xref: /openbmc/linux/drivers/dma/amba-pl08x.c (revision 3b64b188)
1 /*
2  * Copyright (c) 2006 ARM Ltd.
3  * Copyright (c) 2010 ST-Ericsson SA
4  *
5  * Author: Peter Pearse <peter.pearse@arm.com>
6  * Author: Linus Walleij <linus.walleij@stericsson.com>
7  *
8  * This program is free software; you can redistribute it and/or modify it
9  * under the terms of the GNU General Public License as published by the Free
10  * Software Foundation; either version 2 of the License, or (at your option)
11  * any later version.
12  *
13  * This program is distributed in the hope that it will be useful, but WITHOUT
14  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
15  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
16  * more details.
17  *
18  * You should have received a copy of the GNU General Public License along with
19  * this program; if not, write to the Free Software Foundation, Inc., 59
20  * Temple Place - Suite 330, Boston, MA  02111-1307, USA.
21  *
22  * The full GNU General Public License is in this distribution in the file
23  * called COPYING.
24  *
25  * Documentation: ARM DDI 0196G == PL080
26  * Documentation: ARM DDI 0218E == PL081
27  *
28  * PL080 & PL081 both have 16 sets of DMA signals that can be routed to any
29  * channel.
30  *
31  * The PL080 has 8 channels available for simultaneous use, and the PL081
32  * has only two channels. So on these DMA controllers the number of channels
33  * and the number of incoming DMA signals are two totally different things.
34  * It is usually not possible to theoretically handle all physical signals,
35  * so a multiplexing scheme with possible denial of use is necessary.
36  *
37  * The PL080 has a dual bus master, PL081 has a single master.
38  *
39  * Memory to peripheral transfer may be visualized as
40  *	Get data from memory to DMAC
41  *	Until no data left
42  *		On burst request from peripheral
43  *			Destination burst from DMAC to peripheral
44  *			Clear burst request
45  *	Raise terminal count interrupt
46  *
47  * For peripherals with a FIFO:
48  * Source      burst size == half the depth of the peripheral FIFO
49  * Destination burst size == the depth of the peripheral FIFO
50  *
51  * (Bursts are irrelevant for mem to mem transfers - there are no burst
52  * signals, the DMA controller will simply facilitate its AHB master.)
53  *
54  * ASSUMES default (little) endianness for DMA transfers
55  *
56  * The PL08x has two flow control settings:
57  *  - DMAC flow control: the transfer size defines the number of transfers
58  *    which occur for the current LLI entry, and the DMAC raises TC at the
59  *    end of every LLI entry.  Observed behaviour shows the DMAC listening
60  *    to both the BREQ and SREQ signals (contrary to documented),
61  *    transferring data if either is active.  The LBREQ and LSREQ signals
62  *    are ignored.
63  *
64  *  - Peripheral flow control: the transfer size is ignored (and should be
65  *    zero).  The data is transferred from the current LLI entry, until
66  *    after the final transfer signalled by LBREQ or LSREQ.  The DMAC
67  *    will then move to the next LLI entry.
68  *
69  * Global TODO:
70  * - Break out common code from arch/arm/mach-s3c64xx and share
71  */
72 #include <linux/amba/bus.h>
73 #include <linux/amba/pl08x.h>
74 #include <linux/debugfs.h>
75 #include <linux/delay.h>
76 #include <linux/device.h>
77 #include <linux/dmaengine.h>
78 #include <linux/dmapool.h>
79 #include <linux/dma-mapping.h>
80 #include <linux/init.h>
81 #include <linux/interrupt.h>
82 #include <linux/module.h>
83 #include <linux/pm_runtime.h>
84 #include <linux/seq_file.h>
85 #include <linux/slab.h>
86 #include <asm/hardware/pl080.h>
87 
88 #include "dmaengine.h"
89 #include "virt-dma.h"
90 
91 #define DRIVER_NAME	"pl08xdmac"
92 
93 static struct amba_driver pl08x_amba_driver;
94 struct pl08x_driver_data;
95 
96 /**
97  * struct vendor_data - vendor-specific config parameters for PL08x derivatives
98  * @channels: the number of channels available in this variant
99  * @dualmaster: whether this version supports dual AHB masters or not.
100  * @nomadik: whether the channels have Nomadik security extension bits
101  *	that need to be checked for permission before use and some registers are
102  *	missing
103  */
104 struct vendor_data {
105 	u8 channels;
106 	bool dualmaster;
107 	bool nomadik;
108 };
109 
110 /*
111  * PL08X private data structures
112  * An LLI struct - see PL08x TRM.  Note that next uses bit[0] as a bus bit,
113  * start & end do not - their bus bit info is in cctl.  Also note that these
114  * are fixed 32-bit quantities.
115  */
116 struct pl08x_lli {
117 	u32 src;
118 	u32 dst;
119 	u32 lli;
120 	u32 cctl;
121 };
122 
123 /**
124  * struct pl08x_bus_data - information of source or destination
125  * busses for a transfer
126  * @addr: current address
127  * @maxwidth: the maximum width of a transfer on this bus
128  * @buswidth: the width of this bus in bytes: 1, 2 or 4
129  */
130 struct pl08x_bus_data {
131 	dma_addr_t addr;
132 	u8 maxwidth;
133 	u8 buswidth;
134 };
135 
136 /**
137  * struct pl08x_phy_chan - holder for the physical channels
138  * @id: physical index to this channel
139  * @lock: a lock to use when altering an instance of this struct
140  * @serving: the virtual channel currently being served by this physical
141  * channel
142  * @locked: channel unavailable for the system, e.g. dedicated to secure
143  * world
144  */
145 struct pl08x_phy_chan {
146 	unsigned int id;
147 	void __iomem *base;
148 	spinlock_t lock;
149 	struct pl08x_dma_chan *serving;
150 	bool locked;
151 };
152 
153 /**
154  * struct pl08x_sg - structure containing data per sg
155  * @src_addr: src address of sg
156  * @dst_addr: dst address of sg
157  * @len: transfer len in bytes
158  * @node: node for txd's dsg_list
159  */
160 struct pl08x_sg {
161 	dma_addr_t src_addr;
162 	dma_addr_t dst_addr;
163 	size_t len;
164 	struct list_head node;
165 };
166 
167 /**
168  * struct pl08x_txd - wrapper for struct dma_async_tx_descriptor
169  * @vd: virtual DMA descriptor
170  * @dsg_list: list of children sg's
171  * @llis_bus: DMA memory address (physical) start for the LLIs
172  * @llis_va: virtual memory address start for the LLIs
173  * @cctl: control reg values for current txd
174  * @ccfg: config reg values for current txd
175  * @done: this marks completed descriptors, which should not have their
176  *   mux released.
177  */
178 struct pl08x_txd {
179 	struct virt_dma_desc vd;
180 	struct list_head dsg_list;
181 	dma_addr_t llis_bus;
182 	struct pl08x_lli *llis_va;
183 	/* Default cctl value for LLIs */
184 	u32 cctl;
185 	/*
186 	 * Settings to be put into the physical channel when we
187 	 * trigger this txd.  Other registers are in llis_va[0].
188 	 */
189 	u32 ccfg;
190 	bool done;
191 };
192 
193 /**
194  * struct pl08x_dma_chan_state - holds the PL08x specific virtual channel
195  * states
196  * @PL08X_CHAN_IDLE: the channel is idle
197  * @PL08X_CHAN_RUNNING: the channel has allocated a physical transport
198  * channel and is running a transfer on it
199  * @PL08X_CHAN_PAUSED: the channel has allocated a physical transport
200  * channel, but the transfer is currently paused
201  * @PL08X_CHAN_WAITING: the channel is waiting for a physical transport
202  * channel to become available (only pertains to memcpy channels)
203  */
204 enum pl08x_dma_chan_state {
205 	PL08X_CHAN_IDLE,
206 	PL08X_CHAN_RUNNING,
207 	PL08X_CHAN_PAUSED,
208 	PL08X_CHAN_WAITING,
209 };
210 
211 /**
212  * struct pl08x_dma_chan - this structure wraps a DMA ENGINE channel
213  * @vc: wrappped virtual channel
214  * @phychan: the physical channel utilized by this channel, if there is one
215  * @name: name of channel
216  * @cd: channel platform data
217  * @runtime_addr: address for RX/TX according to the runtime config
218  * @at: active transaction on this channel
219  * @lock: a lock for this channel data
220  * @host: a pointer to the host (internal use)
221  * @state: whether the channel is idle, paused, running etc
222  * @slave: whether this channel is a device (slave) or for memcpy
223  * @signal: the physical DMA request signal which this channel is using
224  * @mux_use: count of descriptors using this DMA request signal setting
225  */
226 struct pl08x_dma_chan {
227 	struct virt_dma_chan vc;
228 	struct pl08x_phy_chan *phychan;
229 	const char *name;
230 	const struct pl08x_channel_data *cd;
231 	struct dma_slave_config cfg;
232 	struct pl08x_txd *at;
233 	struct pl08x_driver_data *host;
234 	enum pl08x_dma_chan_state state;
235 	bool slave;
236 	int signal;
237 	unsigned mux_use;
238 };
239 
240 /**
241  * struct pl08x_driver_data - the local state holder for the PL08x
242  * @slave: slave engine for this instance
243  * @memcpy: memcpy engine for this instance
244  * @base: virtual memory base (remapped) for the PL08x
245  * @adev: the corresponding AMBA (PrimeCell) bus entry
246  * @vd: vendor data for this PL08x variant
247  * @pd: platform data passed in from the platform/machine
248  * @phy_chans: array of data for the physical channels
249  * @pool: a pool for the LLI descriptors
250  * @lli_buses: bitmask to or in to LLI pointer selecting AHB port for LLI
251  * fetches
252  * @mem_buses: set to indicate memory transfers on AHB2.
253  * @lock: a spinlock for this struct
254  */
255 struct pl08x_driver_data {
256 	struct dma_device slave;
257 	struct dma_device memcpy;
258 	void __iomem *base;
259 	struct amba_device *adev;
260 	const struct vendor_data *vd;
261 	struct pl08x_platform_data *pd;
262 	struct pl08x_phy_chan *phy_chans;
263 	struct dma_pool *pool;
264 	u8 lli_buses;
265 	u8 mem_buses;
266 };
267 
268 /*
269  * PL08X specific defines
270  */
271 
272 /* Size (bytes) of each LLI buffer allocated for one transfer */
273 # define PL08X_LLI_TSFR_SIZE	0x2000
274 
275 /* Maximum times we call dma_pool_alloc on this pool without freeing */
276 #define MAX_NUM_TSFR_LLIS	(PL08X_LLI_TSFR_SIZE/sizeof(struct pl08x_lli))
277 #define PL08X_ALIGN		8
278 
279 static inline struct pl08x_dma_chan *to_pl08x_chan(struct dma_chan *chan)
280 {
281 	return container_of(chan, struct pl08x_dma_chan, vc.chan);
282 }
283 
284 static inline struct pl08x_txd *to_pl08x_txd(struct dma_async_tx_descriptor *tx)
285 {
286 	return container_of(tx, struct pl08x_txd, vd.tx);
287 }
288 
289 /*
290  * Mux handling.
