xref: /openbmc/linux/drivers/dma/amba-pl08x.c (revision f220d3eb)
1 /*
2  * Copyright (c) 2006 ARM Ltd.
3  * Copyright (c) 2010 ST-Ericsson SA
4  * Copyirght (c) 2017 Linaro Ltd.
5  *
6  * Author: Peter Pearse <peter.pearse@arm.com>
7  * Author: Linus Walleij <linus.walleij@linaro.org>
8  *
9  * This program is free software; you can redistribute it and/or modify it
10  * under the terms of the GNU General Public License as published by the Free
11  * Software Foundation; either version 2 of the License, or (at your option)
12  * any later version.
13  *
14  * This program is distributed in the hope that it will be useful, but WITHOUT
15  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
16  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
17  * more details.
18  *
19  * The full GNU General Public License is in this distribution in the file
20  * called COPYING.
21  *
22  * Documentation: ARM DDI 0196G == PL080
23  * Documentation: ARM DDI 0218E == PL081
24  * Documentation: S3C6410 User's Manual == PL080S
25  *
26  * PL080 & PL081 both have 16 sets of DMA signals that can be routed to any
27  * channel.
28  *
29  * The PL080 has 8 channels available for simultaneous use, and the PL081
30  * has only two channels. So on these DMA controllers the number of channels
31  * and the number of incoming DMA signals are two totally different things.
32  * It is usually not possible to theoretically handle all physical signals,
33  * so a multiplexing scheme with possible denial of use is necessary.
34  *
35  * The PL080 has a dual bus master, PL081 has a single master.
36  *
37  * PL080S is a version modified by Samsung and used in S3C64xx SoCs.
38  * It differs in following aspects:
39  * - CH_CONFIG register at different offset,
40  * - separate CH_CONTROL2 register for transfer size,
41  * - bigger maximum transfer size,
42  * - 8-word aligned LLI, instead of 4-word, due to extra CCTL2 word,
43  * - no support for peripheral flow control.
44  *
45  * Memory to peripheral transfer may be visualized as
46  *	Get data from memory to DMAC
47  *	Until no data left
48  *		On burst request from peripheral
49  *			Destination burst from DMAC to peripheral
50  *			Clear burst request
51  *	Raise terminal count interrupt
52  *
53  * For peripherals with a FIFO:
54  * Source      burst size == half the depth of the peripheral FIFO
55  * Destination burst size == the depth of the peripheral FIFO
56  *
57  * (Bursts are irrelevant for mem to mem transfers - there are no burst
58  * signals, the DMA controller will simply facilitate its AHB master.)
59  *
60  * ASSUMES default (little) endianness for DMA transfers
61  *
62  * The PL08x has two flow control settings:
63  *  - DMAC flow control: the transfer size defines the number of transfers
64  *    which occur for the current LLI entry, and the DMAC raises TC at the
65  *    end of every LLI entry.  Observed behaviour shows the DMAC listening
66  *    to both the BREQ and SREQ signals (contrary to documented),
67  *    transferring data if either is active.  The LBREQ and LSREQ signals
68  *    are ignored.
69  *
70  *  - Peripheral flow control: the transfer size is ignored (and should be
71  *    zero).  The data is transferred from the current LLI entry, until
72  *    after the final transfer signalled by LBREQ or LSREQ.  The DMAC
73  *    will then move to the next LLI entry. Unsupported by PL080S.
74  */
75 #include <linux/amba/bus.h>
76 #include <linux/amba/pl08x.h>
77 #include <linux/debugfs.h>
78 #include <linux/delay.h>
79 #include <linux/device.h>
80 #include <linux/dmaengine.h>
81 #include <linux/dmapool.h>
82 #include <linux/dma-mapping.h>
83 #include <linux/export.h>
84 #include <linux/init.h>
85 #include <linux/interrupt.h>
86 #include <linux/module.h>
87 #include <linux/of.h>
88 #include <linux/of_dma.h>
89 #include <linux/pm_runtime.h>
90 #include <linux/seq_file.h>
91 #include <linux/slab.h>
92 #include <linux/amba/pl080.h>
93 
94 #include "dmaengine.h"
95 #include "virt-dma.h"
96 
97 #define DRIVER_NAME	"pl08xdmac"
98 
99 #define PL80X_DMA_BUSWIDTHS \
100 	BIT(DMA_SLAVE_BUSWIDTH_UNDEFINED) | \
101 	BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
102 	BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
103 	BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
104 
105 static struct amba_driver pl08x_amba_driver;
106 struct pl08x_driver_data;
107 
108 /**
109  * struct vendor_data - vendor-specific config parameters for PL08x derivatives
110  * @config_offset: offset to the configuration register
111  * @channels: the number of channels available in this variant
112  * @signals: the number of request signals available from the hardware
113  * @dualmaster: whether this version supports dual AHB masters or not.
114  * @nomadik: whether this variant is a ST Microelectronics Nomadik, where the
115  *	channels have Nomadik security extension bits that need to be checked
116  *	for permission before use and some registers are missing
117  * @pl080s: whether this variant is a Samsung PL080S, which has separate
118  *	register and LLI word for transfer size.
119  * @ftdmac020: whether this variant is a Faraday Technology FTDMAC020
120  * @max_transfer_size: the maximum single element transfer size for this
121  *	PL08x variant.
122  */
123 struct vendor_data {
124 	u8 config_offset;
125 	u8 channels;
126 	u8 signals;
127 	bool dualmaster;
128 	bool nomadik;
129 	bool pl080s;
130 	bool ftdmac020;
131 	u32 max_transfer_size;
132 };
133 
134 /**
135  * struct pl08x_bus_data - information of source or destination
136  * busses for a transfer
137  * @addr: current address
138  * @maxwidth: the maximum width of a transfer on this bus
139  * @buswidth: the width of this bus in bytes: 1, 2 or 4
140  */
141 struct pl08x_bus_data {
142 	dma_addr_t addr;
143 	u8 maxwidth;
144 	u8 buswidth;
145 };
146 
147 #define IS_BUS_ALIGNED(bus) IS_ALIGNED((bus)->addr, (bus)->buswidth)
148 
149 /**
150  * struct pl08x_phy_chan - holder for the physical channels
151  * @id: physical index to this channel
152  * @base: memory base address for this physical channel
153  * @reg_config: configuration address for this physical channel
154  * @reg_control: control address for this physical channel
155  * @reg_src: transfer source address register
156  * @reg_dst: transfer destination address register
157  * @reg_lli: transfer LLI address register
158  * @reg_busy: if the variant has a special per-channel busy register,
159  * this contains a pointer to it
160  * @lock: a lock to use when altering an instance of this struct
161  * @serving: the virtual channel currently being served by this physical
162  * channel
163  * @locked: channel unavailable for the system, e.g. dedicated to secure
164  * world
165  * @ftdmac020: channel is on a FTDMAC020
166  * @pl080s: channel is on a PL08s
167  */
168 struct pl08x_phy_chan {
169 	unsigned int id;
170 	void __iomem *base;
171 	void __iomem *reg_config;
172 	void __iomem *reg_control;
173 	void __iomem *reg_src;
174 	void __iomem *reg_dst;
175 	void __iomem *reg_lli;
176 	void __iomem *reg_busy;
177 	spinlock_t lock;
178 	struct pl08x_dma_chan *serving;
179 	bool locked;
180 	bool ftdmac020;
181 	bool pl080s;
182 };
183 
184 /**
185  * struct pl08x_sg - structure containing data per sg
186  * @src_addr: src address of sg
187  * @dst_addr: dst address of sg
188  * @len: transfer len in bytes
189  * @node: node for txd's dsg_list
190  */
191 struct pl08x_sg {
192 	dma_addr_t src_addr;
193 	dma_addr_t dst_addr;
194 	size_t len;
195 	struct list_head node;
196 };
197 
198 /**
199  * struct pl08x_txd - wrapper for struct dma_async_tx_descriptor
200  * @vd: virtual DMA descriptor
201  * @dsg_list: list of children sg's
202  * @llis_bus: DMA memory address (physical) start for the LLIs
203  * @llis_va: virtual memory address start for the LLIs
204  * @cctl: control reg values for current txd
205  * @ccfg: config reg values for current txd
206  * @done: this marks completed descriptors, which should not have their
207  *   mux released.
208  * @cyclic: indicate cyclic transfers
209  */
210 struct pl08x_txd {
211 	struct virt_dma_desc vd;
212 	struct list_head dsg_list;
213 	dma_addr_t llis_bus;
214 	u32 *llis_va;
215 	/* Default cctl value for LLIs */
216 	u32 cctl;
217 	/*
218 	 * Settings to be put into the physical channel when we
219 	 * trigger this txd.  Other registers are in llis_va[0].
220 	 */
221 	u32 ccfg;
222 	bool done;
223 	bool cyclic;
224 };
225 
226 /**
227  * enum pl08x_dma_chan_state - holds the PL08x specific virtual channel
228  * states
229  * @PL08X_CHAN_IDLE: the channel is idle
230  * @PL08X_CHAN_RUNNING: the channel has allocated a physical transport
231  * channel and is running a transfer on it
232  * @PL08X_CHAN_PAUSED: the channel has allocated a physical transport
233  * channel, but the transfer is currently paused
234  * @PL08X_CHAN_WAITING: the channel is waiting for a physical transport
235  * channel to become available (only pertains to memcpy channels)
236  */
237 enum pl08x_dma_chan_state {
238 	PL08X_CHAN_IDLE,
239 	PL08X_CHAN_RUNNING,
240 	PL08X_CHAN_PAUSED,
241 	PL08X_CHAN_WAITING,
242 };
243 
244 /**
245  * struct pl08x_dma_chan - this structure wraps a DMA ENGINE channel
246  * @vc: wrappped virtual channel
247  * @phychan: the physical channel utilized by this channel, if there is one
248  * @name: name of channel
249  * @cd: channel platform data
250  * @cfg: slave configuration
251  * @at: active transaction on this channel
252  * @host: a pointer to the host (internal use)
253  * @state: whether the channel is idle, paused, running etc
254  * @slave: whether this channel is a device (slave) or for memcpy
255  * @signal: the physical DMA request signal which this channel is using
256  * @mux_use: count of descriptors using this DMA request signal setting
257  */
258 struct pl08x_dma_chan {
259 	struct virt_dma_chan vc;
260 	struct pl08x_phy_chan *phychan;
261 	const char *name;
262 	struct pl08x_channel_data *cd;
263 	struct dma_slave_config cfg;
264 	struct pl08x_txd *at;
265 	struct pl08x_driver_data *host;
266 	enum pl08x_dma_chan_state state;
267 	bool slave;
268 	int signal;
269 	unsigned mux_use;
270 };
271 
272 /**
273  * struct pl08x_driver_data - the local state holder for the PL08x
274  * @slave: optional slave engine for this instance
275  * @memcpy: memcpy engine for this instance
276  * @has_slave: the PL08x has a slave engine (routed signals)
277  * @base: virtual memory base (remapped) for the PL08x
278  * @adev: the corresponding AMBA (PrimeCell) bus entry
279  * @vd: vendor data for this PL08x variant
280  * @pd: platform data passed in from the platform/machine
281  * @phy_chans: array of data for the physical channels
282  * @pool: a pool for the LLI descriptors
283  * @lli_buses: bitmask to or in to LLI pointer selecting AHB port for LLI
284  * fetches
285  * @mem_buses: set to indicate memory transfers on AHB2.
286  * @lli_words: how many words are used in each LLI item for this variant
287  */
288 struct pl08x_driver_data {
289 	struct dma_device slave;
290 	struct dma_device memcpy;
291 	bool has_slave;
292 	void __iomem *base;
293 	struct amba_device *adev;
294 	const struct vendor_data *vd;
295 	struct pl08x_platform_data *pd;
296 	struct pl08x_phy_chan *phy_chans;
297 	struct dma_pool *pool;
298 	u8 lli_buses;
299 	u8 mem_buses;
300 	u8 lli_words;
301 };
302 
303 /*
304  * PL08X specific defines
305  */
306 
307 /* The order of words in an LLI. */
308 #define PL080_LLI_SRC		0
309 #define PL080_LLI_DST		1
310 #define PL080_LLI_LLI		2
311 #define PL080_LLI_CCTL		3
312 #define PL080S_LLI_CCTL2	4
313 
314 /* Total words in an LLI. */
315 #define PL080_LLI_WORDS		4
316 #define PL080S_LLI_WORDS	8
317 
318 /*
319  * Number of LLIs in each LLI buffer allocated for one transfer
320  * (maximum times we call dma_pool_alloc on this pool without freeing)
321  */
322 #define MAX_NUM_TSFR_LLIS	512
323 #define PL08X_ALIGN		8
324 
325 static inline struct pl08x_dma_chan *to_pl08x_chan(struct dma_chan *chan)
326 {
327 	return container_of(chan, struct pl08x_dma_chan, vc.chan);
328 }
329 
330 static inline struct pl08x_txd *to_pl08x_txd(struct dma_async_tx_descriptor *tx)
331 {
332 	return container_of(tx, struct pl08x_txd, vd.tx);
333 }
334 
335 /*
336  * Mux handling.
337  *
338  * This gives us the DMA request input to the PL08x primecell which the
339  * peripheral described by the channel data will be routed to, possibly
340  * via a board/SoC specific external MUX.  One important point to note
341  * here is that this does not depend on the physical channel.
342  */
343 static int pl08x_request_mux(struct pl08x_dma_chan *plchan)
344 {
345 	const struct pl08x_platform_data *pd = plchan->host->pd;
346 	int ret;
347 
348 	if (plchan->mux_use++ == 0 && pd->get_xfer_signal) {
349 		ret = pd->get_xfer_signal(plchan->cd);
350 		if (ret < 0) {
351 			plchan->mux_use = 0;
352 			return ret;
353 		}
354 
355 		plchan->signal = ret;
356 	}
357 	return 0;
358 }
359 
360 static void pl08x_release_mux(struct pl08x_dma_chan *plchan)
361 {
362 	const struct pl08x_platform_data *pd = plchan->host->pd;
363 
364 	if (plchan->signal >= 0) {
365 		WARN_ON(plchan->mux_use == 0);
366 
367 		if (--plchan->mux_use == 0 && pd->put_xfer_signal) {
368 			pd->put_xfer_signal(plchan->cd, plchan->signal);
369 			plchan->signal = -1;
370 		}
371 	}
372 }
373 
374 /*
375  * Physical channel handling
376  */
377 
378 /* Whether a certain channel is busy or not */
379 static int pl08x_phy_channel_busy(struct pl08x_phy_chan *ch)
380 {
381 	unsigned int val;
382 
383 	/* If we have a special busy register, take a shortcut */
384 	if (ch->reg_busy) {
385 		val = readl(ch->reg_busy);
386 		return !!(val & BIT(ch->id));
387 	}
388 	val = readl(ch->reg_config);
389 	return val & PL080_CONFIG_ACTIVE;
390 }
391 
392 /*
393  * pl08x_write_lli() - Write an LLI into the DMA controller.
394  *
395  * The PL08x derivatives support linked lists, but the first item of the
396  * list containing the source, destination, control word and next LLI is
397  * ignored. Instead the driver has to write those values directly into the
398  * SRC, DST, LLI and control registers. On FTDMAC020 also the SIZE
399  * register need to be set up for the first transfer.
400  */
401 static void pl08x_write_lli(struct pl08x_driver_data *pl08x,
402 		struct pl08x_phy_chan *phychan, const u32 *lli, u32 ccfg)
403 {
404 	if (pl08x->vd->pl080s)
405 		dev_vdbg(&pl08x->adev->dev,
406 			"WRITE channel %d: csrc=0x%08x, cdst=0x%08x, "
407 			"clli=0x%08x, cctl=0x%08x, cctl2=0x%08x, ccfg=0x%08x\n",
408 			phychan->id, lli[PL080_LLI_SRC], lli[PL080_LLI_DST],
409 			lli[PL080_LLI_LLI], lli[PL080_LLI_CCTL],
410 			lli[PL080S_LLI_CCTL2], ccfg);
411 	else
412 		dev_vdbg(&pl08x->adev->dev,
413 			"WRITE channel %d: csrc=0x%08x, cdst=0x%08x, "
414 			"clli=0x%08x, cctl=0x%08x, ccfg=0x%08x\n",
415 			phychan->id, lli[PL080_LLI_SRC], lli[PL080_LLI_DST],
416 			lli[PL080_LLI_LLI], lli[PL080_LLI_CCTL], ccfg);
417 
418 	writel_relaxed(lli[PL080_LLI_SRC], phychan->reg_src);
419 	writel_relaxed(lli[PL080_LLI_DST], phychan->reg_dst);
420 	writel_relaxed(lli[PL080_LLI_LLI], phychan->reg_lli);
421 
422 	/*
423 	 * The FTMAC020 has a different layout in the CCTL word of the LLI
424 	 * and the CCTL register which is split in CSR and SIZE registers.
