\n",
2395 			__func__, ret);
2396 		goto out_no_memcpy_reg;
2397 	}
2398 
2399 	ret = dma_async_device_register(&pl08x->slave);
2400 	if (ret) {
2401 		dev_warn(&pl08x->adev->dev,
2402 			"%s failed to register slave as an async device - %d\n",
2403 			__func__, ret);
2404 		goto out_no_slave_reg;
2405 	}
2406 
2407 	amba_set_drvdata(adev, pl08x);
2408 	init_pl08x_debugfs(pl08x);
2409 	dev_info(&pl08x->adev->dev, "DMA: PL%03x%s rev%u at 0x%08llx irq %d\n",
2410 		 amba_part(adev), pl08x->vd->pl080s ? "s" : "", amba_rev(adev),
2411 		 (unsigned long long)adev->res.start, adev->irq[0]);
2412 
2413 	return 0;
2414 
2415 out_no_slave_reg:
2416 	dma_async_device_unregister(&pl08x->memcpy);
2417 out_no_memcpy_reg:
2418 	pl08x_free_virtual_channels(&pl08x->slave);
2419 out_no_slave:
2420 	pl08x_free_virtual_channels(&pl08x->memcpy);
2421 out_no_memcpy:
2422 	kfree(pl08x->phy_chans);
2423 out_no_phychans:
2424 	free_irq(adev->irq[0], pl08x);
2425 out_no_irq:
2426 	iounmap(pl08x->base);
2427 out_no_ioremap:
2428 	dma_pool_destroy(pl08x->pool);
2429 out_no_lli_pool:
2430 out_no_platdata:
2431 	kfree(pl08x);
2432 out_no_pl08x:
2433 	amba_release_regions(adev);
2434 	return ret;
2435 }
2436 
2437 /* PL080 has 8 channels and the PL080 have just 2 */
2438 static struct vendor_data vendor_pl080 = {
2439 	.config_offset = PL080_CH_CONFIG,
2440 	.channels = 8,
2441 	.dualmaster = true,
2442 	.max_transfer_size = PL080_CONTROL_TRANSFER_SIZE_MASK,
2443 };
2444 
2445 static struct vendor_data vendor_nomadik = {
2446 	.config_offset = PL080_CH_CONFIG,
2447 	.channels = 8,
2448 	.dualmaster = true,
2449 	.nomadik = true,
2450 	.max_transfer_size = PL080_CONTROL_TRANSFER_SIZE_MASK,
2451 };
2452 
2453 static struct vendor_data vendor_pl080s = {
2454 	.config_offset = PL080S_CH_CONFIG,
2455 	.channels = 8,
2456 	.pl080s = true,
2457 	.max_transfer_size = PL080S_CONTROL_TRANSFER_SIZE_MASK,
2458 };
2459 
2460 static struct vendor_data vendor_pl081 = {
2461 	.config_offset = PL080_CH_CONFIG,
2462 	.channels = 2,
2463 	.dualmaster = false,
2464 	.max_transfer_size = PL080_CONTROL_TRANSFER_SIZE_MASK,
2465 };
2466 
2467 static struct amba_id pl08x_ids[] = {
2468 	/* Samsung PL080S variant */
2469 	{
2470 		.id	= 0x0a141080,
2471 		.mask	= 0xffffffff,
2472 		.data	= &vendor_pl080s,
2473 	},
2474 	/* PL080 */
2475 	{
2476 		.id	= 0x00041080,
2477 		.mask	= 0x000fffff,
2478 		.data	= &vendor_pl080,
2479 	},
2480 	/* PL081 */
2481 	{
2482 		.id	= 0x00041081,
2483 		.mask	= 0x000fffff,
2484 		.data	= &vendor_pl081,
2485 	},
2486 	/* Nomadik 8815 PL080 variant */
2487 	{
2488 		.id	= 0x00280080,
2489 		.mask	= 0x00ffffff,
2490 		.data	= &vendor_nomadik,
2491 	},
2492 	{ 0, 0 },
2493 };
2494 
2495 MODULE_DEVICE_TABLE(amba, pl08x_ids);
2496 
2497 static struct amba_driver pl08x_amba_driver = {
2498 	.drv.name	= DRIVER_NAME,
2499 	.id_table	= pl08x_ids,
2500 	.probe		= pl08x_probe,
2501 };
2502 
2503 static int __init pl08x_init(void)
2504 {
2505 	int retval;
2506 	retval = amba_driver_register(&pl08x_amba_driver);
2507 	if (retval)
2508 		printk(KERN_WARNING DRIVER_NAME
2509 		       "failed to register as an AMBA device (%d)\n",
2510 		       retval);
2511 	return retval;
2512 }
2513 subsys_initcall(pl08x_init);
2514