291  *
292  * This gives us the DMA request input to the PL08x primecell which the
293  * peripheral described by the channel data will be routed to, possibly
294  * via a board/SoC specific external MUX.  One important point to note
295  * here is that this does not depend on the physical channel.
296  */
297 static int pl08x_request_mux(struct pl08x_dma_chan *plchan)
298 {
299 	const struct pl08x_platform_data *pd = plchan->host->pd;
300 	int ret;
301 
302 	if (plchan->mux_use++ == 0 && pd->get_signal) {
303 		ret = pd->get_signal(plchan->cd);
304 		if (ret < 0) {
305 			plchan->mux_use = 0;
306 			return ret;
307 		}
308 
309 		plchan->signal = ret;
310 	}
311 	return 0;
312 }
313 
314 static void pl08x_release_mux(struct pl08x_dma_chan *plchan)
315 {
316 	const struct pl08x_platform_data *pd = plchan->host->pd;
317 
318 	if (plchan->signal >= 0) {
319 		WARN_ON(plchan->mux_use == 0);
320 
321 		if (--plchan->mux_use == 0 && pd->put_signal) {
322 			pd->put_signal(plchan->cd, plchan->signal);
323 			plchan->signal = -1;
324 		}
325 	}
326 }
327 
328 /*
329  * Physical channel handling
330  */
331 
332 /* Whether a certain channel is busy or not */
333 static int pl08x_phy_channel_busy(struct pl08x_phy_chan *ch)
334 {
335 	unsigned int val;
336 
337 	val = readl(ch->base + PL080_CH_CONFIG);
338 	return val & PL080_CONFIG_ACTIVE;
339 }
340 
341 /*
342  * Set the initial DMA register values i.e. those for the first LLI
343  * The next LLI pointer and the configuration interrupt bit have
344  * been set when the LLIs were constructed.  Poke them into the hardware
345  * and start the transfer.
346  */
347 static void pl08x_start_next_txd(struct pl08x_dma_chan *plchan)
348 {
349 	struct pl08x_driver_data *pl08x = plchan->host;
350 	struct pl08x_phy_chan *phychan = plchan->phychan;
351 	struct virt_dma_desc *vd = vchan_next_desc(&plchan->vc);
352 	struct pl08x_txd *txd = to_pl08x_txd(&vd->tx);
353 	struct pl08x_lli *lli;
354 	u32 val;
355 
356 	list_del(&txd->vd.node);
357 
358 	plchan->at = txd;
359 
360 	/* Wait for channel inactive */
361 	while (pl08x_phy_channel_busy(phychan))
362 		cpu_relax();
363 
364 	lli = &txd->llis_va[0];
365 
366 	dev_vdbg(&pl08x->adev->dev,
367 		"WRITE channel %d: csrc=0x%08x, cdst=0x%08x, "
368 		"clli=0x%08x, cctl=0x%08x, ccfg=0x%08x\n",
369 		phychan->id, lli->src, lli->dst, lli->lli, lli->cctl,
370 		txd->ccfg);
371 
372 	writel(lli->src, phychan->base + PL080_CH_SRC_ADDR);
373 	writel(lli->dst, phychan->base + PL080_CH_DST_ADDR);
374 	writel(lli->lli, phychan->base + PL080_CH_LLI);
375 	writel(lli->cctl, phychan->base + PL080_CH_CONTROL);
376 	writel(txd->ccfg, phychan->base + PL080_CH_CONFIG);
377 
378 	/* Enable the DMA channel */
379 	/* Do not access config register until channel shows as disabled */
380 	while (readl(pl08x->base + PL080_EN_CHAN) & (1 << phychan->id))
381 		cpu_relax();
382 
383 	/* Do not access config register until channel shows as inactive */
384 	val = readl(phychan->base + PL080_CH_CONFIG);
385 	while ((val & PL080_CONFIG_ACTIVE) || (val & PL080_CONFIG_ENABLE))
386 		val = readl(phychan->base + PL080_CH_CONFIG);
387 
388 	writel(val | PL080_CONFIG_ENABLE, phychan->base + PL080_CH_CONFIG);
389 }
390 
391 /*
392  * Pause the channel by setting the HALT bit.
393  *
394  * For M->P transfers, pause the DMAC first and then stop the peripheral -
395  * the FIFO can only drain if the peripheral is still requesting data.
396  * (note: this can still timeout if the DMAC FIFO never drains of data.)
397  *
398  * For P->M transfers, disable the peripheral first to stop it filling
399  * the DMAC FIFO, and then pause the DMAC.
400  */
401 static void pl08x_pause_phy_chan(struct pl08x_phy_chan *ch)
402 {
403 	u32 val;
404 	int timeout;
405 
406 	/* Set the HALT bit and wait for the FIFO to drain */
407 	val = readl(ch->base + PL080_CH_CONFIG);
408 	val |= PL080_CONFIG_HALT;
409 	writel(val, ch->base + PL080_CH_CONFIG);
410 
411 	/* Wait for channel inactive */
412 	for (timeout = 1000; timeout; timeout--) {
413 		if (!pl08x_phy_channel_busy(ch))
414 			break;
415 		udelay(1);
416 	}
417 	if (pl08x_phy_channel_busy(ch))
418 		pr_err("pl08x: channel%u timeout waiting for pause\n", ch->id);
419 }
420 
421 static void pl08x_resume_phy_chan(struct pl08x_phy_chan *ch)
422 {
423 	u32 val;
424 
425 	/* Clear the HALT bit */
426 	val = readl(ch->base + PL080_CH_CONFIG);
427 	val &= ~PL080_CONFIG_HALT;
428 	writel(val, ch->base + PL080_CH_CONFIG);
429 }
430 
431 /*
432  * pl08x_terminate_phy_chan() stops the channel, clears the FIFO and
433  * clears any pending interrupt status.  This should not be used for
434  * an on-going transfer, but as a method of shutting down a channel
435  * (eg, when it's no longer used) or terminating a transfer.
436  */
437 static void pl08x_terminate_phy_chan(struct pl08x_driver_data *pl08x,
438 	struct pl08x_phy_chan *ch)
439 {
440 	u32 val = readl(ch->base + PL080_CH_CONFIG);
441 
442 	val &= ~(PL080_CONFIG_ENABLE | PL080_CONFIG_ERR_IRQ_MASK |
443 	         PL080_CONFIG_TC_IRQ_MASK);
444 
445 	writel(val, ch->base + PL080_CH_CONFIG);
446 
447 	writel(1 << ch->id, pl08x->base + PL080_ERR_CLEAR);
448 	writel(1 << ch->id, pl08x->base + PL080_TC_CLEAR);
449 }
450 
451 static inline u32 get_bytes_in_cctl(u32 cctl)
452 {
453 	/* The source width defines the number of bytes */
454 	u32 bytes = cctl & PL080_CONTROL_TRANSFER_SIZE_MASK;
455 
456 	switch (cctl >> PL080_CONTROL_SWIDTH_SHIFT) {
457 	case PL080_WIDTH_8BIT:
458 		break;
459 	case PL080_WIDTH_16BIT:
460 		bytes *= 2;
461 		break;
462 	case PL080_WIDTH_32BIT:
463 		bytes *= 4;
464 		break;
465 	}
466 	return bytes;
467 }
468 
469 /* The channel should be paused when calling this */
470 static u32 pl08x_getbytes_chan(struct pl08x_dma_chan *plchan)
471 {
472 	struct pl08x_phy_chan *ch;
473 	struct pl08x_txd *txd;
474 	size_t bytes = 0;
475 
476 	ch = plchan->phychan;
477 	txd = plchan->at;
478 
479 	/*
480 	 * Follow the LLIs to get the number of remaining
481 	 * bytes in the currently active transaction.
482 	 */
483 	if (ch && txd) {
484 		u32 clli = readl(ch->base + PL080_CH_LLI) & ~PL080_LLI_LM_AHB2;
485 
486 		/* First get the remaining bytes in the active transfer */
487 		bytes = get_bytes_in_cctl(readl(ch->base + PL080_CH_CONTROL));
488 
489 		if (clli) {
490 			struct pl08x_lli *llis_va = txd->llis_va;
491 			dma_addr_t llis_bus = txd->llis_bus;
492 			int index;
493 
494 			BUG_ON(clli < llis_bus || clli >= llis_bus +
495 				sizeof(struct pl08x_lli) * MAX_NUM_TSFR_LLIS);
496 
497 			/*
498 			 * Locate the next LLI - as this is an array,
499 			 * it's simple maths to find.
500 			 */
501 			index = (clli - llis_bus) / sizeof(struct pl08x_lli);
502 
503 			for (; index < MAX_NUM_TSFR_LLIS; index++) {
504 				bytes += get_bytes_in_cctl(llis_va[index].cctl);
505 
506 				/*
507 				 * A LLI pointer of 0 terminates the LLI list
508 				 */
509 				if (!llis_va[index].lli)
510 					break;
511 			}
512 		}
513 	}
514 
515 	return bytes;
516 }
517 
518 /*
519  * Allocate a physical channel for a virtual channel
520  *
521  * Try to locate a physical channel to be used for this transfer. If all
522  * are taken return NULL and the requester will have to cope by using
523  * some fallback PIO mode or retrying later.
524  */
525 static struct pl08x_phy_chan *
526 pl08x_get_phy_channel(struct pl08x_driver_data *pl08x,
527 		      struct pl08x_dma_chan *virt_chan)
528 {
529 	struct pl08x_phy_chan *ch = NULL;
530 	unsigned long flags;
531 	int i;
532 
533 	for (i = 0; i < pl08x->vd->channels; i++) {
534 		ch = &pl08x->phy_chans[i];
535 
536 		spin_lock_irqsave(&ch->lock, flags);
537 
538 		if (!ch->locked && !ch->serving) {
539 			ch->serving = virt_chan;
540 			spin_unlock_irqrestore(&ch->lock, flags);
541 			break;
542 		}
543 
544 		spin_unlock_irqrestore(&ch->lock, flags);
545 	}
546 
547 	if (i == pl08x->vd->channels) {
548 		/* No physical channel available, cope with it */
549 		return NULL;
550 	}
551 
552 	return ch;
553 }
554 
555 /* Mark the physical channel as free.  Note, this write is atomic. */
556 static inline void pl08x_put_phy_channel(struct pl08x_driver_data *pl08x,
557 					 struct pl08x_phy_chan *ch)
558 {
559 	ch->serving = NULL;
560 }
561 
562 /*
563  * Try to allocate a physical channel.  When successful, assign it to
564  * this virtual channel, and initiate the next descriptor.  The
565  * virtual channel lock must be held at this point.