425 	 * Convert the LLI item CCTL into the proper values to write into
426 	 * the CSR and SIZE registers.
427 	 */
428 	if (phychan->ftdmac020) {
429 		u32 llictl = lli[PL080_LLI_CCTL];
430 		u32 val = 0;
431 
432 		/* Write the transfer size (12 bits) to the size register */
433 		writel_relaxed(llictl & FTDMAC020_LLI_TRANSFER_SIZE_MASK,
434 			       phychan->base + FTDMAC020_CH_SIZE);
435 		/*
436 		 * Then write the control bits 28..16 to the control register
437 		 * by shuffleing the bits around to where they are in the
438 		 * main register. The mapping is as follows:
439 		 * Bit 28: TC_MSK - mask on all except last LLI
440 		 * Bit 27..25: SRC_WIDTH
441 		 * Bit 24..22: DST_WIDTH
442 		 * Bit 21..20: SRCAD_CTRL
443 		 * Bit 19..17: DSTAD_CTRL
444 		 * Bit 17: SRC_SEL
445 		 * Bit 16: DST_SEL
446 		 */
447 		if (llictl & FTDMAC020_LLI_TC_MSK)
448 			val |= FTDMAC020_CH_CSR_TC_MSK;
449 		val |= ((llictl  & FTDMAC020_LLI_SRC_WIDTH_MSK) >>
450 			(FTDMAC020_LLI_SRC_WIDTH_SHIFT -
451 			 FTDMAC020_CH_CSR_SRC_WIDTH_SHIFT));
452 		val |= ((llictl  & FTDMAC020_LLI_DST_WIDTH_MSK) >>
453 			(FTDMAC020_LLI_DST_WIDTH_SHIFT -
454 			 FTDMAC020_CH_CSR_DST_WIDTH_SHIFT));
455 		val |= ((llictl  & FTDMAC020_LLI_SRCAD_CTL_MSK) >>
456 			(FTDMAC020_LLI_SRCAD_CTL_SHIFT -
457 			 FTDMAC020_CH_CSR_SRCAD_CTL_SHIFT));
458 		val |= ((llictl  & FTDMAC020_LLI_DSTAD_CTL_MSK) >>
459 			(FTDMAC020_LLI_DSTAD_CTL_SHIFT -
460 			 FTDMAC020_CH_CSR_DSTAD_CTL_SHIFT));
461 		if (llictl & FTDMAC020_LLI_SRC_SEL)
462 			val |= FTDMAC020_CH_CSR_SRC_SEL;
463 		if (llictl & FTDMAC020_LLI_DST_SEL)
464 			val |= FTDMAC020_CH_CSR_DST_SEL;
465 
466 		/*
467 		 * Set up the bits that exist in the CSR but are not
468 		 * part the LLI, i.e. only gets written to the control
469 		 * register right here.
470 		 *
471 		 * FIXME: do not just handle memcpy, also handle slave DMA.
472 		 */
473 		switch (pl08x->pd->memcpy_burst_size) {
474 		default:
475 		case PL08X_BURST_SZ_1:
476 			val |= PL080_BSIZE_1 <<
477 				FTDMAC020_CH_CSR_SRC_SIZE_SHIFT;
478 			break;
479 		case PL08X_BURST_SZ_4:
480 			val |= PL080_BSIZE_4 <<
481 				FTDMAC020_CH_CSR_SRC_SIZE_SHIFT;
482 			break;
483 		case PL08X_BURST_SZ_8:
484 			val |= PL080_BSIZE_8 <<
485 				FTDMAC020_CH_CSR_SRC_SIZE_SHIFT;
486 			break;
487 		case PL08X_BURST_SZ_16:
488 			val |= PL080_BSIZE_16 <<
489 				FTDMAC020_CH_CSR_SRC_SIZE_SHIFT;
490 			break;
491 		case PL08X_BURST_SZ_32:
492 			val |= PL080_BSIZE_32 <<
493 				FTDMAC020_CH_CSR_SRC_SIZE_SHIFT;
494 			break;
495 		case PL08X_BURST_SZ_64:
496 			val |= PL080_BSIZE_64 <<
497 				FTDMAC020_CH_CSR_SRC_SIZE_SHIFT;
498 			break;
499 		case PL08X_BURST_SZ_128:
500 			val |= PL080_BSIZE_128 <<
501 				FTDMAC020_CH_CSR_SRC_SIZE_SHIFT;
502 			break;
503 		case PL08X_BURST_SZ_256:
504 			val |= PL080_BSIZE_256 <<
505 				FTDMAC020_CH_CSR_SRC_SIZE_SHIFT;
506 			break;
507 		}
508 
509 		/* Protection flags */
510 		if (pl08x->pd->memcpy_prot_buff)
511 			val |= FTDMAC020_CH_CSR_PROT2;
512 		if (pl08x->pd->memcpy_prot_cache)
513 			val |= FTDMAC020_CH_CSR_PROT3;
514 		/* We are the kernel, so we are in privileged mode */
515 		val |= FTDMAC020_CH_CSR_PROT1;
516 
517 		writel_relaxed(val, phychan->reg_control);
518 	} else {
519 		/* Bits are just identical */
520 		writel_relaxed(lli[PL080_LLI_CCTL], phychan->reg_control);
521 	}
522 
523 	/* Second control word on the PL080s */
524 	if (pl08x->vd->pl080s)
525 		writel_relaxed(lli[PL080S_LLI_CCTL2],
526 				phychan->base + PL080S_CH_CONTROL2);
527 
528 	writel(ccfg, phychan->reg_config);
529 }
530 
531 /*
532  * Set the initial DMA register values i.e. those for the first LLI
533  * The next LLI pointer and the configuration interrupt bit have
534  * been set when the LLIs were constructed.  Poke them into the hardware
535  * and start the transfer.
536  */
537 static void pl08x_start_next_txd(struct pl08x_dma_chan *plchan)
538 {
539 	struct pl08x_driver_data *pl08x = plchan->host;
540 	struct pl08x_phy_chan *phychan = plchan->phychan;
541 	struct virt_dma_desc *vd = vchan_next_desc(&plchan->vc);
542 	struct pl08x_txd *txd = to_pl08x_txd(&vd->tx);
543 	u32 val;
544 
545 	list_del(&txd->vd.node);
546 
547 	plchan->at = txd;
548 
549 	/* Wait for channel inactive */
550 	while (pl08x_phy_channel_busy(phychan))
551 		cpu_relax();
552 
553 	pl08x_write_lli(pl08x, phychan, &txd->llis_va[0], txd->ccfg);
554 
555 	/* Enable the DMA channel */
556 	/* Do not access config register until channel shows as disabled */
557 	while (readl(pl08x->base + PL080_EN_CHAN) & BIT(phychan->id))
558 		cpu_relax();
559 
560 	/* Do not access config register until channel shows as inactive */
561 	if (phychan->ftdmac020) {
562 		val = readl(phychan->reg_config);
563 		while (val & FTDMAC020_CH_CFG_BUSY)
564 			val = readl(phychan->reg_config);
565 
566 		val = readl(phychan->reg_control);
567 		while (val & FTDMAC020_CH_CSR_EN)
568 			val = readl(phychan->reg_control);
569 
570 		writel(val | FTDMAC020_CH_CSR_EN,
571 		       phychan->reg_control);
572 	} else {
573 		val = readl(phychan->reg_config);
574 		while ((val & PL080_CONFIG_ACTIVE) ||
575 		       (val & PL080_CONFIG_ENABLE))
576 			val = readl(phychan->reg_config);
577 
578 		writel(val | PL080_CONFIG_ENABLE, phychan->reg_config);
579 	}
580 }
581 
582 /*
583  * Pause the channel by setting the HALT bit.
584  *
585  * For M->P transfers, pause the DMAC first and then stop the peripheral -
586  * the FIFO can only drain if the peripheral is still requesting data.
587  * (note: this can still timeout if the DMAC FIFO never drains of data.)
588  *
589  * For P->M transfers, disable the peripheral first to stop it filling
590  * the DMAC FIFO, and then pause the DMAC.
591  */
592 static void pl08x_pause_phy_chan(struct pl08x_phy_chan *ch)
593 {
594 	u32 val;
595 	int timeout;
596 
597 	if (ch->ftdmac020) {
598 		/* Use the enable bit on the FTDMAC020 */
599 		val = readl(ch->reg_control);
600 		val &= ~FTDMAC020_CH_CSR_EN;
601 		writel(val, ch->reg_control);
602 		return;
603 	}
604 
605 	/* Set the HALT bit and wait for the FIFO to drain */
606 	val = readl(ch->reg_config);
607 	val |= PL080_CONFIG_HALT;
608 	writel(val, ch->reg_config);
609 
610 	/* Wait for channel inactive */
611 	for (timeout = 1000; timeout; timeout--) {
612 		if (!pl08x_phy_channel_busy(ch))
613 			break;
614 		udelay(1);
615 	}
616 	if (pl08x_phy_channel_busy(ch))
617 		pr_err("pl08x: channel%u timeout waiting for pause\n", ch->id);
618 }
619 
620 static void pl08x_resume_phy_chan(struct pl08x_phy_chan *ch)
621 {
622 	u32 val;
623 
624 	/* Use the enable bit on the FTDMAC020 */
625 	if (ch->ftdmac020) {
626 		val = readl(ch->reg_control);
627 		val |= FTDMAC020_CH_CSR_EN;
628 		writel(val, ch->reg_control);
629 		return;
630 	}
631 
632 	/* Clear the HALT bit */
633 	val = readl(ch->reg_config);
634 	val &= ~PL080_CONFIG_HALT;
635 	writel(val, ch->reg_config);
636 }
637 
638 /*
639  * pl08x_terminate_phy_chan() stops the channel, clears the FIFO and
640  * clears any pending interrupt status.  This should not be used for
641  * an on-going transfer, but as a method of shutting down a channel
642  * (eg, when it's no longer used) or terminating a transfer.
643  */
644 static void pl08x_terminate_phy_chan(struct pl08x_driver_data *pl08x,
645 	struct pl08x_phy_chan *ch)
646 {
647 	u32 val;
648 
649 	/* The layout for the FTDMAC020 is different */
650 	if (ch->ftdmac020) {
651 		/* Disable all interrupts */
652 		val = readl(ch->reg_config);
653 		val |= (FTDMAC020_CH_CFG_INT_ABT_MASK |
654 			FTDMAC020_CH_CFG_INT_ERR_MASK |
655 			FTDMAC020_CH_CFG_INT_TC_MASK);
656 		writel(val, ch->reg_config);
657 
658 		/* Abort and disable channel */
659 		val = readl(ch->reg_control);
660 		val &= ~FTDMAC020_CH_CSR_EN;
661 		val |= FTDMAC020_CH_CSR_ABT;
662 		writel(val, ch->reg_control);
663 
664 		/* Clear ABT and ERR interrupt flags */
665 		writel(BIT(ch->id) | BIT(ch->id + 16),
666 		       pl08x->base + PL080_ERR_CLEAR);
667 		writel(BIT(ch->id), pl08x->base + PL080_TC_CLEAR);
668 
669 		return;
670 	}
671 
672 	val = readl(ch->reg_config);
673 	val &= ~(PL080_CONFIG_ENABLE | PL080_CONFIG_ERR_IRQ_MASK |
674 		 PL080_CONFIG_TC_IRQ_MASK);
675 	writel(val, ch->reg_config);
676 
677 	writel(BIT(ch->id), pl08x->base + PL080_ERR_CLEAR);
678 	writel(BIT(ch->id), pl08x->base + PL080_TC_CLEAR);
679 }
680 
681 static u32 get_bytes_in_phy_channel(struct pl08x_phy_chan *ch)
682 {
683 	u32 val;
684 	u32 bytes;
685 
686 	if (ch->ftdmac020) {
687 		bytes = readl(ch->base + FTDMAC020_CH_SIZE);
688 
689 		val = readl(ch->reg_control);
690 		val &= FTDMAC020_CH_CSR_SRC_WIDTH_MSK;
691 		val >>= FTDMAC020_CH_CSR_SRC_WIDTH_SHIFT;
692 	} else if (ch->pl080s) {
693 		val = readl(ch->base + PL080S_CH_CONTROL2);
694 		bytes = val & PL080S_CONTROL_TRANSFER_SIZE_MASK;
695 
696 		val = readl(ch->reg_control);
697 		val &= PL080_CONTROL_SWIDTH_MASK;
698 		val >>= PL080_CONTROL_SWIDTH_SHIFT;
699 	} else {
700 		/* Plain PL08x */
701 		val = readl(ch->reg_control);
702 		bytes = val & PL080_CONTROL_TRANSFER_SIZE_MASK;
703 
704 		val &= PL080_CONTROL_SWIDTH_MASK;
705 		val >>= PL080_CONTROL_SWIDTH_SHIFT;
706 	}
707 
708 	switch (val) {
709 	case PL080_WIDTH_8BIT:
710 		break;
711 	case PL080_WIDTH_16BIT:
712 		bytes *= 2;
713 		break;
714 	case PL080_WIDTH_32BIT:
715 		bytes *= 4;
716 		break;
717 	}
718 	return bytes;
719 }
720 
721 static u32 get_bytes_in_lli(struct pl08x_phy_chan *ch, const u32 *llis_va)
722 {
723 	u32 val;
724 	u32 bytes;
725 
726 	if (ch->ftdmac020) {
727 		val = llis_va[PL080_LLI_CCTL];
728 		bytes = val & FTDMAC020_LLI_TRANSFER_SIZE_MASK;
729 
730 		val = llis_va[PL080_LLI_CCTL];
731 		val &= FTDMAC020_LLI_SRC_WIDTH_MSK;
732 		val >>= FTDMAC020_LLI_SRC_WIDTH_SHIFT;
733 	} else if (ch->pl080s) {
734 		val = llis_va[PL080S_LLI_CCTL2];
735 		bytes = val & PL080S_CONTROL_TRANSFER_SIZE_MASK;
736 
737 		val = llis_va[PL080_LLI_CCTL];
738 		val &= PL080_CONTROL_SWIDTH_MASK;
739 		val >>= PL080_CONTROL_SWIDTH_SHIFT;
740 	} else {
741 		/* Plain PL08x */
742 		val = llis_va[PL080_LLI_CCTL];
743 		bytes = val & PL080_CONTROL_TRANSFER_SIZE_MASK;
744 
745 		val &= PL080_CONTROL_SWIDTH_MASK;
746 		val >>= PL080_CONTROL_SWIDTH_SHIFT;
747 	}
748 
749 	switch (val) {
750 	case PL080_WIDTH_8BIT:
751 		break;
752 	case PL080_WIDTH_16BIT:
753 		bytes *= 2;
754 		break;
755 	case PL080_WIDTH_32BIT:
756 		bytes *= 4;
757 		break;
758 	}
759 	return bytes;
760 }
761 
762 /* The channel should be paused when calling this */
763 static u32 pl08x_getbytes_chan(struct pl08x_dma_chan *plchan)
764 {
765 	struct pl08x_driver_data *pl08x = plchan->host;
766 	const u32 *llis_va, *llis_va_limit;
767 	struct pl08x_phy_chan *ch;
768 	dma_addr_t llis_bus;
769 	struct pl08x_txd *txd;
770 	u32 llis_max_words;
771 	size_t bytes;
772 	u32 clli;
773 
774 	ch = plchan->phychan;
775 	txd = plchan->at;
776 
777 	if (!ch || !txd)
778 		return 0;
779 
780 	/*
781 	 * Follow the LLIs to get the number of remaining
782 	 * bytes in the currently active transaction.