566  */
567 static void pl08x_phy_alloc_and_start(struct pl08x_dma_chan *plchan)
568 {
569 	struct pl08x_driver_data *pl08x = plchan->host;
570 	struct pl08x_phy_chan *ch;
571 
572 	ch = pl08x_get_phy_channel(pl08x, plchan);
573 	if (!ch) {
574 		dev_dbg(&pl08x->adev->dev, "no physical channel available for xfer on %s\n", plchan->name);
575 		plchan->state = PL08X_CHAN_WAITING;
576 		return;
577 	}
578 
579 	dev_dbg(&pl08x->adev->dev, "allocated physical channel %d for xfer on %s\n",
580 		ch->id, plchan->name);
581 
582 	plchan->phychan = ch;
583 	plchan->state = PL08X_CHAN_RUNNING;
584 	pl08x_start_next_txd(plchan);
585 }
586 
587 static void pl08x_phy_reassign_start(struct pl08x_phy_chan *ch,
588 	struct pl08x_dma_chan *plchan)
589 {
590 	struct pl08x_driver_data *pl08x = plchan->host;
591 
592 	dev_dbg(&pl08x->adev->dev, "reassigned physical channel %d for xfer on %s\n",
593 		ch->id, plchan->name);
594 
595 	/*
596 	 * We do this without taking the lock; we're really only concerned
597 	 * about whether this pointer is NULL or not, and we're guaranteed
598 	 * that this will only be called when it _already_ is non-NULL.
599 	 */
600 	ch->serving = plchan;
601 	plchan->phychan = ch;
602 	plchan->state = PL08X_CHAN_RUNNING;
603 	pl08x_start_next_txd(plchan);
604 }
605 
606 /*
607  * Free a physical DMA channel, potentially reallocating it to another
608  * virtual channel if we have any pending.
609  */
610 static void pl08x_phy_free(struct pl08x_dma_chan *plchan)
611 {
612 	struct pl08x_driver_data *pl08x = plchan->host;
613 	struct pl08x_dma_chan *p, *next;
614 
615  retry:
616 	next = NULL;
617 
618 	/* Find a waiting virtual channel for the next transfer. */
619 	list_for_each_entry(p, &pl08x->memcpy.channels, vc.chan.device_node)
620 		if (p->state == PL08X_CHAN_WAITING) {
621 			next = p;
622 			break;
623 		}
624 
625 	if (!next) {
626 		list_for_each_entry(p, &pl08x->slave.channels, vc.chan.device_node)
627 			if (p->state == PL08X_CHAN_WAITING) {
628 				next = p;
629 				break;
630 			}
631 	}
632 
633 	/* Ensure that the physical channel is stopped */
634 	pl08x_terminate_phy_chan(pl08x, plchan->phychan);
635 
636 	if (next) {
637 		bool success;
638 
639 		/*
640 		 * Eww.  We know this isn't going to deadlock
641 		 * but lockdep probably doesn't.
642 		 */
643 		spin_lock(&next->vc.lock);
644 		/* Re-check the state now that we have the lock */
645 		success = next->state == PL08X_CHAN_WAITING;
646 		if (success)
647 			pl08x_phy_reassign_start(plchan->phychan, next);
648 		spin_unlock(&next->vc.lock);
649 
650 		/* If the state changed, try to find another channel */
651 		if (!success)
652 			goto retry;
653 	} else {
654 		/* No more jobs, so free up the physical channel */
655 		pl08x_put_phy_channel(pl08x, plchan->phychan);
656 	}
657 
658 	plchan->phychan = NULL;
659 	plchan->state = PL08X_CHAN_IDLE;
660 }
661 
662 /*
663  * LLI handling
664  */
665 
666 static inline unsigned int pl08x_get_bytes_for_cctl(unsigned int coded)
667 {
668 	switch (coded) {
669 	case PL080_WIDTH_8BIT:
670 		return 1;
671 	case PL080_WIDTH_16BIT:
672 		return 2;
673 	case PL080_WIDTH_32BIT:
674 		return 4;
675 	default:
676 		break;
677 	}
678 	BUG();
679 	return 0;
680 }
681 
682 static inline u32 pl08x_cctl_bits(u32 cctl, u8 srcwidth, u8 dstwidth,
683 				  size_t tsize)
684 {
685 	u32 retbits = cctl;
686 
687 	/* Remove all src, dst and transfer size bits */
688 	retbits &= ~PL080_CONTROL_DWIDTH_MASK;
689 	retbits &= ~PL080_CONTROL_SWIDTH_MASK;
690 	retbits &= ~PL080_CONTROL_TRANSFER_SIZE_MASK;
691 
692 	/* Then set the bits according to the parameters */
693 	switch (srcwidth) {
694 	case 1:
695 		retbits |= PL080_WIDTH_8BIT << PL080_CONTROL_SWIDTH_SHIFT;
696 		break;
697 	case 2:
698 		retbits |= PL080_WIDTH_16BIT << PL080_CONTROL_SWIDTH_SHIFT;
699 		break;
700 	case 4:
701 		retbits |= PL080_WIDTH_32BIT << PL080_CONTROL_SWIDTH_SHIFT;
702 		break;
703 	default:
704 		BUG();
705 		break;
706 	}
707 
708 	switch (dstwidth) {
709 	case 1:
710 		retbits |= PL080_WIDTH_8BIT << PL080_CONTROL_DWIDTH_SHIFT;
711 		break;
712 	case 2:
713 		retbits |= PL080_WIDTH_16BIT << PL080_CONTROL_DWIDTH_SHIFT;
714 		break;
715 	case 4:
716 		retbits |= PL080_WIDTH_32BIT << PL080_CONTROL_DWIDTH_SHIFT;
717 		break;
718 	default:
719 		BUG();
720 		break;
721 	}
722 
723 	retbits |= tsize << PL080_CONTROL_TRANSFER_SIZE_SHIFT;
724 	return retbits;
725 }
726 
727 struct pl08x_lli_build_data {
728 	struct pl08x_txd *txd;
729 	struct pl08x_bus_data srcbus;
730 	struct pl08x_bus_data dstbus;
731 	size_t remainder;
732 	u32 lli_bus;
733 };
734 
735 /*
736  * Autoselect a master bus to use for the transfer. Slave will be the chosen as
737  * victim in case src & dest are not similarly aligned. i.e. If after aligning
738  * masters address with width requirements of transfer (by sending few byte by
739  * byte data), slave is still not aligned, then its width will be reduced to
740  * BYTE.
741  * - prefers the destination bus if both available
742  * - prefers bus with fixed address (i.e. peripheral)
743  */
744 static void pl08x_choose_master_bus(struct pl08x_lli_build_data *bd,
745 	struct pl08x_bus_data **mbus, struct pl08x_bus_data **sbus, u32 cctl)
746 {
747 	if (!(cctl & PL080_CONTROL_DST_INCR)) {
748 		*mbus = &bd->dstbus;
749 		*sbus = &bd->srcbus;
750 	} else if (!(cctl & PL080_CONTROL_SRC_INCR)) {
751 		*mbus = &bd->srcbus;
752 		*sbus = &bd->dstbus;
753 	} else {
754 		if (bd->dstbus.buswidth >= bd->srcbus.buswidth) {
755 			*mbus = &bd->dstbus;
756 			*sbus = &bd->srcbus;
757 		} else {
758 			*mbus = &bd->srcbus;
759 			*sbus = &bd->dstbus;
760 		}
761 	}
762 }
763 
764 /*
765  * Fills in one LLI for a certain transfer descriptor and advance the counter
766  */
767 static void pl08x_fill_lli_for_desc(struct pl08x_lli_build_data *bd,
768 	int num_llis, int len, u32 cctl)
769 {
770 	struct pl08x_lli *llis_va = bd->txd->llis_va;
771 	dma_addr_t llis_bus = bd->txd->llis_bus;
772 
773 	BUG_ON(num_llis >= MAX_NUM_TSFR_LLIS);
774 
775 	llis_va[num_llis].cctl = cctl;
776 	llis_va[num_llis].src = bd->srcbus.addr;
777 	llis_va[num_llis].dst = bd->dstbus.addr;
778 	llis_va[num_llis].lli = llis_bus + (num_llis + 1) *
779 		sizeof(struct pl08x_lli);
780 	llis_va[num_llis].lli |= bd->lli_bus;
781 
782 	if (cctl & PL080_CONTROL_SRC_INCR)
783 		bd->srcbus.addr += len;
784 	if (cctl & PL080_CONTROL_DST_INCR)
785 		bd->dstbus.addr += len;
786 
787 	BUG_ON(bd->remainder < len);
788 
789 	bd->remainder -= len;
790 }
791 
792 static inline void prep_byte_width_lli(struct pl08x_lli_build_data *bd,
793 		u32 *cctl, u32 len, int num_llis, size_t *total_bytes)
794 {
795 	*cctl = pl08x_cctl_bits(*cctl, 1, 1, len);
796 	pl08x_fill_lli_for_desc(bd, num_llis, len, *cctl);
797 	(*total_bytes) += len;
798 }
799 
800 /*
801  * This fills in the table of LLIs for the transfer descriptor
802  * Note that we assume we never have to change the burst sizes
803  * Return 0 for error
804  */
805 static int pl08x_fill_llis_for_desc(struct pl08x_driver_data *pl08x,
806 			      struct pl08x_txd *txd)
807 {
808 	struct pl08x_bus_data *mbus, *sbus;
809 	struct pl08x_lli_build_data bd;
810 	int num_llis = 0;
811 	u32 cctl, early_bytes = 0;
812 	size_t max_bytes_per_lli, total_bytes;
813 	struct pl08x_lli *llis_va;
814 	struct pl08x_sg *dsg;
815 
816 	txd->llis_va = dma_pool_alloc(pl08x->pool, GFP_NOWAIT, &txd->llis_bus);
817 	if (!txd->llis_va) {
818 		dev_err(&pl08x->adev->dev, "%s no memory for llis\n", __func__);
819 		return 0;
820 	}
821 
822 	bd.txd = txd;
823 	bd.lli_bus = (pl08x->lli_buses & PL08X_AHB2) ? PL080_LLI_LM_AHB2 : 0;
824 	cctl = txd->cctl;
825 
826 	/* Find maximum width of the source bus */
827 	bd.srcbus.maxwidth =
828 		pl08x_get_bytes_for_cctl((cctl & PL080_CONTROL_SWIDTH_MASK) >>
829 				       PL080_CONTROL_SWIDTH_SHIFT);
830 
831 	/* Find maximum width of the destination bus */
832 	bd.dstbus.maxwidth =
833 		pl08x_get_bytes_for_cctl((cctl & PL080_CONTROL_DWIDTH_MASK) >>
834 				       PL080_CONTROL_DWIDTH_SHIFT);
835 
836 	list_for_each_entry(dsg, &txd->dsg_list, node) {
837 		total_bytes = 0;
838 		cctl = txd->cctl;
839 
840 		bd.srcbus.addr = dsg->src_addr;
841 		bd.dstbus.addr = dsg->dst_addr;
842 		bd.remainder = dsg->len;
843 		bd.srcbus.buswidth = bd.srcbus.maxwidth;
844 		bd.dstbus.buswidth = bd.dstbus.maxwidth;
845 
846 		pl08x_choose_master_bus(&bd, &mbus, &sbus, cctl);
847 
848 		dev_vdbg(&pl08x->adev->dev, "src=0x%08x%s/%u dst=0x%08x%s/%u len=%zu\n",
849 			bd.srcbus.addr, cctl & PL080_CONTROL_SRC_INCR ? "+" : "",
850 			bd.srcbus.buswidth,
851 			bd.dstbus.addr, cctl & PL080_CONTROL_DST_INCR ? "+" : "",
852 			bd.dstbus.buswidth,
853 			bd.remainder);
854 		dev_vdbg(&pl08x->adev->dev, "mbus=%s sbus=%s\n",
855 			mbus == &bd.srcbus ? "src" : "dst",
856 			sbus == &bd.srcbus ? "src" : "dst");
857 
858 		/*
859 		 * Zero length is only allowed if all these requirements are
860 		 * met:
861 		 * - flow controller is peripheral.
862 		 * - src.addr is aligned to src.width
863 		 * - dst.addr is aligned to dst.width
864 		 *
865 		 * sg_len == 1 should be true, as there can be two cases here:
866 		 *
867 		 * - Memory addresses are contiguous and are not scattered.
868 		 *   Here, Only one sg will be passed by user driver, with
869 		 *   memory address and zero length. We pass this to controller
870 		 *   and after the transfer it will receive the last burst
871 		 *   request from peripheral and so transfer finishes.
872 		 *
873 		 * - Memory addresses are scattered and are not contiguous.
874 		 *   Here, Obviously as DMA controller doesn't know when a lli's
875 		 *   transfer gets over, it can't load next lli. So in this
876 		 *   case, there has to be an assumption that only one lli is
877 		 *   supported. Thus, we can't have scattered addresses.
878 		 */
879 		if (!bd.remainder) {
880 			u32 fc = (txd->ccfg & PL080_CONFIG_FLOW_CONTROL_MASK) >>
881 				PL080_CONFIG_FLOW_CONTROL_SHIFT;
882 			if (!((fc >= PL080_FLOW_SRC2DST_DST) &&
883 					(fc <= PL080_FLOW_SRC2DST_SRC))) {
884 				dev_err(&pl08x->adev->dev, "%s sg len can't be zero",
885 					__func__);
886 				return 0;
887 			}
888 
889 			if ((bd.srcbus.addr % bd.srcbus.buswidth) ||
890 					(bd.dstbus.addr % bd.dstbus.buswidth)) {
891 				dev_err(&pl08x->adev->dev,
892 					"%s src & dst address must be aligned to src"
893 					" & dst width if peripheral is flow controller",
894 					__func__);
895 				return 0;
896 			}
897 
898 			cctl = pl08x_cctl_bits(cctl, bd.srcbus.buswidth,
899 					bd.dstbus.buswidth, 0);
900 			pl08x_fill_lli_for_desc(&bd, num_llis++, 0, cctl);
901 			break;
902 		}
903 
904 		/*
905 		 * Send byte by byte for following cases
906 		 * - Less than a bus width available
907 		 * - until master bus is aligned
908 		 */
909 		if (bd.remainder < mbus->buswidth)
910 			early_bytes = bd.remainder;
911 		else if ((mbus->addr) % (mbus->buswidth)) {
912 			early_bytes = mbus->buswidth - (mbus->addr) %
913 				(mbus->buswidth);
914 			if ((bd.remainder - early_bytes) < mbus->buswidth)
915 				early_bytes = bd.remainder;
916 		}
917 
918 		if (early_bytes) {
919 			dev_vdbg(&pl08x->adev->dev,
920 				"%s byte width LLIs (remain 0x%08x)\n",
921 				__func__, bd.remainder);
922 			prep_byte_width_lli(&bd, &cctl, early_bytes, num_llis++,
923 				&total_bytes);
924 		}
925 
926 		if (bd.remainder) {
927 			/*
928 			 * Master now aligned
929 			 * - if slave is not then we must set its width down
930 			 */
931 			if (sbus->addr % sbus->buswidth) {
932 				dev_dbg(&pl08x->adev->dev,
933 					"%s set down bus width to one byte\n",
934 					__func__);
935 
936 				sbus->buswidth = 1;
937 			}
938 
939 			/*
940 			 * Bytes transferred = tsize * src width, not
941 			 * MIN(buswidths)
942 			 */
943 			max_bytes_per_lli = bd.srcbus.buswidth *
944 				PL080_CONTROL_TRANSFER_SIZE_MASK;
945 			dev_vdbg(&pl08x->adev->dev,
946 				"%s max bytes per lli = %zu\n",
947 				__func__, max_bytes_per_lli);
948 
949 			/*
950 			 * Make largest possible LLIs until less than one bus
951 			 * width left
952 			 */
953 			while (bd.remainder > (mbus->buswidth - 1)) {
954 				size_t lli_len, tsize, width;
955 
956 				/*
957 				 * If enough left try to send max possible,
958 				 * otherwise try to send the remainder
959 				 */
960 				lli_len = min(bd.remainder, max_bytes_per_lli);
961 
962 				/*
963 				 * Check against maximum bus alignment:
964 				 * Calculate actual transfer size in relation to
965 				 * bus width an get a maximum remainder of the
966 				 * highest bus width - 1
967 				 */
968 				width = max(mbus->buswidth, sbus->buswidth);
969 				lli_len = (lli_len / width) * width;
970 				tsize = lli_len / bd.srcbus.buswidth;
971 
972 				dev_vdbg(&pl08x->adev->dev,
973 					"%s fill lli with single lli chunk of "
974 					"size 0x%08zx (remainder 0x%08zx)\n",
975 					__func__, lli_len, bd.remainder);
976 
977 				cctl = pl08x_cctl_bits(cctl, bd.srcbus.buswidth,
978 					bd.dstbus.buswidth, tsize);
979 				pl08x_fill_lli_for_desc(&bd, num_llis++,
980 						lli_len, cctl);
981 				total_bytes += lli_len;
982 			}
983 
984 			/*
985 			 * Send any odd bytes
986 			 */
987 			if (bd.remainder) {
988 				dev_vdbg(&pl08x->adev->dev,
989 					"%s align with boundary, send odd bytes (remain %zu)\n",
990 					__func__, bd.remainder);
991 				prep_byte_width_lli(&bd, &cctl, bd.remainder,
992 						num_llis++, &total_bytes);
993 			}
994 		}
995 
996 		if (total_bytes != dsg->len) {
997 			dev_err(&pl08x->adev->dev,
998 				"%s size of encoded lli:s don't match total txd, transferred 0x%08zx from size 0x%08zx\n",
999 				__func__, total_bytes, dsg->len);
1000 			return 0;
1001 		}
1002 
1003 		if (num_llis >= MAX_NUM_TSFR_LLIS) {
1004 			dev_err(&pl08x->adev->dev,
1005 				"%s need to increase MAX_NUM_TSFR_LLIS from 0x%08x\n",
1006 				__func__, (u32) MAX_NUM_TSFR_LLIS);
1007 			return 0;
1008 		}
1009 	}
1010 
1011 	llis_va = txd->llis_va;
1012 	/* The final LLI terminates the LLI. */
1013 	llis_va[num_llis - 1].lli = 0;
1014 	/* The final LLI element shall also fire an interrupt. */
1015 	llis_va[num_llis - 1].cctl |= PL080_CONTROL_TC_IRQ_EN;
1016 
1017 #ifdef VERBOSE_DEBUG
1018 	{
1019 		int i;
1020 
1021 		dev_vdbg(&pl08x->adev->dev,
1022 			 "%-3s %-9s  %-10s %-10s %-10s %s\n",
1023 			 "lli", "", "csrc", "cdst", "clli", "cctl");
1024 		for (i = 0; i < num_llis; i++) {
1025 			dev_vdbg(&pl08x->adev->dev,
1026 				 "%3d @%p: 0x%08x 0x%08x 0x%08x 0x%08x\n",
1027 				 i, &llis_va[i], llis_va[i].src,
1028 				 llis_va[i].dst, llis_va[i].lli, llis_va[i].cctl
1029 				);
1030 		}
1031 	}
1032 #endif
1033 
1034 	return num_llis;
1035 }
1036 
1037 static void pl08x_free_txd(struct pl08x_driver_data *pl08x,
1038 			   struct pl08x_txd *txd)
1039 {
1040 	struct pl08x_sg *dsg, *_dsg;
1041 
1042 	if (txd->llis_va)
1043 		dma_pool_free(pl08x->pool, txd->llis_va, txd->llis_bus);
1044 
1045 	list_for_each_entry_safe(dsg, _dsg, &txd->dsg_list, node) {
1046 		list_del(&dsg->node);
1047 		kfree(dsg);
1048 	}
1049 
1050 	kfree(txd);
1051 }
1052 
1053 static void pl08x_unmap_buffers(struct pl08x_txd *txd)
1054 {
1055 	struct device *dev = txd->vd.tx.chan->device->dev;
1056 	struct pl08x_sg *dsg;
1057 
1058 	if (!(txd->vd.tx.flags & DMA_COMPL_SKIP_SRC_UNMAP)) {
1059 		if (txd->vd.tx.flags & DMA_COMPL_SRC_UNMAP_SINGLE)
1060 			list_for_each_entry(dsg, &txd->dsg_list, node)
1061 				dma_unmap_single(dev, dsg->src_addr, dsg->len,
1062 						DMA_TO_DEVICE);
1063 		else {
1064 			list_for_each_entry(dsg, &txd->dsg_list, node)
1065 				dma_unmap_page(dev, dsg->src_addr, dsg->len,
1066 						DMA_TO_DEVICE);
1067 		}
1068 	}
1069 	if (!(txd->vd.tx.flags & DMA_COMPL_SKIP_DEST_UNMAP)) {
1070 		if (txd->vd.tx.flags & DMA_COMPL_DEST_UNMAP_SINGLE)
1071 			list_for_each_entry(dsg, &txd->dsg_list, node)
1072 				dma_unmap_single(dev, dsg->dst_addr, dsg->len,
1073 						DMA_FROM_DEVICE);
1074 		else
1075 			list_for_each_entry(dsg, &txd->dsg_list, node)
1076 				dma_unmap_page(dev, dsg->dst_addr, dsg->len,
1077 						DMA_FROM_DEVICE);
1078 	}
1079 }
1080 
1081 static void pl08x_desc_free(struct virt_dma_desc *vd)
1082 {
1083 	struct pl08x_txd *txd = to_pl08x_txd(&vd->tx);
1084 	struct pl08x_dma_chan *plchan = to_pl08x_chan(vd->tx.chan);
1085 
1086 	if (!plchan->slave)
1087 		pl08x_unmap_buffers(txd);
1088 
1089 	if (!txd->done)
1090 		pl08x_release_mux(plchan);
1091 
1092 	pl08x_free_txd(plchan->host, txd);
1093 }
1094 
1095 static void pl08x_free_txd_list(struct pl08x_driver_data *pl08x,
1096 				struct pl08x_dma_chan *plchan)
1097 {
1098 	LIST_HEAD(head);
1099 	struct pl08x_txd *txd;
1100 
1101 	vchan_get_all_descriptors(&plchan->vc, &head);
1102 
1103 	while (!list_empty(&head)) {
1104 		txd = list_first_entry(&head, struct pl08x_txd, vd.node);
1105 		list_del(&txd->vd.node);
1106 		pl08x_desc_free(&txd->vd);
1107 	}
1108 }
1109 
1110 /*
1111  * The DMA ENGINE API
1112  */
1113 static int pl08x_alloc_chan_resources(struct dma_chan *chan)
1114 {
1115 	return 0;
1116 }
1117 
1118 static void pl08x_free_chan_resources(struct dma_chan *chan)
1119 {
1120 	/* Ensure all queued descriptors are freed */
1121 	vchan_free_chan_resources(to_virt_chan(chan));
1122 }
1123 
1124 static struct dma_async_tx_descriptor *pl08x_prep_dma_interrupt(
1125 		struct dma_chan *chan, unsigned long flags)
1126 {
1127 	struct dma_async_tx_descriptor *retval = NULL;
1128 
1129 	return retval;
1130 }
1131 
1132 /*
1133  * Code accessing dma_async_is_complete() in a tight loop may give problems.