783 	 */
784 	clli = readl(ch->reg_lli) & ~PL080_LLI_LM_AHB2;
785 
786 	/* First get the remaining bytes in the active transfer */
787 	bytes = get_bytes_in_phy_channel(ch);
788 
789 	if (!clli)
790 		return bytes;
791 
792 	llis_va = txd->llis_va;
793 	llis_bus = txd->llis_bus;
794 
795 	llis_max_words = pl08x->lli_words * MAX_NUM_TSFR_LLIS;
796 	BUG_ON(clli < llis_bus || clli >= llis_bus +
797 						sizeof(u32) * llis_max_words);
798 
799 	/*
800 	 * Locate the next LLI - as this is an array,
801 	 * it's simple maths to find.
802 	 */
803 	llis_va += (clli - llis_bus) / sizeof(u32);
804 
805 	llis_va_limit = llis_va + llis_max_words;
806 
807 	for (; llis_va < llis_va_limit; llis_va += pl08x->lli_words) {
808 		bytes += get_bytes_in_lli(ch, llis_va);
809 
810 		/*
811 		 * A LLI pointer going backward terminates the LLI list
812 		 */
813 		if (llis_va[PL080_LLI_LLI] <= clli)
814 			break;
815 	}
816 
817 	return bytes;
818 }
819 
820 /*
821  * Allocate a physical channel for a virtual channel
822  *
823  * Try to locate a physical channel to be used for this transfer. If all
824  * are taken return NULL and the requester will have to cope by using
825  * some fallback PIO mode or retrying later.
826  */
827 static struct pl08x_phy_chan *
828 pl08x_get_phy_channel(struct pl08x_driver_data *pl08x,
829 		      struct pl08x_dma_chan *virt_chan)
830 {
831 	struct pl08x_phy_chan *ch = NULL;
832 	unsigned long flags;
833 	int i;
834 
835 	for (i = 0; i < pl08x->vd->channels; i++) {
836 		ch = &pl08x->phy_chans[i];
837 
838 		spin_lock_irqsave(&ch->lock, flags);
839 
840 		if (!ch->locked && !ch->serving) {
841 			ch->serving = virt_chan;
842 			spin_unlock_irqrestore(&ch->lock, flags);
843 			break;
844 		}
845 
846 		spin_unlock_irqrestore(&ch->lock, flags);
847 	}
848 
849 	if (i == pl08x->vd->channels) {
850 		/* No physical channel available, cope with it */
851 		return NULL;
852 	}
853 
854 	return ch;
855 }
856 
857 /* Mark the physical channel as free.  Note, this write is atomic. */
858 static inline void pl08x_put_phy_channel(struct pl08x_driver_data *pl08x,
859 					 struct pl08x_phy_chan *ch)
860 {
861 	ch->serving = NULL;
862 }
863 
864 /*
865  * Try to allocate a physical channel.  When successful, assign it to
866  * this virtual channel, and initiate the next descriptor.  The
867  * virtual channel lock must be held at this point.
868  */
869 static void pl08x_phy_alloc_and_start(struct pl08x_dma_chan *plchan)
870 {
871 	struct pl08x_driver_data *pl08x = plchan->host;
872 	struct pl08x_phy_chan *ch;
873 
874 	ch = pl08x_get_phy_channel(pl08x, plchan);
875 	if (!ch) {
876 		dev_dbg(&pl08x->adev->dev, "no physical channel available for xfer on %s\n", plchan->name);
877 		plchan->state = PL08X_CHAN_WAITING;
878 		return;
879 	}
880 
881 	dev_dbg(&pl08x->adev->dev, "allocated physical channel %d for xfer on %s\n",
882 		ch->id, plchan->name);
883 
884 	plchan->phychan = ch;
885 	plchan->state = PL08X_CHAN_RUNNING;
886 	pl08x_start_next_txd(plchan);
887 }
888 
889 static void pl08x_phy_reassign_start(struct pl08x_phy_chan *ch,
890 	struct pl08x_dma_chan *plchan)
891 {
892 	struct pl08x_driver_data *pl08x = plchan->host;
893 
894 	dev_dbg(&pl08x->adev->dev, "reassigned physical channel %d for xfer on %s\n",
895 		ch->id, plchan->name);
896 
897 	/*
898 	 * We do this without taking the lock; we're really only concerned
899 	 * about whether this pointer is NULL or not, and we're guaranteed
900 	 * that this will only be called when it _already_ is non-NULL.
901 	 */
902 	ch->serving = plchan;
903 	plchan->phychan = ch;
904 	plchan->state = PL08X_CHAN_RUNNING;
905 	pl08x_start_next_txd(plchan);
906 }
907 
908 /*
909  * Free a physical DMA channel, potentially reallocating it to another
910  * virtual channel if we have any pending.
911  */
912 static void pl08x_phy_free(struct pl08x_dma_chan *plchan)
913 {
914 	struct pl08x_driver_data *pl08x = plchan->host;
915 	struct pl08x_dma_chan *p, *next;
916 
917  retry:
918 	next = NULL;
919 
920 	/* Find a waiting virtual channel for the next transfer. */
921 	list_for_each_entry(p, &pl08x->memcpy.channels, vc.chan.device_node)
922 		if (p->state == PL08X_CHAN_WAITING) {
923 			next = p;
924 			break;
925 		}
926 
927 	if (!next && pl08x->has_slave) {
928 		list_for_each_entry(p, &pl08x->slave.channels, vc.chan.device_node)
929 			if (p->state == PL08X_CHAN_WAITING) {
930 				next = p;
931 				break;
932 			}
933 	}
934 
935 	/* Ensure that the physical channel is stopped */
936 	pl08x_terminate_phy_chan(pl08x, plchan->phychan);
937 
938 	if (next) {
939 		bool success;
940 
941 		/*
942 		 * Eww.  We know this isn't going to deadlock
943 		 * but lockdep probably doesn't.
944 		 */
945 		spin_lock(&next->vc.lock);
946 		/* Re-check the state now that we have the lock */
947 		success = next->state == PL08X_CHAN_WAITING;
948 		if (success)
949 			pl08x_phy_reassign_start(plchan->phychan, next);
950 		spin_unlock(&next->vc.lock);
951 
952 		/* If the state changed, try to find another channel */
953 		if (!success)
954 			goto retry;
955 	} else {
956 		/* No more jobs, so free up the physical channel */
957 		pl08x_put_phy_channel(pl08x, plchan->phychan);
958 	}
959 
960 	plchan->phychan = NULL;
961 	plchan->state = PL08X_CHAN_IDLE;
962 }
963 
964 /*
965  * LLI handling
966  */
967 
968 static inline unsigned int
969 pl08x_get_bytes_for_lli(struct pl08x_driver_data *pl08x,
970 			u32 cctl,
971 			bool source)
972 {
973 	u32 val;
974 
975 	if (pl08x->vd->ftdmac020) {
976 		if (source)
977 			val = (cctl & FTDMAC020_LLI_SRC_WIDTH_MSK) >>
978 				FTDMAC020_LLI_SRC_WIDTH_SHIFT;
979 		else
980 			val = (cctl & FTDMAC020_LLI_DST_WIDTH_MSK) >>
981 				FTDMAC020_LLI_DST_WIDTH_SHIFT;
982 	} else {
983 		if (source)
984 			val = (cctl & PL080_CONTROL_SWIDTH_MASK) >>
985 				PL080_CONTROL_SWIDTH_SHIFT;
986 		else
987 			val = (cctl & PL080_CONTROL_DWIDTH_MASK) >>
988 				PL080_CONTROL_DWIDTH_SHIFT;
989 	}
990 
991 	switch (val) {
992 	case PL080_WIDTH_8BIT:
993 		return 1;
994 	case PL080_WIDTH_16BIT:
995 		return 2;
996 	case PL080_WIDTH_32BIT:
997 		return 4;
998 	default:
999 		break;
1000 	}
1001 	BUG();
1002 	return 0;
1003 }
1004 
1005 static inline u32 pl08x_lli_control_bits(struct pl08x_driver_data *pl08x,
1006 					 u32 cctl,
1007 					 u8 srcwidth, u8 dstwidth,
1008 					 size_t tsize)
1009 {
1010 	u32 retbits = cctl;
1011 
1012 	/*
1013 	 * Remove all src, dst and transfer size bits, then set the
1014 	 * width and size according to the parameters. The bit offsets
1015 	 * are different in the FTDMAC020 so we need to accound for this.
1016 	 */
1017 	if (pl08x->vd->ftdmac020) {
1018 		retbits &= ~FTDMAC020_LLI_DST_WIDTH_MSK;
1019 		retbits &= ~FTDMAC020_LLI_SRC_WIDTH_MSK;
1020 		retbits &= ~FTDMAC020_LLI_TRANSFER_SIZE_MASK;
1021 
1022 		switch (srcwidth) {
1023 		case 1:
1024 			retbits |= PL080_WIDTH_8BIT <<
1025 				FTDMAC020_LLI_SRC_WIDTH_SHIFT;
1026 			break;
1027 		case 2:
1028 			retbits |= PL080_WIDTH_16BIT <<
1029 				FTDMAC020_LLI_SRC_WIDTH_SHIFT;
1030 			break;
1031 		case 4:
1032 			retbits |= PL080_WIDTH_32BIT <<
1033 				FTDMAC020_LLI_SRC_WIDTH_SHIFT;
1034 			break;
1035 		default:
1036 			BUG();
1037 			break;
1038 		}
1039 
1040 		switch (dstwidth) {
1041 		case 1:
1042 			retbits |= PL080_WIDTH_8BIT <<
1043 				FTDMAC020_LLI_DST_WIDTH_SHIFT;
1044 			break;
1045 		case 2:
1046 			retbits |= PL080_WIDTH_16BIT <<
1047 				FTDMAC020_LLI_DST_WIDTH_SHIFT;
1048 			break;
1049 		case 4:
1050 			retbits |= PL080_WIDTH_32BIT <<
1051 				FTDMAC020_LLI_DST_WIDTH_SHIFT;
1052 			break;
1053 		default:
1054 			BUG();
1055 			break;
1056 		}
1057 
1058 		tsize &= FTDMAC020_LLI_TRANSFER_SIZE_MASK;
1059 		retbits |= tsize << FTDMAC020_LLI_TRANSFER_SIZE_SHIFT;
1060 	} else {
1061 		retbits &= ~PL080_CONTROL_DWIDTH_MASK;
1062 		retbits &= ~PL080_CONTROL_SWIDTH_MASK;
1063 		retbits &= ~PL080_CONTROL_TRANSFER_SIZE_MASK;
1064 
1065 		switch (srcwidth) {
1066 		case 1:
1067 			retbits |= PL080_WIDTH_8BIT <<
1068 				PL080_CONTROL_SWIDTH_SHIFT;
1069 			break;
1070 		case 2:
1071 			retbits |= PL080_WIDTH_16BIT <<
1072 				PL080_CONTROL_SWIDTH_SHIFT;
1073 			break;
1074 		case 4:
1075 			retbits |= PL080_WIDTH_32BIT <<
1076 				PL080_CONTROL_SWIDTH_SHIFT;
1077 			break;
1078 		default:
1079 			BUG();
1080 			break;
1081 		}
1082 
1083 		switch (dstwidth) {
1084 		case 1:
1085 			retbits |= PL080_WIDTH_8BIT <<
1086 				PL080_CONTROL_DWIDTH_SHIFT;
1087 			break;
1088 		case 2:
1089 			retbits |= PL080_WIDTH_16BIT <<
1090 				PL080_CONTROL_DWIDTH_SHIFT;
1091 			break;
1092 		case 4:
1093 			retbits |= PL080_WIDTH_32BIT <<
1094 				PL080_CONTROL_DWIDTH_SHIFT;
1095 			break;
1096 		default:
1097 			BUG();
1098 			break;
1099 		}
1100 
1101 		tsize &= PL080_CONTROL_TRANSFER_SIZE_MASK;
1102 		retbits |= tsize << PL080_CONTROL_TRANSFER_SIZE_SHIFT;
1103 	}
1104 
1105 	return retbits;
1106 }
1107 
1108 struct pl08x_lli_build_data {
1109 	struct pl08x_txd *txd;
1110 	struct pl08x_bus_data srcbus;
1111 	struct pl08x_bus_data dstbus;
1112 	size_t remainder;
1113 	u32 lli_bus;
1114 };
1115 
1116 /*
1117  * Autoselect a master bus to use for the transfer. Slave will be the chosen as
1118  * victim in case src & dest are not similarly aligned. i.e. If after aligning
1119  * masters address with width requirements of transfer (by sending few byte by
1120  * byte data), slave is still not aligned, then its width will be reduced to
1121  * BYTE.
1122  * - prefers the destination bus if both available
1123  * - prefers bus with fixed address (i.e. peripheral)
1124  */
1125 static void pl08x_choose_master_bus(struct pl08x_driver_data *pl08x,
1126 				    struct pl08x_lli_build_data *bd,
1127 				    struct pl08x_bus_data **mbus,
1128 				    struct pl08x_bus_data **sbus,
1129 				    u32 cctl)
1130 {
1131 	bool dst_incr;
1132 	bool src_incr;
1133 
1134 	/*
1135 	 * The FTDMAC020 only supports memory-to-memory transfer, so
1136 	 * source and destination always increase.