1134  * If slaves are relying on interrupts to signal completion this function
1135  * must not be called with interrupts disabled.
1136  */
1137 static enum dma_status pl08x_dma_tx_status(struct dma_chan *chan,
1138 		dma_cookie_t cookie, struct dma_tx_state *txstate)
1139 {
1140 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1141 	struct virt_dma_desc *vd;
1142 	unsigned long flags;
1143 	enum dma_status ret;
1144 	size_t bytes = 0;
1145 
1146 	ret = dma_cookie_status(chan, cookie, txstate);
1147 	if (ret == DMA_SUCCESS)
1148 		return ret;
1149 
1150 	/*
1151 	 * There's no point calculating the residue if there's
1152 	 * no txstate to store the value.
1153 	 */
1154 	if (!txstate) {
1155 		if (plchan->state == PL08X_CHAN_PAUSED)
1156 			ret = DMA_PAUSED;
1157 		return ret;
1158 	}
1159 
1160 	spin_lock_irqsave(&plchan->vc.lock, flags);
1161 	ret = dma_cookie_status(chan, cookie, txstate);
1162 	if (ret != DMA_SUCCESS) {
1163 		vd = vchan_find_desc(&plchan->vc, cookie);
1164 		if (vd) {
1165 			/* On the issued list, so hasn't been processed yet */
1166 			struct pl08x_txd *txd = to_pl08x_txd(&vd->tx);
1167 			struct pl08x_sg *dsg;
1168 
1169 			list_for_each_entry(dsg, &txd->dsg_list, node)
1170 				bytes += dsg->len;
1171 		} else {
1172 			bytes = pl08x_getbytes_chan(plchan);
1173 		}
1174 	}
1175 	spin_unlock_irqrestore(&plchan->vc.lock, flags);
1176 
1177 	/*
1178 	 * This cookie not complete yet
1179 	 * Get number of bytes left in the active transactions and queue
1180 	 */
1181 	dma_set_residue(txstate, bytes);
1182 
1183 	if (plchan->state == PL08X_CHAN_PAUSED && ret == DMA_IN_PROGRESS)
1184 		ret = DMA_PAUSED;
1185 
1186 	/* Whether waiting or running, we're in progress */
1187 	return ret;
1188 }
1189 
1190 /* PrimeCell DMA extension */
1191 struct burst_table {
1192 	u32 burstwords;
1193 	u32 reg;
1194 };
1195 
1196 static const struct burst_table burst_sizes[] = {
1197 	{
1198 		.burstwords = 256,
1199 		.reg = PL080_BSIZE_256,
1200 	},
1201 	{
1202 		.burstwords = 128,
1203 		.reg = PL080_BSIZE_128,
1204 	},
1205 	{
1206 		.burstwords = 64,
1207 		.reg = PL080_BSIZE_64,
1208 	},
1209 	{
1210 		.burstwords = 32,
1211 		.reg = PL080_BSIZE_32,
1212 	},
1213 	{
1214 		.burstwords = 16,
1215 		.reg = PL080_BSIZE_16,
1216 	},
1217 	{
1218 		.burstwords = 8,
1219 		.reg = PL080_BSIZE_8,
1220 	},
1221 	{
1222 		.burstwords = 4,
1223 		.reg = PL080_BSIZE_4,
1224 	},
1225 	{
1226 		.burstwords = 0,
1227 		.reg = PL080_BSIZE_1,
1228 	},
1229 };
1230 
1231 /*
1232  * Given the source and destination available bus masks, select which
1233  * will be routed to each port.  We try to have source and destination
1234  * on separate ports, but always respect the allowable settings.
1235  */
1236 static u32 pl08x_select_bus(u8 src, u8 dst)
1237 {
1238 	u32 cctl = 0;
1239 
1240 	if (!(dst & PL08X_AHB1) || ((dst & PL08X_AHB2) && (src & PL08X_AHB1)))
1241 		cctl |= PL080_CONTROL_DST_AHB2;
1242 	if (!(src & PL08X_AHB1) || ((src & PL08X_AHB2) && !(dst & PL08X_AHB2)))
1243 		cctl |= PL080_CONTROL_SRC_AHB2;
1244 
1245 	return cctl;
1246 }
1247 
1248 static u32 pl08x_cctl(u32 cctl)
1249 {
1250 	cctl &= ~(PL080_CONTROL_SRC_AHB2 | PL080_CONTROL_DST_AHB2 |
1251 		  PL080_CONTROL_SRC_INCR | PL080_CONTROL_DST_INCR |
1252 		  PL080_CONTROL_PROT_MASK);
1253 
1254 	/* Access the cell in privileged mode, non-bufferable, non-cacheable */
1255 	return cctl | PL080_CONTROL_PROT_SYS;
1256 }
1257 
1258 static u32 pl08x_width(enum dma_slave_buswidth width)
1259 {
1260 	switch (width) {
1261 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
1262 		return PL080_WIDTH_8BIT;
1263 	case DMA_SLAVE_BUSWIDTH_2_BYTES:
1264 		return PL080_WIDTH_16BIT;
1265 	case DMA_SLAVE_BUSWIDTH_4_BYTES:
1266 		return PL080_WIDTH_32BIT;
1267 	default:
1268 		return ~0;
1269 	}
1270 }
1271 
1272 static u32 pl08x_burst(u32 maxburst)
1273 {
1274 	int i;
1275 
1276 	for (i = 0; i < ARRAY_SIZE(burst_sizes); i++)
1277 		if (burst_sizes[i].burstwords <= maxburst)
1278 			break;
1279 
1280 	return burst_sizes[i].reg;
1281 }
1282 
1283 static u32 pl08x_get_cctl(struct pl08x_dma_chan *plchan,
1284 	enum dma_slave_buswidth addr_width, u32 maxburst)
1285 {
1286 	u32 width, burst, cctl = 0;
1287 
1288 	width = pl08x_width(addr_width);
1289 	if (width == ~0)
1290 		return ~0;
1291 
1292 	cctl |= width << PL080_CONTROL_SWIDTH_SHIFT;
1293 	cctl |= width << PL080_CONTROL_DWIDTH_SHIFT;
1294 
1295 	/*
1296 	 * If this channel will only request single transfers, set this
1297 	 * down to ONE element.  Also select one element if no maxburst
1298 	 * is specified.