1137 	 */
1138 	if (pl08x->vd->ftdmac020) {
1139 		dst_incr = true;
1140 		src_incr = true;
1141 	} else {
1142 		dst_incr = !!(cctl & PL080_CONTROL_DST_INCR);
1143 		src_incr = !!(cctl & PL080_CONTROL_SRC_INCR);
1144 	}
1145 
1146 	/*
1147 	 * If either bus is not advancing, i.e. it is a peripheral, that
1148 	 * one becomes master
1149 	 */
1150 	if (!dst_incr) {
1151 		*mbus = &bd->dstbus;
1152 		*sbus = &bd->srcbus;
1153 	} else if (!src_incr) {
1154 		*mbus = &bd->srcbus;
1155 		*sbus = &bd->dstbus;
1156 	} else {
1157 		if (bd->dstbus.buswidth >= bd->srcbus.buswidth) {
1158 			*mbus = &bd->dstbus;
1159 			*sbus = &bd->srcbus;
1160 		} else {
1161 			*mbus = &bd->srcbus;
1162 			*sbus = &bd->dstbus;
1163 		}
1164 	}
1165 }
1166 
1167 /*
1168  * Fills in one LLI for a certain transfer descriptor and advance the counter
1169  */
1170 static void pl08x_fill_lli_for_desc(struct pl08x_driver_data *pl08x,
1171 				    struct pl08x_lli_build_data *bd,
1172 				    int num_llis, int len, u32 cctl, u32 cctl2)
1173 {
1174 	u32 offset = num_llis * pl08x->lli_words;
1175 	u32 *llis_va = bd->txd->llis_va + offset;
1176 	dma_addr_t llis_bus = bd->txd->llis_bus;
1177 
1178 	BUG_ON(num_llis >= MAX_NUM_TSFR_LLIS);
1179 
1180 	/* Advance the offset to next LLI. */
1181 	offset += pl08x->lli_words;
1182 
1183 	llis_va[PL080_LLI_SRC] = bd->srcbus.addr;
1184 	llis_va[PL080_LLI_DST] = bd->dstbus.addr;
1185 	llis_va[PL080_LLI_LLI] = (llis_bus + sizeof(u32) * offset);
1186 	llis_va[PL080_LLI_LLI] |= bd->lli_bus;
1187 	llis_va[PL080_LLI_CCTL] = cctl;
1188 	if (pl08x->vd->pl080s)
1189 		llis_va[PL080S_LLI_CCTL2] = cctl2;
1190 
1191 	if (pl08x->vd->ftdmac020) {
1192 		/* FIXME: only memcpy so far so both increase */
1193 		bd->srcbus.addr += len;
1194 		bd->dstbus.addr += len;
1195 	} else {
1196 		if (cctl & PL080_CONTROL_SRC_INCR)
1197 			bd->srcbus.addr += len;
1198 		if (cctl & PL080_CONTROL_DST_INCR)
1199 			bd->dstbus.addr += len;
1200 	}
1201 
1202 	BUG_ON(bd->remainder < len);
1203 
1204 	bd->remainder -= len;
1205 }
1206 
1207 static inline void prep_byte_width_lli(struct pl08x_driver_data *pl08x,
1208 			struct pl08x_lli_build_data *bd, u32 *cctl, u32 len,
1209 			int num_llis, size_t *total_bytes)
1210 {
1211 	*cctl = pl08x_lli_control_bits(pl08x, *cctl, 1, 1, len);
1212 	pl08x_fill_lli_for_desc(pl08x, bd, num_llis, len, *cctl, len);
1213 	(*total_bytes) += len;
1214 }
1215 
1216 #if 1
1217 static void pl08x_dump_lli(struct pl08x_driver_data *pl08x,
1218 			   const u32 *llis_va, int num_llis)
1219 {
1220 	int i;
1221 
1222 	if (pl08x->vd->pl080s) {
1223 		dev_vdbg(&pl08x->adev->dev,
1224 			"%-3s %-9s  %-10s %-10s %-10s %-10s %s\n",
1225 			"lli", "", "csrc", "cdst", "clli", "cctl", "cctl2");
1226 		for (i = 0; i < num_llis; i++) {
1227 			dev_vdbg(&pl08x->adev->dev,
1228 				"%3d @%p: 0x%08x 0x%08x 0x%08x 0x%08x 0x%08x\n",
1229 				i, llis_va, llis_va[PL080_LLI_SRC],
1230 				llis_va[PL080_LLI_DST], llis_va[PL080_LLI_LLI],
1231 				llis_va[PL080_LLI_CCTL],
1232 				llis_va[PL080S_LLI_CCTL2]);
1233 			llis_va += pl08x->lli_words;
1234 		}
1235 	} else {
1236 		dev_vdbg(&pl08x->adev->dev,
1237 			"%-3s %-9s  %-10s %-10s %-10s %s\n",
1238 			"lli", "", "csrc", "cdst", "clli", "cctl");
1239 		for (i = 0; i < num_llis; i++) {
1240 			dev_vdbg(&pl08x->adev->dev,
1241 				"%3d @%p: 0x%08x 0x%08x 0x%08x 0x%08x\n",
1242 				i, llis_va, llis_va[PL080_LLI_SRC],
1243 				llis_va[PL080_LLI_DST], llis_va[PL080_LLI_LLI],
1244 				llis_va[PL080_LLI_CCTL]);
1245 			llis_va += pl08x->lli_words;
1246 		}
1247 	}
1248 }
1249 #else
1250 static inline void pl08x_dump_lli(struct pl08x_driver_data *pl08x,
1251 				  const u32 *llis_va, int num_llis) {}
1252 #endif
1253 
1254 /*
1255  * This fills in the table of LLIs for the transfer descriptor
1256  * Note that we assume we never have to change the burst sizes
1257  * Return 0 for error
1258  */
1259 static int pl08x_fill_llis_for_desc(struct pl08x_driver_data *pl08x,
1260 			      struct pl08x_txd *txd)
1261 {
1262 	struct pl08x_bus_data *mbus, *sbus;
1263 	struct pl08x_lli_build_data bd;
1264 	int num_llis = 0;
1265 	u32 cctl, early_bytes = 0;
1266 	size_t max_bytes_per_lli, total_bytes;
1267 	u32 *llis_va, *last_lli;
1268 	struct pl08x_sg *dsg;
1269 
1270 	txd->llis_va = dma_pool_alloc(pl08x->pool, GFP_NOWAIT, &txd->llis_bus);
1271 	if (!txd->llis_va) {
1272 		dev_err(&pl08x->adev->dev, "%s no memory for llis\n", __func__);
1273 		return 0;
1274 	}
1275 
1276 	bd.txd = txd;
1277 	bd.lli_bus = (pl08x->lli_buses & PL08X_AHB2) ? PL080_LLI_LM_AHB2 : 0;
1278 	cctl = txd->cctl;
1279 
1280 	/* Find maximum width of the source bus */
1281 	bd.srcbus.maxwidth = pl08x_get_bytes_for_lli(pl08x, cctl, true);
1282 
1283 	/* Find maximum width of the destination bus */
1284 	bd.dstbus.maxwidth = pl08x_get_bytes_for_lli(pl08x, cctl, false);
1285 
1286 	list_for_each_entry(dsg, &txd->dsg_list, node) {
1287 		total_bytes = 0;
1288 		cctl = txd->cctl;
1289 
1290 		bd.srcbus.addr = dsg->src_addr;
1291 		bd.dstbus.addr = dsg->dst_addr;
1292 		bd.remainder = dsg->len;
1293 		bd.srcbus.buswidth = bd.srcbus.maxwidth;
1294 		bd.dstbus.buswidth = bd.dstbus.maxwidth;
1295 
1296 		pl08x_choose_master_bus(pl08x, &bd, &mbus, &sbus, cctl);
1297 
1298 		dev_vdbg(&pl08x->adev->dev,
1299 			"src=0x%08llx%s/%u dst=0x%08llx%s/%u len=%zu\n",
1300 			(u64)bd.srcbus.addr,
1301 			cctl & PL080_CONTROL_SRC_INCR ? "+" : "",
1302 			bd.srcbus.buswidth,
1303 			(u64)bd.dstbus.addr,
1304 			cctl & PL080_CONTROL_DST_INCR ? "+" : "",
1305 			bd.dstbus.buswidth,
1306 			bd.remainder);
1307 		dev_vdbg(&pl08x->adev->dev, "mbus=%s sbus=%s\n",
1308 			mbus == &bd.srcbus ? "src" : "dst",
1309 			sbus == &bd.srcbus ? "src" : "dst");
1310 
1311 		/*
1312 		 * Zero length is only allowed if all these requirements are
1313 		 * met:
1314 		 * - flow controller is peripheral.
1315 		 * - src.addr is aligned to src.width
1316 		 * - dst.addr is aligned to dst.width
1317 		 *
1318 		 * sg_len == 1 should be true, as there can be two cases here:
1319 		 *
1320 		 * - Memory addresses are contiguous and are not scattered.
1321 		 *   Here, Only one sg will be passed by user driver, with
1322 		 *   memory address and zero length. We pass this to controller
1323 		 *   and after the transfer it will receive the last burst
1324 		 *   request from peripheral and so transfer finishes.
1325 		 *
1326 		 * - Memory addresses are scattered and are not contiguous.
1327 		 *   Here, Obviously as DMA controller doesn't know when a lli's
1328 		 *   transfer gets over, it can't load next lli. So in this
1329 		 *   case, there has to be an assumption that only one lli is
1330 		 *   supported. Thus, we can't have scattered addresses.
1331 		 */
1332 		if (!bd.remainder) {
1333 			u32 fc;
1334 
1335 			/* FTDMAC020 only does memory-to-memory */
1336 			if (pl08x->vd->ftdmac020)
1337 				fc = PL080_FLOW_MEM2MEM;
1338 			else
1339 				fc = (txd->ccfg & PL080_CONFIG_FLOW_CONTROL_MASK) >>
1340 					PL080_CONFIG_FLOW_CONTROL_SHIFT;
1341 			if (!((fc >= PL080_FLOW_SRC2DST_DST) &&
1342 					(fc <= PL080_FLOW_SRC2DST_SRC))) {
1343 				dev_err(&pl08x->adev->dev, "%s sg len can't be zero",
1344 					__func__);
1345 				return 0;
1346 			}
1347 
1348 			if (!IS_BUS_ALIGNED(&bd.srcbus) ||
1349 				!IS_BUS_ALIGNED(&bd.dstbus)) {
1350 				dev_err(&pl08x->adev->dev,
1351 					"%s src & dst address must be aligned to src"
1352 					" & dst width if peripheral is flow controller",
1353 					__func__);
1354 				return 0;
1355 			}
1356 
1357 			cctl = pl08x_lli_control_bits(pl08x, cctl,
1358 					bd.srcbus.buswidth, bd.dstbus.buswidth,
1359 					0);
1360 			pl08x_fill_lli_for_desc(pl08x, &bd, num_llis++,
1361 					0, cctl, 0);
1362 			break;
1363 		}
1364 
1365 		/*
1366 		 * Send byte by byte for following cases
1367 		 * - Less than a bus width available
1368 		 * - until master bus is aligned
1369 		 */
1370 		if (bd.remainder < mbus->buswidth)
1371 			early_bytes = bd.remainder;
1372 		else if (!IS_BUS_ALIGNED(mbus)) {
1373 			early_bytes = mbus->buswidth -
1374 				(mbus->addr & (mbus->buswidth - 1));
1375 			if ((bd.remainder - early_bytes) < mbus->buswidth)
1376 				early_bytes = bd.remainder;
1377 		}
1378 
1379 		if (early_bytes) {
1380 			dev_vdbg(&pl08x->adev->dev,
1381 				"%s byte width LLIs (remain 0x%08zx)\n",
1382 				__func__, bd.remainder);
1383 			prep_byte_width_lli(pl08x, &bd, &cctl, early_bytes,
1384 				num_llis++, &total_bytes);
1385 		}
1386 
1387 		if (bd.remainder) {
1388 			/*
1389 			 * Master now aligned
1390 			 * - if slave is not then we must set its width down
1391 			 */
1392 			if (!IS_BUS_ALIGNED(sbus)) {
1393 				dev_dbg(&pl08x->adev->dev,
1394 					"%s set down bus width to one byte\n",
1395 					__func__);
1396 
1397 				sbus->buswidth = 1;
1398 			}
1399 
1400 			/*
1401 			 * Bytes transferred = tsize * src width, not
1402 			 * MIN(buswidths)
1403 			 */
1404 			max_bytes_per_lli = bd.srcbus.buswidth *
1405 						pl08x->vd->max_transfer_size;
1406 			dev_vdbg(&pl08x->adev->dev,
1407 				"%s max bytes per lli = %zu\n",
1408 				__func__, max_bytes_per_lli);
1409 
1410 			/*
1411 			 * Make largest possible LLIs until less than one bus
1412 			 * width left
1413 			 */
1414 			while (bd.remainder > (mbus->buswidth - 1)) {
1415 				size_t lli_len, tsize, width;
1416 
1417 				/*
1418 				 * If enough left try to send max possible,
1419 				 * otherwise try to send the remainder
1420 				 */
1421 				lli_len = min(bd.remainder, max_bytes_per_lli);
1422 
1423 				/*
1424 				 * Check against maximum bus alignment:
1425 				 * Calculate actual transfer size in relation to
1426 				 * bus width an get a maximum remainder of the
1427 				 * highest bus width - 1
1428 				 */
1429 				width = max(mbus->buswidth, sbus->buswidth);
1430 				lli_len = (lli_len / width) * width;
1431 				tsize = lli_len / bd.srcbus.buswidth;
1432 
1433 				dev_vdbg(&pl08x->adev->dev,
1434 					"%s fill lli with single lli chunk of "
1435 					"size 0x%08zx (remainder 0x%08zx)\n",
1436 					__func__, lli_len, bd.remainder);
1437 
1438 				cctl = pl08x_lli_control_bits(pl08x, cctl,
1439 					bd.srcbus.buswidth, bd.dstbus.buswidth,
1440 					tsize);
1441 				pl08x_fill_lli_for_desc(pl08x, &bd, num_llis++,
1442 						lli_len, cctl, tsize);
1443 				total_bytes += lli_len;
1444 			}
1445 
1446 			/*
1447 			 * Send any odd bytes
1448 			 */
1449 			if (bd.remainder) {
1450 				dev_vdbg(&pl08x->adev->dev,
1451 					"%s align with boundary, send odd bytes (remain %zu)\n",
1452 					__func__, bd.remainder);
1453 				prep_byte_width_lli(pl08x, &bd, &cctl,
1454 					bd.remainder, num_llis++, &total_bytes);
1455 			}
1456 		}
1457 
1458 		if (total_bytes != dsg->len) {
1459 			dev_err(&pl08x->adev->dev,
1460 				"%s size of encoded lli:s don't match total txd, transferred 0x%08zx from size 0x%08zx\n",
1461 				__func__, total_bytes, dsg->len);
1462 			return 0;
1463 		}
1464 
1465 		if (num_llis >= MAX_NUM_TSFR_LLIS) {
1466 			dev_err(&pl08x->adev->dev,
1467 				"%s need to increase MAX_NUM_TSFR_LLIS from 0x%08x\n",
1468 				__func__, MAX_NUM_TSFR_LLIS);
1469 			return 0;
1470 		}
1471 	}
1472 
1473 	llis_va = txd->llis_va;
1474 	last_lli = llis_va + (num_llis - 1) * pl08x->lli_words;
1475 
1476 	if (txd->cyclic) {
1477 		/* Link back to the first LLI. */
1478 		last_lli[PL080_LLI_LLI] = txd->llis_bus | bd.lli_bus;
1479 	} else {
1480 		/* The final LLI terminates the LLI. */
1481 		last_lli[PL080_LLI_LLI] = 0;
1482 		/* The final LLI element shall also fire an interrupt. */
1483 		if (pl08x->vd->ftdmac020)
1484 			last_lli[PL080_LLI_CCTL] &= ~FTDMAC020_LLI_TC_MSK;
1485 		else
1486 			last_lli[PL080_LLI_CCTL] |= PL080_CONTROL_TC_IRQ_EN;
1487 	}
1488 
1489 	pl08x_dump_lli(pl08x, llis_va, num_llis);
1490 
1491 	return num_llis;
1492 }
1493 
1494 static void pl08x_free_txd(struct pl08x_driver_data *pl08x,
1495 			   struct pl08x_txd *txd)
1496 {
1497 	struct pl08x_sg *dsg, *_dsg;
1498 
1499 	if (txd->llis_va)
1500 		dma_pool_free(pl08x->pool, txd->llis_va, txd->llis_bus);
1501 
1502 	list_for_each_entry_safe(dsg, _dsg, &txd->dsg_list, node) {
1503 		list_del(&dsg->node);
1504 		kfree(dsg);
1505 	}
1506 
1507 	kfree(txd);
1508 }
1509 
1510 static void pl08x_desc_free(struct virt_dma_desc *vd)
1511 {
1512 	struct pl08x_txd *txd = to_pl08x_txd(&vd->tx);
1513 	struct pl08x_dma_chan *plchan = to_pl08x_chan(vd->tx.chan);
1514 
1515 	dma_descriptor_unmap(&vd->tx);
1516 	if (!txd->done)
1517 		pl08x_release_mux(plchan);
1518 
1519 	pl08x_free_txd(plchan->host, txd);
1520 }
1521 
1522 static void pl08x_free_txd_list(struct pl08x_driver_data *pl08x,
1523 				struct pl08x_dma_chan *plchan)
1524 {
1525 	LIST_HEAD(head);
1526 
1527 	vchan_get_all_descriptors(&plchan->vc, &head);
1528 	vchan_dma_desc_free_list(&plchan->vc, &head);
1529 }
1530 
1531 /*
1532  * The DMA ENGINE API
1533  */
1534 static void pl08x_free_chan_resources(struct dma_chan *chan)
1535 {
1536 	/* Ensure all queued descriptors are freed */
1537 	vchan_free_chan_resources(to_virt_chan(chan));
1538 }
1539 
1540 static struct dma_async_tx_descriptor *pl08x_prep_dma_interrupt(
1541 		struct dma_chan *chan, unsigned long flags)
1542 {
1543 	struct dma_async_tx_descriptor *retval = NULL;
1544 
1545 	return retval;
1546 }
1547 
1548 /*
1549  * Code accessing dma_async_is_complete() in a tight loop may give problems.
1550  * If slaves are relying on interrupts to signal completion this function
1551  * must not be called with interrupts disabled.
1552  */
1553 static enum dma_status pl08x_dma_tx_status(struct dma_chan *chan,
1554 		dma_cookie_t cookie, struct dma_tx_state *txstate)
1555 {
1556 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1557 	struct virt_dma_desc *vd;
1558 	unsigned long flags;
1559 	enum dma_status ret;
1560 	size_t bytes = 0;
1561 
1562 	ret = dma_cookie_status(chan, cookie, txstate);
1563 	if (ret == DMA_COMPLETE)
1564 		return ret;
1565 
1566 	/*
1567 	 * There's no point calculating the residue if there's
1568 	 * no txstate to store the value.
1569 	 */
1570 	if (!txstate) {
1571 		if (plchan->state == PL08X_CHAN_PAUSED)
1572 			ret = DMA_PAUSED;
1573 		return ret;
1574 	}
1575 
1576 	spin_lock_irqsave(&plchan->vc.lock, flags);
1577 	ret = dma_cookie_status(chan, cookie, txstate);
1578 	if (ret != DMA_COMPLETE) {
1579 		vd = vchan_find_desc(&plchan->vc, cookie);
1580 		if (vd) {
1581 			/* On the issued list, so hasn't been processed yet */
1582 			struct pl08x_txd *txd = to_pl08x_txd(&vd->tx);
1583 			struct pl08x_sg *dsg;
1584 
1585 			list_for_each_entry(dsg, &txd->dsg_list, node)
1586 				bytes += dsg->len;
1587 		} else {
1588 			bytes = pl08x_getbytes_chan(plchan);
1589 		}
1590 	}
1591 	spin_unlock_irqrestore(&plchan->vc.lock, flags);
1592 
1593 	/*
1594 	 * This cookie not complete yet
1595 	 * Get number of bytes left in the active transactions and queue
1596 	 */
1597 	dma_set_residue(txstate, bytes);
1598 
1599 	if (plchan->state == PL08X_CHAN_PAUSED && ret == DMA_IN_PROGRESS)
1600 		ret = DMA_PAUSED;
1601 
1602 	/* Whether waiting or running, we're in progress */
1603 	return ret;
1604 }
1605 
1606 /* PrimeCell DMA extension */
1607 struct burst_table {
1608 	u32 burstwords;
1609 	u32 reg;
1610 };
1611 
1612 static const struct burst_table burst_sizes[] = {
1613 	{
1614 		.burstwords = 256,
1615 		.reg = PL080_BSIZE_256,
1616 	},
1617 	{
1618 		.burstwords = 128,
1619 		.reg = PL080_BSIZE_128,
1620 	},
1621 	{
1622 		.burstwords = 64,
1623 		.reg = PL080_BSIZE_64,
1624 	},
1625 	{
1626 		.burstwords = 32,
1627 		.reg = PL080_BSIZE_32,
1628 	},
1629 	{
1630 		.burstwords = 16,
1631 		.reg = PL080_BSIZE_16,
1632 	},
1633 	{
1634 		.burstwords = 8,
1635 		.reg = PL080_BSIZE_8,
1636 	},
1637 	{
1638 		.burstwords = 4,
1639 		.reg = PL080_BSIZE_4,
1640 	},
1641 	{
1642 		.burstwords = 0,
1643 		.reg = PL080_BSIZE_1,
1644 	},
1645 };
1646 
1647 /*
1648  * Given the source and destination available bus masks, select which
1649  * will be routed to each port.  We try to have source and destination
1650  * on separate ports, but always respect the allowable settings.
1651  */
1652 static u32 pl08x_select_bus(bool ftdmac020, u8 src, u8 dst)
1653 {
1654 	u32 cctl = 0;
1655 	u32 dst_ahb2;
1656 	u32 src_ahb2;
1657 
1658 	/* The FTDMAC020 use different bits to indicate src/dst bus */
1659 	if (ftdmac020) {
1660 		dst_ahb2 = FTDMAC020_LLI_DST_SEL;
1661 		src_ahb2 = FTDMAC020_LLI_SRC_SEL;
1662 	} else {
1663 		dst_ahb2 = PL080_CONTROL_DST_AHB2;
1664 		src_ahb2 = PL080_CONTROL_SRC_AHB2;
1665 	}
1666 
1667 	if (!(dst & PL08X_AHB1) || ((dst & PL08X_AHB2) && (src & PL08X_AHB1)))
1668 		cctl |= dst_ahb2;
1669 	if (!(src & PL08X_AHB1) || ((src & PL08X_AHB2) && !(dst & PL08X_AHB2)))
1670 		cctl |= src_ahb2;
1671 
1672 	return cctl;
1673 }
1674 
1675 static u32 pl08x_cctl(u32 cctl)
1676 {
1677 	cctl &= ~(PL080_CONTROL_SRC_AHB2 | PL080_CONTROL_DST_AHB2 |
1678 		  PL080_CONTROL_SRC_INCR | PL080_CONTROL_DST_INCR |
1679 		  PL080_CONTROL_PROT_MASK);
1680 
1681 	/* Access the cell in privileged mode, non-bufferable, non-cacheable */
1682 	return cctl | PL080_CONTROL_PROT_SYS;
1683 }
1684 
1685 static u32 pl08x_width(enum dma_slave_buswidth width)
1686 {
1687 	switch (width) {
1688 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
1689 		return PL080_WIDTH_8BIT;
1690 	case DMA_SLAVE_BUSWIDTH_2_BYTES:
1691 		return PL080_WIDTH_16BIT;
1692 	case DMA_SLAVE_BUSWIDTH_4_BYTES:
1693 		return PL080_WIDTH_32BIT;
1694 	default:
1695 		return ~0;
1696 	}
1697 }
1698 
1699 static u32 pl08x_burst(u32 maxburst)
1700 {
1701 	int i;
1702 
1703 	for (i = 0; i < ARRAY_SIZE(burst_sizes); i++)
1704 		if (burst_sizes[i].burstwords <= maxburst)
1705 			break;
1706 
1707 	return burst_sizes[i].reg;
1708 }
1709 
1710 static u32 pl08x_get_cctl(struct pl08x_dma_chan *plchan,
1711 	enum dma_slave_buswidth addr_width, u32 maxburst)
1712 {
1713 	u32 width, burst, cctl = 0;
1714 
1715 	width = pl08x_width(addr_width);
1716 	if (width == ~0)
1717 		return ~0;
1718 
1719 	cctl |= width << PL080_CONTROL_SWIDTH_SHIFT;
1720 	cctl |= width << PL080_CONTROL_DWIDTH_SHIFT;
1721 
1722 	/*
1723 	 * If this channel will only request single transfers, set this
1724 	 * down to ONE element.  Also select one element if no maxburst
1725 	 * is specified.
1726 	 */
1727 	if (plchan->cd->single)
1728 		maxburst = 1;
1729 
1730 	burst = pl08x_burst(maxburst);
1731 	cctl |= burst << PL080_CONTROL_SB_SIZE_SHIFT;
1732 	cctl |= burst << PL080_CONTROL_DB_SIZE_SHIFT;
1733 
1734 	return pl08x_cctl(cctl);
1735 }
1736 
1737 /*
1738  * Slave transactions callback to the slave device to allow
1739  * synchronization of slave DMA signals with the DMAC enable
1740  */
1741 static void pl08x_issue_pending(struct dma_chan *chan)
1742 {
1743 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1744 	unsigned long flags;
1745 
1746 	spin_lock_irqsave(&plchan->vc.lock, flags);
1747 	if (vchan_issue_pending(&plchan->vc)) {
1748 		if (!plchan->phychan && plchan->state != PL08X_CHAN_WAITING)
1749 			pl08x_phy_alloc_and_start(plchan);
1750 	}
1751 	spin_unlock_irqrestore(&plchan->vc.lock, flags);
1752 }
1753 
1754 static struct pl08x_txd *pl08x_get_txd(struct pl08x_dma_chan *plchan)
1755 {
1756 	struct pl08x_txd *txd = kzalloc(sizeof(*txd), GFP_NOWAIT);
1757 
1758 	if (txd)
1759 		INIT_LIST_HEAD(&txd->dsg_list);
1760 	return txd;
1761 }
1762 
1763 static u32 pl08x_memcpy_cctl(struct pl08x_driver_data *pl08x)
1764 {
1765 	u32 cctl = 0;
1766 
1767 	/* Conjure cctl */
1768 	switch (pl08x->pd->memcpy_burst_size) {
1769 	default:
1770 		dev_err(&pl08x->adev->dev,
1771 			"illegal burst size for memcpy, set to 1\n");
1772 		/* Fall through */
1773 	case PL08X_BURST_SZ_1:
1774 		cctl |= PL080_BSIZE_1 << PL080_CONTROL_SB_SIZE_SHIFT |
1775 			PL080_BSIZE_1 << PL080_CONTROL_DB_SIZE_SHIFT;
1776 		break;
1777 	case PL08X_BURST_SZ_4:
1778 		cctl |= PL080_BSIZE_4 << PL080_CONTROL_SB_SIZE_SHIFT |
1779 			PL080_BSIZE_4 << PL080_CONTROL_DB_SIZE_SHIFT;
1780 		break;
1781 	case PL08X_BURST_SZ_8:
1782 		cctl |= PL080_BSIZE_8 << PL080_CONTROL_SB_SIZE_SHIFT |
1783 			PL080_BSIZE_8 << PL080_CONTROL_DB_SIZE_SHIFT;
1784 		break;
1785 	case PL08X_BURST_SZ_16:
1786 		cctl |= PL080_BSIZE_16 << PL080_CONTROL_SB_SIZE_SHIFT |
1787 			PL080_BSIZE_16 << PL080_CONTROL_DB_SIZE_SHIFT;
1788 		break;
1789 	case PL08X_BURST_SZ_32:
1790 		cctl |= PL080_BSIZE_32 << PL080_CONTROL_SB_SIZE_SHIFT |
1791 			PL080_BSIZE_32 << PL080_CONTROL_DB_SIZE_SHIFT;
1792 		break;
1793 	case PL08X_BURST_SZ_64:
1794 		cctl |= PL080_BSIZE_64 << PL080_CONTROL_SB_SIZE_SHIFT |
1795 			PL080_BSIZE_64 << PL080_CONTROL_DB_SIZE_SHIFT;
1796 		break;
1797 	case PL08X_BURST_SZ_128:
1798 		cctl |= PL080_BSIZE_128 << PL080_CONTROL_SB_SIZE_SHIFT |
1799 			PL080_BSIZE_128 << PL080_CONTROL_DB_SIZE_SHIFT;
1800 		break;
1801 	case PL08X_BURST_SZ_256:
1802 		cctl |= PL080_BSIZE_256 << PL080_CONTROL_SB_SIZE_SHIFT |
1803 			PL080_BSIZE_256 << PL080_CONTROL_DB_SIZE_SHIFT;
1804 		break;
1805 	}
1806 
1807 	switch (pl08x->pd->memcpy_bus_width) {
1808 	default:
1809 		dev_err(&pl08x->adev->dev,
1810 			"illegal bus width for memcpy, set to 8 bits\n");
1811 		/* Fall through */
1812 	case PL08X_BUS_WIDTH_8_BITS:
1813 		cctl |= PL080_WIDTH_8BIT << PL080_CONTROL_SWIDTH_SHIFT |
1814 			PL080_WIDTH_8BIT << PL080_CONTROL_DWIDTH_SHIFT;
1815 		break;
1816 	case PL08X_BUS_WIDTH_16_BITS:
1817 		cctl |= PL080_WIDTH_16BIT << PL080_CONTROL_SWIDTH_SHIFT |
1818 			PL080_WIDTH_16BIT << PL080_CONTROL_DWIDTH_SHIFT;
1819 		break;
1820 	case PL08X_BUS_WIDTH_32_BITS:
1821 		cctl |= PL080_WIDTH_32BIT << PL080_CONTROL_SWIDTH_SHIFT |
1822 			PL080_WIDTH_32BIT << PL080_CONTROL_DWIDTH_SHIFT;
1823 		break;
1824 	}
1825 
1826 	/* Protection flags */
1827 	if (pl08x->pd->memcpy_prot_buff)
1828 		cctl |= PL080_CONTROL_PROT_BUFF;
1829 	if (pl08x->pd->memcpy_prot_cache)
1830 		cctl |= PL080_CONTROL_PROT_CACHE;
1831 
1832 	/* We are the kernel, so we are in privileged mode */
1833 	cctl |= PL080_CONTROL_PROT_SYS;
1834 
1835 	/* Both to be incremented or the code will break */
1836 	cctl |= PL080_CONTROL_SRC_INCR | PL080_CONTROL_DST_INCR;
1837 
1838 	if (pl08x->vd->dualmaster)
1839 		cctl |= pl08x_select_bus(false,
1840 					 pl08x->mem_buses,
1841 					 pl08x->mem_buses);
1842 
1843 	return cctl;
1844 }
1845 
1846 static u32 pl08x_ftdmac020_memcpy_cctl(struct pl08x_driver_data *pl08x)
1847 {
1848 	u32 cctl = 0;
1849 
1850 	/* Conjure cctl */
1851 	switch (pl08x->pd->memcpy_bus_width) {
1852 	default:
1853 		dev_err(&pl08x->adev->dev,
1854 			"illegal bus width for memcpy, set to 8 bits\n");
1855 		/* Fall through */
1856 	case PL08X_BUS_WIDTH_8_BITS:
1857 		cctl |= PL080_WIDTH_8BIT << FTDMAC020_LLI_SRC_WIDTH_SHIFT |
1858 			PL080_WIDTH_8BIT << FTDMAC020_LLI_DST_WIDTH_SHIFT;
1859 		break;
1860 	case PL08X_BUS_WIDTH_16_BITS:
1861 		cctl |= PL080_WIDTH_16BIT << FTDMAC020_LLI_SRC_WIDTH_SHIFT |
1862 			PL080_WIDTH_16BIT << FTDMAC020_LLI_DST_WIDTH_SHIFT;
1863 		break;
1864 	case PL08X_BUS_WIDTH_32_BITS:
1865 		cctl |= PL080_WIDTH_32BIT << FTDMAC020_LLI_SRC_WIDTH_SHIFT |
1866 			PL080_WIDTH_32BIT << FTDMAC020_LLI_DST_WIDTH_SHIFT;
1867 		break;
1868 	}
1869 
1870 	/*
1871 	 * By default mask the TC IRQ on all LLIs, it will be unmasked on
1872 	 * the last LLI item by other code.
1873 	 */
1874 	cctl |= FTDMAC020_LLI_TC_MSK;
1875 
1876 	/*
1877 	 * Both to be incremented so leave bits FTDMAC020_LLI_SRCAD_CTL
1878 	 * and FTDMAC020_LLI_DSTAD_CTL as zero
1879 	 */
1880 	if (pl08x->vd->dualmaster)
1881 		cctl |= pl08x_select_bus(true,
1882 					 pl08x->mem_buses,
1883 					 pl08x->mem_buses);
1884 
1885 	return cctl;
1886 }
1887 
1888 /*
1889  * Initialize a descriptor to be used by memcpy submit
1890  */
1891 static struct dma_async_tx_descriptor *pl08x_prep_dma_memcpy(
1892 		struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
1893 		size_t len, unsigned long flags)
1894 {
1895 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1896 	struct pl08x_driver_data *pl08x = plchan->host;
1897 	struct pl08x_txd *txd;
1898 	struct pl08x_sg *dsg;
1899 	int ret;
1900 
1901 	txd = pl08x_get_txd(plchan);
1902 	if (!txd) {
1903 		dev_err(&pl08x->adev->dev,
1904 			"%s no memory for descriptor\n", __func__);
1905 		return NULL;
1906 	}
1907 
1908 	dsg = kzalloc(sizeof(struct pl08x_sg), GFP_NOWAIT);
1909 	if (!dsg) {
1910 		pl08x_free_txd(pl08x, txd);
1911 		return NULL;
1912 	}
1913 	list_add_tail(&dsg->node, &txd->dsg_list);
1914 
1915 	dsg->src_addr = src;
1916 	dsg->dst_addr = dest;
1917 	dsg->len = len;
1918 	if (pl08x->vd->ftdmac020) {
1919 		/* Writing CCFG zero ENABLES all interrupts */
1920 		txd->ccfg = 0;
1921 		txd->cctl = pl08x_ftdmac020_memcpy_cctl(pl08x);
1922 	} else {
1923 		txd->ccfg = PL080_CONFIG_ERR_IRQ_MASK |
1924 			PL080_CONFIG_TC_IRQ_MASK |
1925 			PL080_FLOW_MEM2MEM << PL080_CONFIG_FLOW_CONTROL_SHIFT;
1926 		txd->cctl = pl08x_memcpy_cctl(pl08x);
1927 	}
1928 
1929 	ret = pl08x_fill_llis_for_desc(plchan->host, txd);
1930 	if (!ret) {
1931 		pl08x_free_txd(pl08x, txd);
1932 		return NULL;
1933 	}
1934 
1935 	return vchan_tx_prep(&plchan->vc, &txd->vd, flags);
1936 }
1937 
1938 static struct pl08x_txd *pl08x_init_txd(
1939 		struct dma_chan *chan,
1940 		enum dma_transfer_direction direction,
1941 		dma_addr_t *slave_addr)
1942 {
1943 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1944 	struct pl08x_driver_data *pl08x = plchan->host;
1945 	struct pl08x_txd *txd;
1946 	enum dma_slave_buswidth addr_width;
1947 	int ret, tmp;
1948 	u8 src_buses, dst_buses;
1949 	u32 maxburst, cctl;
1950 
1951 	txd = pl08x_get_txd(plchan);
1952 	if (!txd) {
1953 		dev_err(&pl08x->adev->dev, "%s no txd\n", __func__);
1954 		return NULL;
1955 	}
1956 
1957 	/*
1958 	 * Set up addresses, the PrimeCell configured address
1959 	 * will take precedence since this may configure the
1960 	 * channel target address dynamically at runtime.