1299 	 */
1300 	if (plchan->cd->single)
1301 		maxburst = 1;
1302 
1303 	burst = pl08x_burst(maxburst);
1304 	cctl |= burst << PL080_CONTROL_SB_SIZE_SHIFT;
1305 	cctl |= burst << PL080_CONTROL_DB_SIZE_SHIFT;
1306 
1307 	return pl08x_cctl(cctl);
1308 }
1309 
1310 static int dma_set_runtime_config(struct dma_chan *chan,
1311 				  struct dma_slave_config *config)
1312 {
1313 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1314 
1315 	if (!plchan->slave)
1316 		return -EINVAL;
1317 
1318 	/* Reject definitely invalid configurations */
1319 	if (config->src_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES ||
1320 	    config->dst_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES)
1321 		return -EINVAL;
1322 
1323 	plchan->cfg = *config;
1324 
1325 	return 0;
1326 }
1327 
1328 /*
1329  * Slave transactions callback to the slave device to allow
1330  * synchronization of slave DMA signals with the DMAC enable
1331  */
1332 static void pl08x_issue_pending(struct dma_chan *chan)
1333 {
1334 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1335 	unsigned long flags;
1336 
1337 	spin_lock_irqsave(&plchan->vc.lock, flags);
1338 	if (vchan_issue_pending(&plchan->vc)) {
1339 		if (!plchan->phychan && plchan->state != PL08X_CHAN_WAITING)
1340 			pl08x_phy_alloc_and_start(plchan);
1341 	}
1342 	spin_unlock_irqrestore(&plchan->vc.lock, flags);
1343 }
1344 
1345 static struct pl08x_txd *pl08x_get_txd(struct pl08x_dma_chan *plchan)
1346 {
1347 	struct pl08x_txd *txd = kzalloc(sizeof(*txd), GFP_NOWAIT);
1348 
1349 	if (txd) {
1350 		INIT_LIST_HEAD(&txd->dsg_list);
1351 
1352 		/* Always enable error and terminal interrupts */
1353 		txd->ccfg = PL080_CONFIG_ERR_IRQ_MASK |
1354 			    PL080_CONFIG_TC_IRQ_MASK;
1355 	}
1356 	return txd;
1357 }
1358 
1359 /*
1360  * Initialize a descriptor to be used by memcpy submit
1361  */
1362 static struct dma_async_tx_descriptor *pl08x_prep_dma_memcpy(
1363 		struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
1364 		size_t len, unsigned long flags)
1365 {
1366 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1367 	struct pl08x_driver_data *pl08x = plchan->host;
1368 	struct pl08x_txd *txd;
1369 	struct pl08x_sg *dsg;
1370 	int ret;
1371 
1372 	txd = pl08x_get_txd(plchan);
1373 	if (!txd) {
1374 		dev_err(&pl08x->adev->dev,
1375 			"%s no memory for descriptor\n", __func__);
1376 		return NULL;
1377 	}
1378 
1379 	dsg = kzalloc(sizeof(struct pl08x_sg), GFP_NOWAIT);
1380 	if (!dsg) {
1381 		pl08x_free_txd(pl08x, txd);
1382 		dev_err(&pl08x->adev->dev, "%s no memory for pl080 sg\n",
1383 				__func__);
1384 		return NULL;
1385 	}
1386 	list_add_tail(&dsg->node, &txd->dsg_list);
1387 
1388 	dsg->src_addr = src;
1389 	dsg->dst_addr = dest;
1390 	dsg->len = len;
1391 
1392 	/* Set platform data for m2m */
1393 	txd->ccfg |= PL080_FLOW_MEM2MEM << PL080_CONFIG_FLOW_CONTROL_SHIFT;
1394 	txd->cctl = pl08x->pd->memcpy_channel.cctl_memcpy &
1395 			~(PL080_CONTROL_DST_AHB2 | PL080_CONTROL_SRC_AHB2);
1396 
1397 	/* Both to be incremented or the code will break */
1398 	txd->cctl |= PL080_CONTROL_SRC_INCR | PL080_CONTROL_DST_INCR;
1399 
1400 	if (pl08x->vd->dualmaster)
1401 		txd->cctl |= pl08x_select_bus(pl08x->mem_buses,
1402 					      pl08x->mem_buses);
1403 
1404 	ret = pl08x_fill_llis_for_desc(plchan->host, txd);
1405 	if (!ret) {
1406 		pl08x_free_txd(pl08x, txd);
1407 		return NULL;
1408 	}
1409 
1410 	return vchan_tx_prep(&plchan->vc, &txd->vd, flags);
1411 }
1412 
1413 static struct dma_async_tx_descriptor *pl08x_prep_slave_sg(
1414 		struct dma_chan *chan, struct scatterlist *sgl,
1415 		unsigned int sg_len, enum dma_transfer_direction direction,
1416 		unsigned long flags, void *context)
1417 {
1418 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1419 	struct pl08x_driver_data *pl08x = plchan->host;
1420 	struct pl08x_txd *txd;
1421 	struct pl08x_sg *dsg;
1422 	struct scatterlist *sg;
1423 	enum dma_slave_buswidth addr_width;
1424 	dma_addr_t slave_addr;
1425 	int ret, tmp;
1426 	u8 src_buses, dst_buses;
1427 	u32 maxburst, cctl;
1428 
1429 	dev_dbg(&pl08x->adev->dev, "%s prepare transaction of %d bytes from %s\n",
1430 			__func__, sg_dma_len(sgl), plchan->name);
1431 
1432 	txd = pl08x_get_txd(plchan);
1433 	if (!txd) {
1434 		dev_err(&pl08x->adev->dev, "%s no txd\n", __func__);
1435 		return NULL;
1436 	}
1437 
1438 	/*
1439 	 * Set up addresses, the PrimeCell configured address
1440 	 * will take precedence since this may configure the
1441 	 * channel target address dynamically at runtime.
1442 	 */
1443 	if (direction == DMA_MEM_TO_DEV) {
1444 		cctl = PL080_CONTROL_SRC_INCR;
1445 		slave_addr = plchan->cfg.dst_addr;
1446 		addr_width = plchan->cfg.dst_addr_width;
1447 		maxburst = plchan->cfg.dst_maxburst;
1448 		src_buses = pl08x->mem_buses;
1449 		dst_buses = plchan->cd->periph_buses;
1450 	} else if (direction == DMA_DEV_TO_MEM) {
1451 		cctl = PL080_CONTROL_DST_INCR;
1452 		slave_addr = plchan->cfg.src_addr;
1453 		addr_width = plchan->cfg.src_addr_width;
1454 		maxburst = plchan->cfg.src_maxburst;
1455 		src_buses = plchan->cd->periph_buses;
1456 		dst_buses = pl08x->mem_buses;
1457 	} else {
1458 		pl08x_free_txd(pl08x, txd);
1459 		dev_err(&pl08x->adev->dev,
1460 			"%s direction unsupported\n", __func__);
1461 		return NULL;
1462 	}
1463 
1464 	cctl |= pl08x_get_cctl(plchan, addr_width, maxburst);
1465 	if (cctl == ~0) {
1466 		pl08x_free_txd(pl08x, txd);
1467 		dev_err(&pl08x->adev->dev,
1468 			"DMA slave configuration botched?\n");
1469 		return NULL;
1470 	}
1471 
1472 	txd->cctl = cctl | pl08x_select_bus(src_buses, dst_buses);
1473 
1474 	if (plchan->cfg.device_fc)
1475 		tmp = (direction == DMA_MEM_TO_DEV) ? PL080_FLOW_MEM2PER_PER :
1476 			PL080_FLOW_PER2MEM_PER;
1477 	else
1478 		tmp = (direction == DMA_MEM_TO_DEV) ? PL080_FLOW_MEM2PER :
1479 			PL080_FLOW_PER2MEM;
1480 
1481 	txd->ccfg |= tmp << PL080_CONFIG_FLOW_CONTROL_SHIFT;
1482 
1483 	ret = pl08x_request_mux(plchan);
1484 	if (ret < 0) {
1485 		pl08x_free_txd(pl08x, txd);
1486 		dev_dbg(&pl08x->adev->dev,
1487 			"unable to mux for transfer on %s due to platform restrictions\n",
1488 			plchan->name);
1489 		return NULL;
1490 	}
1491 
1492 	dev_dbg(&pl08x->adev->dev, "allocated DMA request signal %d for xfer on %s\n",
1493 		 plchan->signal, plchan->name);
1494 
1495 	/* Assign the flow control signal to this channel */
1496 	if (direction == DMA_MEM_TO_DEV)
1497 		txd->ccfg |= plchan->signal << PL080_CONFIG_DST_SEL_SHIFT;
1498 	else
1499 		txd->ccfg |= plchan->signal << PL080_CONFIG_SRC_SEL_SHIFT;
1500 
1501 	for_each_sg(sgl, sg, sg_len, tmp) {
1502 		dsg = kzalloc(sizeof(struct pl08x_sg), GFP_NOWAIT);
1503 		if (!dsg) {
1504 			pl08x_release_mux(plchan);
1505 			pl08x_free_txd(pl08x, txd);
1506 			dev_err(&pl08x->adev->dev, "%s no mem for pl080 sg\n",
1507 					__func__);
1508 			return NULL;
1509 		}
1510 		list_add_tail(&dsg->node, &txd->dsg_list);
1511 
1512 		dsg->len = sg_dma_len(sg);
1513 		if (direction == DMA_MEM_TO_DEV) {
1514 			dsg->src_addr = sg_dma_address(sg);
1515 			dsg->dst_addr = slave_addr;
1516 		} else {
1517 			dsg->src_addr = slave_addr;
1518 			dsg->dst_addr = sg_dma_address(sg);
1519 		}
1520 	}
1521 
1522 	ret = pl08x_fill_llis_for_desc(plchan->host, txd);
1523 	if (!ret) {
1524 		pl08x_release_mux(plchan);
1525 		pl08x_free_txd(pl08x, txd);
1526 		return NULL;
1527 	}
1528 
1529 	return vchan_tx_prep(&plchan->vc, &txd->vd, flags);
1530 }
1531 
1532 static int pl08x_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
1533 			 unsigned long arg)
1534 {
1535 	struct pl08x_dma_chan *plchan = to_pl08x_chan(chan);
1536 	struct pl08x_driver_data *pl08x = plchan->host;
1537 	unsigned long flags;
1538 	int ret = 0;
1539 
1540 	/* Controls applicable to inactive channels */
1541 	if (cmd == DMA_SLAVE_CONFIG) {
1542 		return dma_set_runtime_config(chan,
1543 					      (struct dma_slave_config *)arg);
1544 	}
1545 
1546 	/*
1547 	 * Anything succeeds on channels with no physical allocation and
1548 	 * no queued transfers.
1549 	 */
1550 	spin_lock_irqsave(&plchan->vc.lock, flags);
1551 	if (!plchan->phychan && !plchan->at) {
1552 		spin_unlock_irqrestore(&plchan->vc.lock, flags);
1553 		return 0;
1554 	}
1555 
1556 	switch (cmd) {
1557 	case DMA_TERMINATE_ALL:
1558 		plchan->state = PL08X_CHAN_IDLE;
1559 
1560 		if (plchan->phychan) {
1561 			/*
1562 			 * Mark physical channel as free and free any slave
1563 			 * signal
1564 			 */
1565 			pl08x_phy_free(plchan);
1566 		}
1567 		/* Dequeue jobs and free LLIs */
1568 		if (plchan->at) {
1569 			pl08x_desc_free(&plchan->at->vd);
1570 			plchan->at = NULL;
1571 		}
1572 		/* Dequeue jobs not yet fired as well */
1573 		pl08x_free_txd_list(pl08x, plchan);
1574 		break;
1575 	case DMA_PAUSE:
1576 		pl08x_pause_phy_chan(plchan->phychan);
1577 		plchan->state = PL08X_CHAN_PAUSED;
1578 		break;
1579 	case DMA_RESUME:
1580 		pl08x_resume_phy_chan(plchan->phychan);
1581 		plchan->state = PL08X_CHAN_RUNNING;
1582 		break;
1583 	default:
1584 		/* Unknown command */
1585 		ret = -ENXIO;
1586 		break;
1587 	}
1588 
1589 	spin_unlock_irqrestore(&plchan->vc.lock, flags);
1590 
1591 	return ret;
1592 }
1593 
1594 bool pl08x_filter_id(struct dma_chan *chan, void *chan_id)
1595 {
1596 	struct pl08x_dma_chan *plchan;
1597 	char *name = chan_id;
1598 
1599 	/* Reject channels for devices not bound to this driver */
1600 	if (chan->device->dev->driver != &pl08x_amba_driver.drv)
1601 		return false;
1602 
1603 	plchan = to_pl08x_chan(chan);
1604 
1605 	/* Check that the channel is not taken! */
1606 	if (!strcmp(plchan->name, name))
1607 		return true;
1608 
1609 	return false;
1610 }
1611 
1612 /*
1613  * Just check that the device is there and active
1614  * TODO: turn this bit on/off depending on the number of physical channels
1615  * actually used, if it is zero... well shut it off. That will save some
1616  * power. Cut the clock at the same time.
1617  */
1618 static void pl08x_ensure_on(struct pl08x_driver_data *pl08x)
1619 {
1620 	/* The Nomadik variant does not have the config register */
1621 	if (pl08x->vd->nomadik)
1622 		return;
1623 	writel(PL080_CONFIG_ENABLE, pl08x->base + PL080_CONFIG);
1624 }
1625 
1626 static irqreturn_t pl08x_irq(int irq, void *dev)
1627 {
1628 	struct pl08x_driver_data *pl08x = dev;
1629 	u32 mask = 0, err, tc, i;
1630 
1631 	/* check & clear - ERR & TC interrupts */
1632 	err = readl(pl08x->base + PL080_ERR_STATUS);
1633 	if (err) {
1634 		dev_err(&pl08x->adev->dev, "%s error interrupt, register value 0x%08x\n",
1635 			__func__, err);
1636 		writel(err, pl08x->base + PL080_ERR_CLEAR);
1637 	}
1638 	tc = readl(pl08x->base + PL080_TC_STATUS);
1639 	if (tc)
1640 		writel(tc, pl08x->base + PL080_TC_CLEAR);
1641 
1642 	if (!err && !tc)
1643 		return IRQ_NONE;
1644 
1645 	for (i = 0; i < pl08x->vd->channels; i++) {
1646 		if (((1 << i) & err) || ((1 << i) & tc)) {
1647 			/* Locate physical channel */
1648 			struct pl08x_phy_chan *phychan = &pl08x->phy_chans[i];
1649 			struct pl08x_dma_chan *plchan = phychan->serving;
1650 			struct pl08x_txd *tx;
1651 
1652 			if (!plchan) {
1653 				dev_err(&pl08x->adev->dev,
1654 					"%s Error TC interrupt on unused channel: 0x%08x\n",
1655 					__func__, i);
1656 				continue;
1657 			}
1658 
1659 			spin_lock(&plchan->vc.lock);
1660 			tx = plchan->at;
1661 			if (tx) {
1662 				plchan->at = NULL;
1663 				/*
1664 				 * This descriptor is done, release its mux
1665 				 * reservation.
1666 				 */
1667 				pl08x_release_mux(plchan);
1668 				tx->done = true;
1669 				vchan_cookie_complete(&tx->vd);
1670 
1671 				/*
1672 				 * And start the next descriptor (if any),
1673 				 * otherwise free this channel.
1674 				 */
1675 				if (vchan_next_desc(&plchan->vc))
1676 					pl08x_start_next_txd(plchan);
1677 				else
1678 					pl08x_phy_free(plchan);
1679 			}
1680 			spin_unlock(&plchan->vc.lock);
1681 
1682 			mask |= (1 << i);
1683 		}
1684 	}
1685 
1686 	return mask ? IRQ_HANDLED : IRQ_NONE;
1687 }
1688 
1689 static void pl08x_dma_slave_init(struct pl08x_dma_chan *chan)
1690 {
1691 	chan->slave = true;
1692 	chan->name = chan->cd->bus_id;
1693 	chan->cfg.src_addr = chan->cd->addr;
1694 	chan->cfg.dst_addr = chan->cd->addr;
1695 }
1696 
1697 /*
1698  * Initialise the DMAC memcpy/slave channels.
1699  * Make a local wrapper to hold required data
1700  */
1701 static int pl08x_dma_init_virtual_channels(struct pl08x_driver_data *pl08x,
1702 		struct dma_device *dmadev, unsigned int channels, bool slave)
1703 {
1704 	struct pl08x_dma_chan *chan;
1705 	int i;
1706 
1707 	INIT_LIST_HEAD(&dmadev->channels);
1708 
1709 	/*
1710 	 * Register as many many memcpy as we have physical channels,
1711 	 * we won't always be able to use all but the code will have
1712 	 * to cope with that situation.
1713 	 */
1714 	for (i = 0; i < channels; i++) {
1715 		chan = kzalloc(sizeof(*chan), GFP_KERNEL);
1716 		if (!chan) {
1717 			dev_err(&pl08x->adev->dev,
1718 				"%s no memory for channel\n", __func__);
1719 			return -ENOMEM;
1720 		}
1721 
1722 		chan->host = pl08x;
1723 		chan->state = PL08X_CHAN_IDLE;
1724 		chan->signal = -1;
1725 
1726 		if (slave) {
1727 			chan->cd = &pl08x->pd->slave_channels[i];
1728 			pl08x_dma_slave_init(chan);
1729 		} else {
1730 			chan->cd = &pl08x->pd->memcpy_channel;
1731 			chan->name = kasprintf(GFP_KERNEL, "memcpy%d", i);
1732 			if (!chan->name) {
1733 				kfree(chan);
1734 				return -ENOMEM;
1735 			}
1736 		}
1737 		dev_dbg(&pl08x->adev->dev,
1738 			 "initialize virtual channel \"%s\"\n",
1739 			 chan->name);
1740 
1741 		chan->vc.desc_free = pl08x_desc_free;
1742 		vchan_init(&chan->vc, dmadev);
1743 	}
1744 	dev_info(&pl08x->adev->dev, "initialized %d virtual %s channels\n",
1745 		 i, slave ? "slave" : "memcpy");
1746 	return i;
1747 }
1748 
1749 static void pl08x_free_virtual_channels(struct dma_device *dmadev)
1750 {
1751 	struct pl08x_dma_chan *chan = NULL;
1752 	struct pl08x_dma_chan *next;
1753 
1754 	list_for_each_entry_safe(chan,
1755 				 next, &dmadev->channels, vc.chan.device_node) {
1756 		list_del(&chan->vc.chan.device_node);
1757 		kfree(chan);
1758 	}
1759 }
1760 
1761 #ifdef CONFIG_DEBUG_FS
1762 static const char *pl08x_state_str(enum pl08x_dma_chan_state state)
1763 {
1764 	switch (state) {
1765 	case PL08X_CHAN_IDLE:
1766 		return "idle";
1767 	case PL08X_CHAN_RUNNING:
1768 		return "running";
1769 	case PL08X_CHAN_PAUSED:
1770 		return "paused";
1771 	case PL08X_CHAN_WAITING:
1772 		return "waiting";
1773 	default:
1774 		break;
1775 	}
1776 	return "UNKNOWN STATE";
1777 }
1778 
1779 static int pl08x_debugfs_show(struct seq_file *s, void *data)
1780 {
1781 	struct pl08x_driver_data *pl08x = s->private;
1782 	struct pl08x_dma_chan *chan;
1783 	struct pl08x_phy_chan *ch;
1784 	unsigned long flags;
1785 	int i;
1786 
1787 	seq_printf(s, "PL08x physical channels:\n");
1788 	seq_printf(s, "CHANNEL:\tUSER:\n");
1789 	seq_printf(s, "--------\t-----\n");
1790 	for (i = 0; i < pl08x->vd->channels; i++) {
1791 		struct pl08x_dma_chan *virt_chan;
1792 
1793 		ch = &pl08x->phy_chans[i];
1794 
1795 		spin_lock_irqsave(&ch->lock, flags);
1796 		virt_chan = ch->serving;
1797 
1798 		seq_printf(s, "%d\t\t%s%s\n",
1799 			   ch->id,
1800 			   virt_chan ? virt_chan->name : "(none)",
1801 			   ch->locked ? " LOCKED" : "");
1802 
1803 		spin_unlock_irqrestore(&ch->lock, flags);
1804 	}
1805 
1806 	seq_printf(s, "\nPL08x virtual memcpy channels:\n");
1807 	seq_printf(s, "CHANNEL:\tSTATE:\n");
1808 	seq_printf(s, "--------\t------\n");
1809 	list_for_each_entry(chan, &pl08x->memcpy.channels, vc.chan.device_node) {
1810 		seq_printf(s, "%s\t\t%s\n", chan->name,
1811 			   pl08x_state_str(chan->state));
1812 	}
1813 
1814 	seq_printf(s, "\nPL08x virtual slave channels:\n");
1815 	seq_printf(s, "CHANNEL:\tSTATE:\n");
1816 	seq_printf(s, "--------\t------\n");
1817 	list_for_each_entry(chan, &pl08x->slave.channels, vc.chan.device_node) {
1818 		seq_printf(s, "%s\t\t%s\n", chan->name,
1819 			   pl08x_state_str(chan->state));
1820 	}
1821 
1822 	return 0;
1823 }
1824 
1825 static int pl08x_debugfs_open(struct inode *inode, struct file *file)
1826 {
1827 	return single_open(file, pl08x_debugfs_show, inode->i_private);
1828 }
1829 
1830 static const struct file_operations pl08x_debugfs_operations = {
1831 	.open		= pl08x_debugfs_open,
1832 	.read		= seq_read,
1833 	.llseek		= seq_lseek,
1834 	.release	= single_release,
1835 };
1836 
1837 static void init_pl08x_debugfs(struct pl08x_driver_data *pl08x)
1838 {
1839 	/* Expose a simple debugfs interface to view all clocks */
1840 	(void) debugfs_create_file(dev_name(&pl08x->adev->dev),
1841 			S_IFREG | S_IRUGO, NULL, pl08x,
1842 			&pl08x_debugfs_operations);
1843 }
1844 
1845 #else
1846 static inline void init_pl08x_debugfs(struct pl08x_driver_data *pl08x)
1847 {
1848 }
1849 #endif
1850 
1851 static int pl08x_probe(struct amba_device *adev, const struct amba_id *id)
1852 {
1853 	struct pl08x_driver_data *pl08x;
1854 	const struct vendor_data *vd = id->data;
1855 	int ret = 0;
1856 	int i;
1857 
1858 	ret = amba_request_regions(adev, NULL);
1859 	if (ret)
1860 		return ret;
1861 
1862 	/* Create the driver state holder */
1863 	pl08x = kzalloc(sizeof(*pl08x), GFP_KERNEL);
1864 	if (!pl08x) {
1865 		ret = -ENOMEM;
1866 		goto out_no_pl08x;
1867 	}
1868 