1961 	 */
1962 	if (direction == DMA_MEM_TO_DEV) {
1963 		cctl = PL080_CONTROL_SRC_INCR;
1964 		*slave_addr = plchan->cfg.dst_addr;
1965 		addr_width = plchan->cfg.dst_addr_width;
1966 		maxburst = plchan->cfg.dst_maxburst;
1967 		src_buses = pl08x->mem_buses;
1968 		dst_buses = plchan->cd->periph_buses;
1969 	} else if (direction == DMA_DEV_TO_MEM) {
1970 		cctl = PL080_CONTROL_DST_INCR;
1971 		*slave_addr = plchan->cfg.src_addr;
1972 		addr_width = plchan->cfg.src_addr_width;
1973 		maxburst = plchan->cfg.src_maxburst;
1974 		src_buses = plchan->cd->periph_buses;
1975 		dst_buses = pl08x->mem_buses;
1976 	} else {
1977 		pl08x_free_txd(pl08x, txd);
1978 		dev_err(&pl08x->adev->dev,
1979 			"%s direction unsupported\n", __func__);
1980 		return NULL;
1981 	}
1982 
1983 	cctl |= pl08x_get_cctl(plchan, addr_width, maxburst);
1984 	if (cctl == ~0) {
1985 		pl08x_free_txd(pl08x, txd);
1986 		dev_err(&pl08x->adev->dev,
1987 			"DMA slave configuration botched?\n");
1988 		return NULL;
1989 	}
1990 
1991 	txd->cctl = cctl | pl08x_select_bus(false, src_buses, dst_buses);
1992 
1993 	if (plchan->cfg.device_fc)
1994 		tmp = (direction == DMA_MEM_TO_DEV) ? PL080_FLOW_MEM2PER_PER :
1995 			PL080_FLOW_PER2MEM_PER;
1996 	else
1997 		tmp = (direction == DMA_MEM_TO_DEV) ? PL080_FLOW_MEM2PER :
1998 			PL080_FLOW_PER2MEM;
1999 
2000 	txd->ccfg = PL080_CONFIG_ERR_IRQ_MASK |
2001 		PL080_CONFIG_TC_IRQ_MASK |
2002 		tmp << PL080_CONFIG_FLOW_CONTROL_SHIFT;
2003 
2004 	ret = pl08x_request_mux(plchan);
2005 	if (ret < 0) {
2006 		pl08x_free_txd(pl08x, txd);
2007 		dev_dbg(&pl08x->adev->dev,
2008 			"unable to mux for transfer on %s due to platform restrictions\n",
2009 			plchan->name);
2010 		return NULL;
2011 	}
2012 
2013 	dev_dbg(&pl08x->adev->dev, "allocated DMA request signal %d for xfer on %s\n",
2014 		 plchan->signal, plchan->name);
2015 
2016 	/* Assign the flow control signal to this channel */
2017 	if (direction == DMA_MEM_TO_DEV)
2018 		txd->ccfg |= plchan->signal << PL080_CONFIG_DST_SEL_SHIFT;
2019 	else
2020 		txd->ccfg |= plchan->signal << PL080_CONFIG_SRC_SEL_SHIFT;
2021 
2022 	return txd;
2023 }
2024 
2025 static int pl08x_tx_add_sg(struct pl08x_txd *txd,
2026 			   enum dma_transfer_direction direction,
2027 			   dma_addr_t slave_addr,
2028 			   dma_addr_t buf_addr,
2029 			   unsigned int len)
2030 {
2031 	struct pl08x_sg *dsg;
2032 
2033 	dsg = kzalloc(sizeof(struct pl08x_sg), GFP_NOWAIT);
2034 	if (!dsg)
2035 		return -ENOMEM;
2036 
2037 	list_add_tail(&dsg->node, &txd->dsg_list);
2038 
2039 	dsg->len = len;
2040 	if (direction == DMA_MEM_TO_DEV) {
2041 		dsg->src_addr = buf_addr;
2042 		dsg->dst_addr = slave_addr;
2043 	} else {
2044 		dsg->src_addr = slave_addr;
2045 		dsg->dst_addr = buf_addr;
2046 	}
2047 
2048 	return 0;
2049 }
2050 
2051 static struct dma_async_tx_descriptor *pl08x_prep_slave_sg(
2052 		struct dma_chan *chan, struct scatterlist *sgl,
2053 		unsigned int sg_len, enum dma_transfer_direction direction,
2054 		unsigned long flags, void *context)
2055 {
2056 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
2057 	struct pl08x_driver_data *pl08x = plchan->host;
2058 	struct pl08x_txd *txd;
2059 	struct scatterlist *sg;
2060 	int ret, tmp;
2061 	dma_addr_t slave_addr;
2062 
2063 	dev_dbg(&pl08x->adev->dev, "%s prepare transaction of %d bytes from %s\n",
2064 			__func__, sg_dma_len(sgl), plchan->name);
2065 
2066 	txd = pl08x_init_txd(chan, direction, &slave_addr);
2067 	if (!txd)
2068 		return NULL;
2069 
2070 	for_each_sg(sgl, sg, sg_len, tmp) {
2071 		ret = pl08x_tx_add_sg(txd, direction, slave_addr,
2072 				      sg_dma_address(sg),
2073 				      sg_dma_len(sg));
2074 		if (ret) {
2075 			pl08x_release_mux(plchan);
2076 			pl08x_free_txd(pl08x, txd);
2077 			dev_err(&pl08x->adev->dev, "%s no mem for pl080 sg\n",
2078 					__func__);
2079 			return NULL;
2080 		}
2081 	}
2082 
2083 	ret = pl08x_fill_llis_for_desc(plchan->host, txd);
2084 	if (!ret) {
2085 		pl08x_release_mux(plchan);
2086 		pl08x_free_txd(pl08x, txd);
2087 		return NULL;
2088 	}
2089 
2090 	return vchan_tx_prep(&plchan->vc, &txd->vd, flags);
2091 }
2092 
2093 static struct dma_async_tx_descriptor *pl08x_prep_dma_cyclic(
2094 		struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len,
2095 		size_t period_len, enum dma_transfer_direction direction,
2096 		unsigned long flags)
2097 {
2098 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
2099 	struct pl08x_driver_data *pl08x = plchan->host;
2100 	struct pl08x_txd *txd;
2101 	int ret, tmp;
2102 	dma_addr_t slave_addr;
2103 
2104 	dev_dbg(&pl08x->adev->dev,
2105 		"%s prepare cyclic transaction of %zd/%zd bytes %s %s\n",
2106 		__func__, period_len, buf_len,
2107 		direction == DMA_MEM_TO_DEV ? "to" : "from",
2108 		plchan->name);
2109 
2110 	txd = pl08x_init_txd(chan, direction, &slave_addr);
2111 	if (!txd)
2112 		return NULL;
2113 
2114 	txd->cyclic = true;
2115 	txd->cctl |= PL080_CONTROL_TC_IRQ_EN;
2116 	for (tmp = 0; tmp < buf_len; tmp += period_len) {
2117 		ret = pl08x_tx_add_sg(txd, direction, slave_addr,
2118 				      buf_addr + tmp, period_len);
2119 		if (ret) {
2120 			pl08x_release_mux(plchan);
2121 			pl08x_free_txd(pl08x, txd);
2122 			return NULL;
2123 		}
2124 	}
2125 
2126 	ret = pl08x_fill_llis_for_desc(plchan->host, txd);
2127 	if (!ret) {
2128 		pl08x_release_mux(plchan);
2129 		pl08x_free_txd(pl08x, txd);
2130 		return NULL;
2131 	}
2132 
2133 	return vchan_tx_prep(&plchan->vc, &txd->vd, flags);
2134 }
2135 
2136 static int pl08x_config(struct dma_chan *chan,
2137 			struct dma_slave_config *config)
2138 {
2139 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
2140 	struct pl08x_driver_data *pl08x = plchan->host;
2141 
2142 	if (!plchan->slave)
2143 		return -EINVAL;
2144 
2145 	/* Reject definitely invalid configurations */
2146 	if (config->src_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES ||
2147 	    config->dst_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES)
2148 		return -EINVAL;
2149 
2150 	if (config->device_fc && pl08x->vd->pl080s) {
2151 		dev_err(&pl08x->adev->dev,
2152 			"%s: PL080S does not support peripheral flow control\n",
2153 			__func__);
2154 		return -EINVAL;
2155 	}
2156 
2157 	plchan->cfg = *config;
2158 
2159 	return 0;
2160 }
2161 
2162 static int pl08x_terminate_all(struct dma_chan *chan)
2163 {
2164 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
2165 	struct pl08x_driver_data *pl08x = plchan->host;
2166 	unsigned long flags;
2167 
2168 	spin_lock_irqsave(&plchan->vc.lock, flags);
2169 	if (!plchan->phychan && !plchan->at) {
2170 		spin_unlock_irqrestore(&plchan->vc.lock, flags);
2171 		return 0;
2172 	}
2173 
2174 	plchan->state = PL08X_CHAN_IDLE;
2175 
2176 	if (plchan->phychan) {
2177 		/*
2178 		 * Mark physical channel as free and free any slave
2179 		 * signal
2180 		 */
2181 		pl08x_phy_free(plchan);
2182 	}
2183 	/* Dequeue jobs and free LLIs */
2184 	if (plchan->at) {
2185 		vchan_terminate_vdesc(&plchan->at->vd);
2186 		plchan->at = NULL;
2187 	}
2188 	/* Dequeue jobs not yet fired as well */
2189 	pl08x_free_txd_list(pl08x, plchan);
2190 
2191 	spin_unlock_irqrestore(&plchan->vc.lock, flags);
2192 
2193 	return 0;
2194 }
2195 
2196 static void pl08x_synchronize(struct dma_chan *chan)
2197 {
2198 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
2199 
2200 	vchan_synchronize(&plchan->vc);
2201 }
2202 
2203 static int pl08x_pause(struct dma_chan *chan)
2204 {
2205 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
2206 	unsigned long flags;
2207 
2208 	/*
2209 	 * Anything succeeds on channels with no physical allocation and
2210 	 * no queued transfers.
2211 	 */
2212 	spin_lock_irqsave(&plchan->vc.lock, flags);
2213 	if (!plchan->phychan && !plchan->at) {
2214 		spin_unlock_irqrestore(&plchan->vc.lock, flags);
2215 		return 0;
2216 	}
2217 
2218 	pl08x_pause_phy_chan(plchan->phychan);
2219 	plchan->state = PL08X_CHAN_PAUSED;
2220 
2221 	spin_unlock_irqrestore(&plchan->vc.lock, flags);
2222 
2223 	return 0;
2224 }
2225 
2226 static int pl08x_resume(struct dma_chan *chan)
2227 {
2228 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
2229 	unsigned long flags;
2230 
2231 	/*
2232 	 * Anything succeeds on channels with no physical allocation and
2233 	 * no queued transfers.
2234 	 */
2235 	spin_lock_irqsave(&plchan->vc.lock, flags);
2236 	if (!plchan->phychan && !plchan->at) {
2237 		spin_unlock_irqrestore(&plchan->vc.lock, flags);
2238 		return 0;
2239 	}
2240 
2241 	pl08x_resume_phy_chan(plchan->phychan);
2242 	plchan->state = PL08X_CHAN_RUNNING;
2243 
2244 	spin_unlock_irqrestore(&plchan->vc.lock, flags);
2245 
2246 	return 0;
2247 }
2248 
2249 bool pl08x_filter_id(struct dma_chan *chan, void *chan_id)
2250 {
2251 	struct pl08x_dma_chan *plchan;
2252 	char *name = chan_id;
2253 
2254 	/* Reject channels for devices not bound to this driver */
2255 	if (chan->device->dev->driver != &pl08x_amba_driver.drv)
2256 		return false;
2257 
2258 	plchan = to_pl08x_chan(chan);
2259 
2260 	/* Check that the channel is not taken! */
2261 	if (!strcmp(plchan->name, name))
2262 		return true;
2263 
2264 	return false;
2265 }
2266 EXPORT_SYMBOL_GPL(pl08x_filter_id);
2267 
2268 static bool pl08x_filter_fn(struct dma_chan *chan, void *chan_id)
2269 {
2270 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
2271 
2272 	return plchan->cd == chan_id;
2273 }
2274 
2275 /*
2276  * Just check that the device is there and active
2277  * TODO: turn this bit on/off depending on the number of physical channels
2278  * actually used, if it is zero... well shut it off. That will save some
2279  * power. Cut the clock at the same time.
2280  */
2281 static void pl08x_ensure_on(struct pl08x_driver_data *pl08x)
2282 {
2283 	/* The Nomadik variant does not have the config register */
2284 	if (pl08x->vd->nomadik)
2285 		return;
2286 	/* The FTDMAC020 variant does this in another register */
2287 	if (pl08x->vd->ftdmac020) {
2288 		writel(PL080_CONFIG_ENABLE, pl08x->base + FTDMAC020_CSR);
2289 		return;
2290 	}
2291 	writel(PL080_CONFIG_ENABLE, pl08x->base + PL080_CONFIG);
2292 }
2293 
2294 static irqreturn_t pl08x_irq(int irq, void *dev)
2295 {
2296 	struct pl08x_driver_data *pl08x = dev;
2297 	u32 mask = 0, err, tc, i;
2298 
2299 	/* check & clear - ERR & TC interrupts */
2300 	err = readl(pl08x->base + PL080_ERR_STATUS);
2301 	if (err) {
2302 		dev_err(&pl08x->adev->dev, "%s error interrupt, register value 0x%08x\n",
2303 			__func__, err);
2304 		writel(err, pl08x->base + PL080_ERR_CLEAR);
2305 	}
2306 	tc = readl(pl08x->base + PL080_TC_STATUS);
2307 	if (tc)
2308 		writel(tc, pl08x->base + PL080_TC_CLEAR);
2309 
2310 	if (!err && !tc)
2311 		return IRQ_NONE;
2312 
2313 	for (i = 0; i < pl08x->vd->channels; i++) {
2314 		if ((BIT(i) & err) || (BIT(i) & tc)) {
2315 			/* Locate physical channel */
2316 			struct pl08x_phy_chan *phychan = &pl08x->phy_chans[i];
2317 			struct pl08x_dma_chan *plchan = phychan->serving;
2318 			struct pl08x_txd *tx;
2319 
2320 			if (!plchan) {
2321 				dev_err(&pl08x->adev->dev,
2322 					"%s Error TC interrupt on unused channel: 0x%08x\n",
2323 					__func__, i);
2324 				continue;
2325 			}
2326 
2327 			spin_lock(&plchan->vc.lock);
2328 			tx = plchan->at;
2329 			if (tx && tx->cyclic) {
2330 				vchan_cyclic_callback(&tx->vd);
2331 			} else if (tx) {
2332 				plchan->at = NULL;
2333 				/*
2334 				 * This descriptor is done, release its mux
2335 				 * reservation.
2336 				 */
2337 				pl08x_release_mux(plchan);
2338 				tx->done = true;
2339 				vchan_cookie_complete(&tx->vd);
2340 
2341 				/*
2342 				 * And start the next descriptor (if any),
2343 				 * otherwise free this channel.