1869 	/* Initialize memcpy engine */
1870 	dma_cap_set(DMA_MEMCPY, pl08x->memcpy.cap_mask);
1871 	pl08x->memcpy.dev = &adev->dev;
1872 	pl08x->memcpy.device_alloc_chan_resources = pl08x_alloc_chan_resources;
1873 	pl08x->memcpy.device_free_chan_resources = pl08x_free_chan_resources;
1874 	pl08x->memcpy.device_prep_dma_memcpy = pl08x_prep_dma_memcpy;
1875 	pl08x->memcpy.device_prep_dma_interrupt = pl08x_prep_dma_interrupt;
1876 	pl08x->memcpy.device_tx_status = pl08x_dma_tx_status;
1877 	pl08x->memcpy.device_issue_pending = pl08x_issue_pending;
1878 	pl08x->memcpy.device_control = pl08x_control;
1879 
1880 	/* Initialize slave engine */
1881 	dma_cap_set(DMA_SLAVE, pl08x->slave.cap_mask);
1882 	pl08x->slave.dev = &adev->dev;
1883 	pl08x->slave.device_alloc_chan_resources = pl08x_alloc_chan_resources;
1884 	pl08x->slave.device_free_chan_resources = pl08x_free_chan_resources;
1885 	pl08x->slave.device_prep_dma_interrupt = pl08x_prep_dma_interrupt;
1886 	pl08x->slave.device_tx_status = pl08x_dma_tx_status;
1887 	pl08x->slave.device_issue_pending = pl08x_issue_pending;
1888 	pl08x->slave.device_prep_slave_sg = pl08x_prep_slave_sg;
1889 	pl08x->slave.device_control = pl08x_control;
1890 
1891 	/* Get the platform data */
1892 	pl08x->pd = dev_get_platdata(&adev->dev);
1893 	if (!pl08x->pd) {
1894 		dev_err(&adev->dev, "no platform data supplied\n");
1895 		goto out_no_platdata;
1896 	}
1897 
1898 	/* Assign useful pointers to the driver state */
1899 	pl08x->adev = adev;
1900 	pl08x->vd = vd;
1901 
1902 	/* By default, AHB1 only.  If dualmaster, from platform */
1903 	pl08x->lli_buses = PL08X_AHB1;
1904 	pl08x->mem_buses = PL08X_AHB1;
1905 	if (pl08x->vd->dualmaster) {
1906 		pl08x->lli_buses = pl08x->pd->lli_buses;
1907 		pl08x->mem_buses = pl08x->pd->mem_buses;
1908 	}
1909 
1910 	/* A DMA memory pool for LLIs, align on 1-byte boundary */
1911 	pl08x->pool = dma_pool_create(DRIVER_NAME, &pl08x->adev->dev,
1912 			PL08X_LLI_TSFR_SIZE, PL08X_ALIGN, 0);
1913 	if (!pl08x->pool) {
1914 		ret = -ENOMEM;
1915 		goto out_no_lli_pool;
1916 	}
1917 
1918 	pl08x->base = ioremap(adev->res.start, resource_size(&adev->res));
1919 	if (!pl08x->base) {
1920 		ret = -ENOMEM;
1921 		goto out_no_ioremap;
1922 	}
1923 
1924 	/* Turn on the PL08x */
1925 	pl08x_ensure_on(pl08x);
1926 
1927 	/* Attach the interrupt handler */
1928 	writel(0x000000FF, pl08x->base + PL080_ERR_CLEAR);
1929 	writel(0x000000FF, pl08x->base + PL080_TC_CLEAR);
1930 
1931 	ret = request_irq(adev->irq[0], pl08x_irq, IRQF_DISABLED,
1932 			  DRIVER_NAME, pl08x);
1933 	if (ret) {
1934 		dev_err(&adev->dev, "%s failed to request interrupt %d\n",
1935 			__func__, adev->irq[0]);
1936 		goto out_no_irq;
1937 	}
1938 
1939 	/* Initialize physical channels */
1940 	pl08x->phy_chans = kzalloc((vd->channels * sizeof(*pl08x->phy_chans)),
1941 			GFP_KERNEL);
1942 	if (!pl08x->phy_chans) {
1943 		dev_err(&adev->dev, "%s failed to allocate "
1944 			"physical channel holders\n",
1945 			__func__);
1946 		goto out_no_phychans;
1947 	}
1948 
1949 	for (i = 0; i < vd->channels; i++) {
1950 		struct pl08x_phy_chan *ch = &pl08x->phy_chans[i];
1951 
1952 		ch->id = i;
1953 		ch->base = pl08x->base + PL080_Cx_BASE(i);
1954 		spin_lock_init(&ch->lock);
1955 
1956 		/*
1957 		 * Nomadik variants can have channels that are locked
1958 		 * down for the secure world only. Lock up these channels
1959 		 * by perpetually serving a dummy virtual channel.
1960 		 */
1961 		if (vd->nomadik) {
1962 			u32 val;
1963 
1964 			val = readl(ch->base + PL080_CH_CONFIG);
1965 			if (val & (PL080N_CONFIG_ITPROT | PL080N_CONFIG_SECPROT)) {
1966 				dev_info(&adev->dev, "physical channel %d reserved for secure access only\n", i);
1967 				ch->locked = true;
1968 			}
1969 		}
1970 
1971 		dev_dbg(&adev->dev, "physical channel %d is %s\n",
1972 			i, pl08x_phy_channel_busy(ch) ? "BUSY" : "FREE");
1973 	}
1974 
1975 	/* Register as many memcpy channels as there are physical channels */
1976 	ret = pl08x_dma_init_virtual_channels(pl08x, &pl08x->memcpy,
1977 					      pl08x->vd->channels, false);
1978 	if (ret <= 0) {
1979 		dev_warn(&pl08x->adev->dev,
1980 			 "%s failed to enumerate memcpy channels - %d\n",
1981 			 __func__, ret);
1982 		goto out_no_memcpy;
1983 	}
1984 	pl08x->memcpy.chancnt = ret;
1985 
1986 	/* Register slave channels */
1987 	ret = pl08x_dma_init_virtual_channels(pl08x, &pl08x->slave,
1988 			pl08x->pd->num_slave_channels, true);
1989 	if (ret <= 0) {
1990 		dev_warn(&pl08x->adev->dev,
1991 			"%s failed to enumerate slave channels - %d\n",
1992 				__func__, ret);
1993 		goto out_no_slave;
1994 	}
1995 	pl08x->slave.chancnt = ret;
1996 
1997 	ret = dma_async_device_register(&pl08x->memcpy);
1998 	if (ret) {
1999 		dev_warn(&pl08x->adev->dev,
2000 			"%s failed to register memcpy as an async device - %d\n",
2001 			__func__, ret);
2002 		goto out_no_memcpy_reg;
2003 	}
2004 
2005 	ret = dma_async_device_register(&pl08x->slave);
2006 	if (ret) {
2007 		dev_warn(&pl08x->adev->dev,
2008 			"%s failed to register slave as an async device - %d\n",
2009 			__func__, ret);
2010 		goto out_no_slave_reg;
2011 	}
2012 
2013 	amba_set_drvdata(adev, pl08x);
2014 	init_pl08x_debugfs(pl08x);
2015 	dev_info(&pl08x->adev->dev, "DMA: PL%03x rev%u at 0x%08llx irq %d\n",
2016 		 amba_part(adev), amba_rev(adev),
2017 		 (unsigned long long)adev->res.start, adev->irq[0]);
2018 
2019 	return 0;
2020 
2021 out_no_slave_reg:
2022 	dma_async_device_unregister(&pl08x->memcpy);
2023 out_no_memcpy_reg:
2024 	pl08x_free_virtual_channels(&pl08x->slave);
2025 out_no_slave:
2026 	pl08x_free_virtual_channels(&pl08x->memcpy);
2027 out_no_memcpy:
2028 	kfree(pl08x->phy_chans);
2029 out_no_phychans:
2030 	free_irq(adev->irq[0], pl08x);
2031 out_no_irq:
2032 	iounmap(pl08x->base);
2033 out_no_ioremap:
2034 	dma_pool_destroy(pl08x->pool);
2035 out_no_lli_pool:
2036 out_no_platdata:
2037 	kfree(pl08x);
2038 out_no_pl08x:
2039 	amba_release_regions(adev);
2040 	return ret;
2041 }
2042 
2043 /* PL080 has 8 channels and the PL080 have just 2 */
2044 static struct vendor_data vendor_pl080 = {
2045 	.channels = 8,
2046 	.dualmaster = true,
2047 };
2048 
2049 static struct vendor_data vendor_nomadik = {
2050 	.channels = 8,
2051 	.dualmaster = true,
2052 	.nomadik = true,
2053 };
2054 
2055 static struct vendor_data vendor_pl081 = {
2056 	.channels = 2,
2057 	.dualmaster = false,
2058 };
2059 
2060 static struct amba_id pl08x_ids[] = {
2061 	/* PL080 */
2062 	{
2063 		.id	= 0x00041080,
2064 		.mask	= 0x000fffff,
2065 		.data	= &vendor_pl080,
2066 	},
2067 	/* PL081 */
2068 	{
2069 		.id	= 0x00041081,
2070 		.mask	= 0x000fffff,
2071 		.data	= &vendor_pl081,
2072 	},
2073 	/* Nomadik 8815 PL080 variant */
2074 	{
2075 		.id	= 0x00280080,
2076 		.mask	= 0x00ffffff,
2077 		.data	= &vendor_nomadik,
2078 	},
2079 	{ 0, 0 },
2080 };
2081 
2082 MODULE_DEVICE_TABLE(amba, pl08x_ids);
2083 
2084 static struct amba_driver pl08x_amba_driver = {
2085 	.drv.name	= DRIVER_NAME,
2086 	.id_table	= pl08x_ids,
2087 	.probe		= pl08x_probe,
2088 };
2089 
2090 static int __init pl08x_init(void)
2091 {
2092 	int retval;
2093 	retval = amba_driver_register(&pl08x_amba_driver);
2094 	if (retval)
2095 		printk(KERN_WARNING DRIVER_NAME
2096 		       "failed to register as an AMBA device (%d)\n",
2097 		       retval);
2098 	return retval;
2099 }
2100 subsys_initcall(pl08x_init);
2101