2344 				 */
2345 				if (vchan_next_desc(&plchan->vc))
2346 					pl08x_start_next_txd(plchan);
2347 				else
2348 					pl08x_phy_free(plchan);
2349 			}
2350 			spin_unlock(&plchan->vc.lock);
2351 
2352 			mask |= BIT(i);
2353 		}
2354 	}
2355 
2356 	return mask ? IRQ_HANDLED : IRQ_NONE;
2357 }
2358 
2359 static void pl08x_dma_slave_init(struct pl08x_dma_chan *chan)
2360 {
2361 	chan->slave = true;
2362 	chan->name = chan->cd->bus_id;
2363 	chan->cfg.src_addr = chan->cd->addr;
2364 	chan->cfg.dst_addr = chan->cd->addr;
2365 }
2366 
2367 /*
2368  * Initialise the DMAC memcpy/slave channels.
2369  * Make a local wrapper to hold required data
2370  */
2371 static int pl08x_dma_init_virtual_channels(struct pl08x_driver_data *pl08x,
2372 		struct dma_device *dmadev, unsigned int channels, bool slave)
2373 {
2374 	struct pl08x_dma_chan *chan;
2375 	int i;
2376 
2377 	INIT_LIST_HEAD(&dmadev->channels);
2378 
2379 	/*
2380 	 * Register as many many memcpy as we have physical channels,
2381 	 * we won't always be able to use all but the code will have
2382 	 * to cope with that situation.
2383 	 */
2384 	for (i = 0; i < channels; i++) {
2385 		chan = kzalloc(sizeof(*chan), GFP_KERNEL);
2386 		if (!chan)
2387 			return -ENOMEM;
2388 
2389 		chan->host = pl08x;
2390 		chan->state = PL08X_CHAN_IDLE;
2391 		chan->signal = -1;
2392 
2393 		if (slave) {
2394 			chan->cd = &pl08x->pd->slave_channels[i];
2395 			/*
2396 			 * Some implementations have muxed signals, whereas some
2397 			 * use a mux in front of the signals and need dynamic
2398 			 * assignment of signals.
2399 			 */
2400 			chan->signal = i;
2401 			pl08x_dma_slave_init(chan);
2402 		} else {
2403 			chan->cd = kzalloc(sizeof(*chan->cd), GFP_KERNEL);
2404 			if (!chan->cd) {
2405 				kfree(chan);
2406 				return -ENOMEM;
2407 			}
2408 			chan->cd->bus_id = "memcpy";
2409 			chan->cd->periph_buses = pl08x->pd->mem_buses;
2410 			chan->name = kasprintf(GFP_KERNEL, "memcpy%d", i);
2411 			if (!chan->name) {
2412 				kfree(chan->cd);
2413 				kfree(chan);
2414 				return -ENOMEM;
2415 			}
2416 		}
2417 		dev_dbg(&pl08x->adev->dev,
2418 			 "initialize virtual channel \"%s\"\n",
2419 			 chan->name);
2420 
2421 		chan->vc.desc_free = pl08x_desc_free;
2422 		vchan_init(&chan->vc, dmadev);
2423 	}
2424 	dev_info(&pl08x->adev->dev, "initialized %d virtual %s channels\n",
2425 		 i, slave ? "slave" : "memcpy");
2426 	return i;
2427 }
2428 
2429 static void pl08x_free_virtual_channels(struct dma_device *dmadev)
2430 {
2431 	struct pl08x_dma_chan *chan = NULL;
2432 	struct pl08x_dma_chan *next;
2433 
2434 	list_for_each_entry_safe(chan,
2435 				 next, &dmadev->channels, vc.chan.device_node) {
2436 		list_del(&chan->vc.chan.device_node);
2437 		kfree(chan);
2438 	}
2439 }
2440 
2441 #ifdef CONFIG_DEBUG_FS
2442 static const char *pl08x_state_str(enum pl08x_dma_chan_state state)
2443 {
2444 	switch (state) {
2445 	case PL08X_CHAN_IDLE:
2446 		return "idle";
2447 	case PL08X_CHAN_RUNNING:
2448 		return "running";
2449 	case PL08X_CHAN_PAUSED:
2450 		return "paused";
2451 	case PL08X_CHAN_WAITING:
2452 		return "waiting";
2453 	default:
2454 		break;
2455 	}
2456 	return "UNKNOWN STATE";
2457 }
2458 
2459 static int pl08x_debugfs_show(struct seq_file *s, void *data)
2460 {
2461 	struct pl08x_driver_data *pl08x = s->private;
2462 	struct pl08x_dma_chan *chan;
2463 	struct pl08x_phy_chan *ch;
2464 	unsigned long flags;
2465 	int i;
2466 
2467 	seq_printf(s, "PL08x physical channels:\n");
2468 	seq_printf(s, "CHANNEL:\tUSER:\n");
2469 	seq_printf(s, "--------\t-----\n");
2470 	for (i = 0; i < pl08x->vd->channels; i++) {
2471 		struct pl08x_dma_chan *virt_chan;
2472 
2473 		ch = &pl08x->phy_chans[i];
2474 
2475 		spin_lock_irqsave(&ch->lock, flags);
2476 		virt_chan = ch->serving;
2477 
2478 		seq_printf(s, "%d\t\t%s%s\n",
2479 			   ch->id,
2480 			   virt_chan ? virt_chan->name : "(none)",
2481 			   ch->locked ? " LOCKED" : "");
2482 
2483 		spin_unlock_irqrestore(&ch->lock, flags);
2484 	}
2485 
2486 	seq_printf(s, "\nPL08x virtual memcpy channels:\n");
2487 	seq_printf(s, "CHANNEL:\tSTATE:\n");
2488 	seq_printf(s, "--------\t------\n");
2489 	list_for_each_entry(chan, &pl08x->memcpy.channels, vc.chan.device_node) {
2490 		seq_printf(s, "%s\t\t%s\n", chan->name,
2491 			   pl08x_state_str(chan->state));
2492 	}
2493 
2494 	if (pl08x->has_slave) {
2495 		seq_printf(s, "\nPL08x virtual slave channels:\n");
2496 		seq_printf(s, "CHANNEL:\tSTATE:\n");
2497 		seq_printf(s, "--------\t------\n");
2498 		list_for_each_entry(chan, &pl08x->slave.channels,
2499 				    vc.chan.device_node) {
2500 			seq_printf(s, "%s\t\t%s\n", chan->name,
2501 				   pl08x_state_str(chan->state));
2502 		}
2503 	}
2504 
2505 	return 0;
2506 }
2507 
2508 static int pl08x_debugfs_open(struct inode *inode, struct file *file)
2509 {
2510 	return single_open(file, pl08x_debugfs_show, inode->i_private);
2511 }
2512 
2513 static const struct file_operations pl08x_debugfs_operations = {
2514 	.open		= pl08x_debugfs_open,
2515 	.read		= seq_read,
2516 	.llseek		= seq_lseek,
2517 	.release	= single_release,
2518 };
2519 
2520 static void init_pl08x_debugfs(struct pl08x_driver_data *pl08x)
2521 {
2522 	/* Expose a simple debugfs interface to view all clocks */
2523 	(void) debugfs_create_file(dev_name(&pl08x->adev->dev),
2524 			S_IFREG | S_IRUGO, NULL, pl08x,
2525 			&pl08x_debugfs_operations);
2526 }
2527 
2528 #else
2529 static inline void init_pl08x_debugfs(struct pl08x_driver_data *pl08x)
2530 {
2531 }
2532 #endif
2533 
2534 #ifdef CONFIG_OF
2535 static struct dma_chan *pl08x_find_chan_id(struct pl08x_driver_data *pl08x,
2536 					 u32 id)
2537 {
2538 	struct pl08x_dma_chan *chan;
2539 
2540 	/* Trying to get a slave channel from something with no slave support */
2541 	if (!pl08x->has_slave)
2542 		return NULL;
2543 
2544 	list_for_each_entry(chan, &pl08x->slave.channels, vc.chan.device_node) {
2545 		if (chan->signal == id)
2546 			return &chan->vc.chan;
2547 	}
2548 
2549 	return NULL;
2550 }
2551 
2552 static struct dma_chan *pl08x_of_xlate(struct of_phandle_args *dma_spec,
2553 				       struct of_dma *ofdma)
2554 {
2555 	struct pl08x_driver_data *pl08x = ofdma->of_dma_data;
2556 	struct dma_chan *dma_chan;
2557 	struct pl08x_dma_chan *plchan;
2558 
2559 	if (!pl08x)
2560 		return NULL;
2561 
2562 	if (dma_spec->args_count != 2) {
2563 		dev_err(&pl08x->adev->dev,
2564 			"DMA channel translation requires two cells\n");
2565 		return NULL;
2566 	}
2567 
2568 	dma_chan = pl08x_find_chan_id(pl08x, dma_spec->args[0]);
2569 	if (!dma_chan) {
2570 		dev_err(&pl08x->adev->dev,
2571 			"DMA slave channel not found\n");
2572 		return NULL;
2573 	}
2574 
2575 	plchan = to_pl08x_chan(dma_chan);
2576 	dev_dbg(&pl08x->adev->dev,
2577 		"translated channel for signal %d\n",
2578 		dma_spec->args[0]);
2579 
2580 	/* Augment channel data for applicable AHB buses */
2581 	plchan->cd->periph_buses = dma_spec->args[1];
2582 	return dma_get_slave_channel(dma_chan);
2583 }
2584 
2585 static int pl08x_of_probe(struct amba_device *adev,
2586 			  struct pl08x_driver_data *pl08x,
2587 			  struct device_node *np)
2588 {
2589 	struct pl08x_platform_data *pd;
2590 	struct pl08x_channel_data *chanp = NULL;
2591 	u32 val;
2592 	int ret;
2593 	int i;
2594 
2595 	pd = devm_kzalloc(&adev->dev, sizeof(*pd), GFP_KERNEL);
2596 	if (!pd)
2597 		return -ENOMEM;
2598 
2599 	/* Eligible bus masters for fetching LLIs */
2600 	if (of_property_read_bool(np, "lli-bus-interface-ahb1"))
2601 		pd->lli_buses |= PL08X_AHB1;
2602 	if (of_property_read_bool(np, "lli-bus-interface-ahb2"))
2603 		pd->lli_buses |= PL08X_AHB2;
2604 	if (!pd->lli_buses) {
2605 		dev_info(&adev->dev, "no bus masters for LLIs stated, assume all\n");
2606 		pd->lli_buses |= PL08X_AHB1 | PL08X_AHB2;
2607 	}
2608 
2609 	/* Eligible bus masters for memory access */
2610 	if (of_property_read_bool(np, "mem-bus-interface-ahb1"))
2611 		pd->mem_buses |= PL08X_AHB1;
2612 	if (of_property_read_bool(np, "mem-bus-interface-ahb2"))
2613 		pd->mem_buses |= PL08X_AHB2;
2614 	if (!pd->mem_buses) {
2615 		dev_info(&adev->dev, "no bus masters for memory stated, assume all\n");
2616 		pd->mem_buses |= PL08X_AHB1 | PL08X_AHB2;
2617 	}
2618 
2619 	/* Parse the memcpy channel properties */
2620 	ret = of_property_read_u32(np, "memcpy-burst-size", &val);
2621 	if (ret) {
2622 		dev_info(&adev->dev, "no memcpy burst size specified, using 1 byte\n");
2623 		val = 1;
2624 	}
2625 	switch (val) {
2626 	default:
2627 		dev_err(&adev->dev, "illegal burst size for memcpy, set to 1\n");
2628 		/* Fall through */
2629 	case 1:
2630 		pd->memcpy_burst_size = PL08X_BURST_SZ_1;
2631 		break;
2632 	case 4:
2633 		pd->memcpy_burst_size = PL08X_BURST_SZ_4;
2634 		break;
2635 	case 8:
2636 		pd->memcpy_burst_size = PL08X_BURST_SZ_8;
2637 		break;
2638 	case 16:
2639 		pd->memcpy_burst_size = PL08X_BURST_SZ_16;
2640 		break;
2641 	case 32:
2642 		pd->memcpy_burst_size = PL08X_BURST_SZ_32;
2643 		break;
2644 	case 64:
2645 		pd->memcpy_burst_size = PL08X_BURST_SZ_64;
2646 		break;
2647 	case 128:
2648 		pd->memcpy_burst_size = PL08X_BURST_SZ_128;
2649 		break;
2650 	case 256:
2651 		pd->memcpy_burst_size = PL08X_BURST_SZ_256;
2652 		break;
2653 	}
2654 
2655 	ret = of_property_read_u32(np, "memcpy-bus-width", &val);
2656 	if (ret) {
2657 		dev_info(&adev->dev, "no memcpy bus width specified, using 8 bits\n");
2658 		val = 8;
2659 	}
2660 	switch (val) {
2661 	default:
2662 		dev_err(&adev->dev, "illegal bus width for memcpy, set to 8 bits\n");
2663 		/* Fall through */
2664 	case 8:
2665 		pd->memcpy_bus_width = PL08X_BUS_WIDTH_8_BITS;
2666 		break;
2667 	case 16:
2668 		pd->memcpy_bus_width = PL08X_BUS_WIDTH_16_BITS;
2669 		break;
2670 	case 32:
2671 		pd->memcpy_bus_width = PL08X_BUS_WIDTH_32_BITS;
2672 		break;
2673 	}
2674 
2675 	/*
2676 	 * Allocate channel data for all possible slave channels (one
2677 	 * for each possible signal), channels will then be allocated
2678 	 * for a device and have it's AHB interfaces set up at
2679 	 * translation time.
2680 	 */
2681 	if (pl08x->vd->signals) {
2682 		chanp = devm_kcalloc(&adev->dev,
2683 				     pl08x->vd->signals,
2684 				     sizeof(struct pl08x_channel_data),
2685 				     GFP_KERNEL);
2686 		if (!chanp)
2687 			return -ENOMEM;
2688 
2689 		pd->slave_channels = chanp;
2690 		for (i = 0; i < pl08x->vd->signals; i++) {
2691 			/*
2692 			 * chanp->periph_buses will be assigned at translation
2693 			 */
2694 			chanp->bus_id = kasprintf(GFP_KERNEL, "slave%d", i);
2695 			chanp++;
2696 		}
2697 		pd->num_slave_channels = pl08x->vd->signals;
2698 	}
2699 
2700 	pl08x->pd = pd;
2701 
2702 	return of_dma_controller_register(adev->dev.of_node, pl08x_of_xlate,
2703 					  pl08x);
2704 }
2705 #else
2706 static inline int pl08x_of_probe(struct amba_device *adev,
2707 				 struct pl08x_driver_data *pl08x,
2708 				 struct device_node *np)
2709 {
2710 	return -EINVAL;
2711 }
2712 #endif
2713 
2714 static int pl08x_probe(struct amba_device *adev, const struct amba_id *id)
2715 {
2716 	struct pl08x_driver_data *pl08x;
2717 	struct vendor_data *vd = id->data;
2718 	struct device_node *np = adev->dev.of_node;
2719 	u32 tsfr_size;
2720 	int ret = 0;
2721 	int i;
2722 
2723 	ret = amba_request_regions(adev, NULL);
2724 	if (ret)
2725 		return ret;
2726 
2727 	/* Ensure that we can do DMA */
2728 	ret = dma_set_mask_and_coherent(&adev->dev, DMA_BIT_MASK(32));
2729 	if (ret)
2730 		goto out_no_pl08x;
2731 
2732 	/* Create the driver state holder */
2733 	pl08x = kzalloc(sizeof(*pl08x), GFP_KERNEL);
2734 	if (!pl08x) {
2735 		ret = -ENOMEM;
2736 		goto out_no_pl08x;
2737 	}
2738 
2739 	/* Assign useful pointers to the driver state */
2740 	pl08x->adev = adev;
2741 	pl08x->vd = vd;
2742 
2743 	pl08x->base = ioremap(adev->res.start, resource_size(&adev->res));
2744 	if (!pl08x->base) {
2745 		ret = -ENOMEM;
2746 		goto out_no_ioremap;
2747 	}
2748 
2749 	if (vd->ftdmac020) {
2750 		u32 val;
2751 
2752 		val = readl(pl08x->base + FTDMAC020_REVISION);
2753 		dev_info(&pl08x->adev->dev, "FTDMAC020 %d.%d rel %d\n",
2754 			 (val >> 16) & 0xff, (val >> 8) & 0xff, val & 0xff);
2755 		val = readl(pl08x->base + FTDMAC020_FEATURE);
2756 		dev_info(&pl08x->adev->dev, "FTDMAC020 %d channels, "
2757 			 "%s built-in bridge, %s, %s linked lists\n",
2758 			 (val >> 12) & 0x0f,
2759 			 (val & BIT(10)) ? "no" : "has",
2760 			 (val & BIT(9)) ? "AHB0 and AHB1" : "AHB0",
2761 			 (val & BIT(8)) ? "supports" : "does not support");
2762 
2763 		/* Vendor data from feature register */
2764 		if (!(val & BIT(8)))
2765 			dev_warn(&pl08x->adev->dev,
2766 				 "linked lists not supported, required\n");
2767 		vd->channels = (val >> 12) & 0x0f;
2768 		vd->dualmaster = !!(val & BIT(9));
2769 	}
2770 
2771 	/* Initialize memcpy engine */
2772 	dma_cap_set(DMA_MEMCPY, pl08x->memcpy.cap_mask);
2773 	pl08x->memcpy.dev = &adev->dev;
2774 	pl08x->memcpy.device_free_chan_resources = pl08x_free_chan_resources;
2775 	pl08x->memcpy.device_prep_dma_memcpy = pl08x_prep_dma_memcpy;
2776 	pl08x->memcpy.device_prep_dma_interrupt = pl08x_prep_dma_interrupt;
2777 	pl08x->memcpy.device_tx_status = pl08x_dma_tx_status;
2778 	pl08x->memcpy.device_issue_pending = pl08x_issue_pending;
2779 	pl08x->memcpy.device_config = pl08x_config;
2780 	pl08x->memcpy.device_pause = pl08x_pause;
2781 	pl08x->memcpy.device_resume = pl08x_resume;
2782 	pl08x->memcpy.device_terminate_all = pl08x_terminate_all;
2783 	pl08x->memcpy.device_synchronize = pl08x_synchronize;
2784 	pl08x->memcpy.src_addr_widths = PL80X_DMA_BUSWIDTHS;
2785 	pl08x->memcpy.dst_addr_widths = PL80X_DMA_BUSWIDTHS;
2786 	pl08x->memcpy.directions = BIT(DMA_MEM_TO_MEM);
2787 	pl08x->memcpy.residue_granularity = DMA_RESIDUE_GRANULARITY_SEGMENT;
2788 	if (vd->ftdmac020)
2789 		pl08x->memcpy.copy_align = DMAENGINE_ALIGN_4_BYTES;
2790 
2791 
2792 	/*
2793 	 * Initialize slave engine, if the block has no signals, that means
2794 	 * we have no slave support.
2795 	 */
2796 	if (vd->signals) {
2797 		pl08x->has_slave = true;
2798 		dma_cap_set(DMA_SLAVE, pl08x->slave.cap_mask);
2799 		dma_cap_set(DMA_CYCLIC, pl08x->slave.cap_mask);
2800 		pl08x->slave.dev = &adev->dev;
2801 		pl08x->slave.device_free_chan_resources =
2802 			pl08x_free_chan_resources;
2803 		pl08x->slave.device_prep_dma_interrupt =
2804 			pl08x_prep_dma_interrupt;
2805 		pl08x->slave.device_tx_status = pl08x_dma_tx_status;
2806 		pl08x->slave.device_issue_pending = pl08x_issue_pending;
2807 		pl08x->slave.device_prep_slave_sg = pl08x_prep_slave_sg;
2808 		pl08x->slave.device_prep_dma_cyclic = pl08x_prep_dma_cyclic;
2809 		pl08x->slave.device_config = pl08x_config;
2810 		pl08x->slave.device_pause = pl08x_pause;
2811 		pl08x->slave.device_resume = pl08x_resume;
2812 		pl08x->slave.device_terminate_all = pl08x_terminate_all;
2813 		pl08x->slave.device_synchronize = pl08x_synchronize;
2814 		pl08x->slave.src_addr_widths = PL80X_DMA_BUSWIDTHS;
2815 		pl08x->slave.dst_addr_widths = PL80X_DMA_BUSWIDTHS;
2816 		pl08x->slave.directions =
2817 			BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
2818 		pl08x->slave.residue_granularity =
2819 			DMA_RESIDUE_GRANULARITY_SEGMENT;
2820 	}
2821 
2822 	/* Get the platform data */
2823 	pl08x->pd = dev_get_platdata(&adev->dev);
2824 	if (!pl08x->pd) {
2825 		if (np) {
2826 			ret = pl08x_of_probe(adev, pl08x, np);
2827 			if (ret)
2828 				goto out_no_platdata;
2829 		} else {
2830 			dev_err(&adev->dev, "no platform data supplied\n");
2831 			ret = -EINVAL;
2832 			goto out_no_platdata;
2833 		}
2834 	} else {
2835 		pl08x->slave.filter.map = pl08x->pd->slave_map;
2836 		pl08x->slave.filter.mapcnt = pl08x->pd->slave_map_len;
2837 		pl08x->slave.filter.fn = pl08x_filter_fn;
2838 	}
2839 
2840 	/* By default, AHB1 only.  If dualmaster, from platform */
2841 	pl08x->lli_buses = PL08X_AHB1;
2842 	pl08x->mem_buses = PL08X_AHB1;
2843 	if (pl08x->vd->dualmaster) {
2844 		pl08x->lli_buses = pl08x->pd->lli_buses;
2845 		pl08x->mem_buses = pl08x->pd->mem_buses;
2846 	}
2847 
2848 	if (vd->pl080s)
2849 		pl08x->lli_words = PL080S_LLI_WORDS;
2850 	else
2851 		pl08x->lli_words = PL080_LLI_WORDS;
2852 	tsfr_size = MAX_NUM_TSFR_LLIS * pl08x->lli_words * sizeof(u32);
2853 
2854 	/* A DMA memory pool for LLIs, align on 1-byte boundary */
2855 	pl08x->pool = dma_pool_create(DRIVER_NAME, &pl08x->adev->dev,
2856 						tsfr_size, PL08X_ALIGN, 0);
2857 	if (!pl08x->pool) {
2858 		ret = -ENOMEM;
2859 		goto out_no_lli_pool;
2860 	}
2861 
2862 	/* Turn on the PL08x */
2863 	pl08x_ensure_on(pl08x);
2864 
2865 	/* Clear any pending interrupts */
2866 	if (vd->ftdmac020)
2867 		/* This variant has error IRQs in bits 16-19 */
2868 		writel(0x0000FFFF, pl08x->base + PL080_ERR_CLEAR);
2869 	else
2870 		writel(0x000000FF, pl08x->base + PL080_ERR_CLEAR);
2871 	writel(0x000000FF, pl08x->base + PL080_TC_CLEAR);
2872 
2873 	/* Attach the interrupt handler */
2874 	ret = request_irq(adev->irq[0], pl08x_irq, 0, DRIVER_NAME, pl08x);
2875 	if (ret) {
2876 		dev_err(&adev->dev, "%s failed to request interrupt %d\n",
2877 			__func__, adev->irq[0]);
2878 		goto out_no_irq;
2879 	}
2880 
2881 	/* Initialize physical channels */
2882 	pl08x->phy_chans = kzalloc((vd->channels * sizeof(*pl08x->phy_chans)),
2883 			GFP_KERNEL);
2884 	if (!pl08x->phy_chans) {
2885 		ret = -ENOMEM;
2886 		goto out_no_phychans;
2887 	}
2888 
2889 	for (i = 0; i < vd->channels; i++) {
2890 		struct pl08x_phy_chan *ch = &pl08x->phy_chans[i];
2891 
2892 		ch->id = i;
2893 		ch->base = pl08x->base + PL080_Cx_BASE(i);
2894 		if (vd->ftdmac020) {
2895 			/* FTDMA020 has a special channel busy register */
2896 			ch->reg_busy = ch->base + FTDMAC020_CH_BUSY;
2897 			ch->reg_config = ch->base + FTDMAC020_CH_CFG;
2898 			ch->reg_control = ch->base + FTDMAC020_CH_CSR;
2899 			ch->reg_src = ch->base + FTDMAC020_CH_SRC_ADDR;
2900 			ch->reg_dst = ch->base + FTDMAC020_CH_DST_ADDR;
2901 			ch->reg_lli = ch->base + FTDMAC020_CH_LLP;
2902 			ch->ftdmac020 = true;
2903 		} else {
2904 			ch->reg_config = ch->base + vd->config_offset;
2905 			ch->reg_control = ch->base + PL080_CH_CONTROL;
2906 			ch->reg_src = ch->base + PL080_CH_SRC_ADDR;
2907 			ch->reg_dst = ch->base + PL080_CH_DST_ADDR;
2908 			ch->reg_lli = ch->base + PL080_CH_LLI;
2909 		}
2910 		if (vd->pl080s)
2911 			ch->pl080s = true;
2912 
2913 		spin_lock_init(&ch->lock);
2914 
2915 		/*
2916 		 * Nomadik variants can have channels that are locked
2917 		 * down for the secure world only. Lock up these channels
2918 		 * by perpetually serving a dummy virtual channel.
2919 		 */
2920 		if (vd->nomadik) {
2921 			u32 val;
2922 
2923 			val = readl(ch->reg_config);
2924 			if (val & (PL080N_CONFIG_ITPROT | PL080N_CONFIG_SECPROT)) {
2925 				dev_info(&adev->dev, "physical channel %d reserved for secure access only\n", i);
2926 				ch->locked = true;
2927 			}
2928 		}
2929 
2930 		dev_dbg(&adev->dev, "physical channel %d is %s\n",
2931 			i, pl08x_phy_channel_busy(ch) ? "BUSY" : "FREE");
2932 	}
2933 
2934 	/* Register as many memcpy channels as there are physical channels */
2935 	ret = pl08x_dma_init_virtual_channels(pl08x, &pl08x->memcpy,
2936 					      pl08x->vd->channels, false);
2937 	if (ret <= 0) {
2938 		dev_warn(&pl08x->adev->dev,
2939 			 "%s failed to enumerate memcpy channels - %d\n",
2940 			 __func__, ret);
2941 		goto out_no_memcpy;
2942 	}
2943 
2944 	/* Register slave channels */
2945 	if (pl08x->has_slave) {
2946 		ret = pl08x_dma_init_virtual_channels(pl08x, &pl08x->slave,
2947 					pl08x->pd->num_slave_channels, true);
2948 		if (ret < 0) {
2949 			dev_warn(&pl08x->adev->dev,
2950 				 "%s failed to enumerate slave channels - %d\n",
2951 				 __func__, ret);
2952 			goto out_no_slave;
2953 		}
2954 	}
2955 
2956 	ret = dma_async_device_register(&pl08x->memcpy);
2957 	if (ret) {
2958 		dev_warn(&pl08x->adev->dev,
2959 			"%s failed to register memcpy as an async device - %d\n",
2960 			__func__, ret);
2961 		goto out_no_memcpy_reg;
2962 	}
2963 
2964 	if (pl08x->has_slave) {
2965 		ret = dma_async_device_register(&pl08x->slave);
2966 		if (ret) {
2967 			dev_warn(&pl08x->adev->dev,
2968 			"%s failed to register slave as an async device - %d\n",
2969 			__func__, ret);
2970 			goto out_no_slave_reg;
2971 		}
2972 	}
2973 
2974 	amba_set_drvdata(adev, pl08x);
2975 	init_pl08x_debugfs(pl08x);
2976 	dev_info(&pl08x->adev->dev, "DMA: PL%03x%s rev%u at 0x%08llx irq %d\n",
2977 		 amba_part(adev), pl08x->vd->pl080s ? "s" : "", amba_rev(adev),
2978 		 (unsigned long long)adev->res.start, adev->irq[0]);
2979 
2980 	return 0;
2981 
2982 out_no_slave_reg:
2983 	dma_async_device_unregister(&pl08x->memcpy);
2984 out_no_memcpy_reg:
2985 	if (pl08x->has_slave)
2986 		pl08x_free_virtual_channels(&pl08x->slave);
2987 out_no_slave:
2988 	pl08x_free_virtual_channels(&pl08x->memcpy);
2989 out_no_memcpy:
2990 	kfree(pl08x->phy_chans);
2991 out_no_phychans:
2992 	free_irq(adev->irq[0], pl08x);
2993 out_no_irq:
2994 	dma_pool_destroy(pl08x->pool);
2995 out_no_lli_pool:
2996 out_no_platdata:
2997 	iounmap(pl08x->base);
2998 out_no_ioremap:
2999 	kfree(pl08x);
3000 out_no_pl08x:
3001 	amba_release_regions(adev);
3002 	return ret;
3003 }
3004 
3005 /* PL080 has 8 channels and the PL080 have just 2 */
3006 static struct vendor_data vendor_pl080 = {
3007 	.config_offset = PL080_CH_CONFIG,
3008 	.channels = 8,
3009 	.signals = 16,
3010 	.dualmaster = true,
3011 	.max_transfer_size = PL080_CONTROL_TRANSFER_SIZE_MASK,
3012 };
3013 
3014 static struct vendor_data vendor_nomadik = {
3015 	.config_offset = PL080_CH_CONFIG,
3016 	.channels = 8,
3017 	.signals = 32,
3018 	.dualmaster = true,
3019 	.nomadik = true,
3020 	.max_transfer_size = PL080_CONTROL_TRANSFER_SIZE_MASK,
3021 };
3022 
3023 static struct vendor_data vendor_pl080s = {
3024 	.config_offset = PL080S_CH_CONFIG,
3025 	.channels = 8,
3026 	.signals = 32,
3027 	.pl080s = true,
3028 	.max_transfer_size = PL080S_CONTROL_TRANSFER_SIZE_MASK,
3029 };
3030 
3031 static struct vendor_data vendor_pl081 = {
3032 	.config_offset = PL080_CH_CONFIG,
3033 	.channels = 2,
3034 	.signals = 16,
3035 	.dualmaster = false,
3036 	.max_transfer_size = PL080_CONTROL_TRANSFER_SIZE_MASK,
3037 };
3038 
3039 static struct vendor_data vendor_ftdmac020 = {
3040 	.config_offset = PL080_CH_CONFIG,
3041 	.ftdmac020 = true,
3042 	.max_transfer_size = PL080_CONTROL_TRANSFER_SIZE_MASK,
3043 };
3044 
3045 static const struct amba_id pl08x_ids[] = {
3046 	/* Samsung PL080S variant */
3047 	{
3048 		.id	= 0x0a141080,
3049 		.mask	= 0xffffffff,
3050 		.data	= &vendor_pl080s,
3051 	},
3052 	/* PL080 */
3053 	{
3054 		.id	= 0x00041080,
3055 		.mask	= 0x000fffff,
3056 		.data	= &vendor_pl080,
3057 	},
3058 	/* PL081 */
3059 	{
3060 		.id	= 0x00041081,
3061 		.mask	= 0x000fffff,
3062 		.data	= &vendor_pl081,
3063 	},
3064 	/* Nomadik 8815 PL080 variant */
3065 	{
3066 		.id	= 0x00280080,
3067 		.mask	= 0x00ffffff,
3068 		.data	= &vendor_nomadik,
3069 	},
3070 	/* Faraday Technology FTDMAC020 */
3071 	{
3072 		.id	= 0x0003b080,
3073 		.mask	= 0x000fffff,
3074 		.data	= &vendor_ftdmac020,
3075 	},
3076 	{ 0, 0 },
3077 };
3078 
3079 MODULE_DEVICE_TABLE(amba, pl08x_ids);
3080 
3081 static struct amba_driver pl08x_amba_driver = {
3082 	.drv.name	= DRIVER_NAME,
3083 	.id_table	= pl08x_ids,
3084 	.probe		= pl08x_probe,
3085 };
3086 
3087 static int __init pl08x_init(void)
3088 {
3089 	int retval;
3090 	retval = amba_driver_register(&pl08x_amba_driver);
3091 	if (retval)
3092 		printk(KERN_WARNING DRIVER_NAME
3093 		       "failed to register as an AMBA device (%d)\n",
3094 		       retval);
3095 	return retval;
3096 }
3097 subsys_initcall(pl08x_init);
3098