xref: /openbmc/linux/drivers/dma/ste_dma40.c (revision 8fdff1dc)
1 /*
2  * Copyright (C) Ericsson AB 2007-2008
3  * Copyright (C) ST-Ericsson SA 2008-2010
4  * Author: Per Forlin <per.forlin@stericsson.com> for ST-Ericsson
5  * Author: Jonas Aaberg <jonas.aberg@stericsson.com> for ST-Ericsson
6  * License terms: GNU General Public License (GPL) version 2
7  */
8 
9 #include <linux/dma-mapping.h>
10 #include <linux/kernel.h>
11 #include <linux/slab.h>
12 #include <linux/export.h>
13 #include <linux/dmaengine.h>
14 #include <linux/platform_device.h>
15 #include <linux/clk.h>
16 #include <linux/delay.h>
17 #include <linux/pm.h>
18 #include <linux/pm_runtime.h>
19 #include <linux/err.h>
20 #include <linux/amba/bus.h>
21 #include <linux/regulator/consumer.h>
22 #include <linux/platform_data/dma-ste-dma40.h>
23 
24 #include "dmaengine.h"
25 #include "ste_dma40_ll.h"
26 
27 #define D40_NAME "dma40"
28 
29 #define D40_PHY_CHAN -1
30 
31 /* For masking out/in 2 bit channel positions */
32 #define D40_CHAN_POS(chan)  (2 * (chan / 2))
33 #define D40_CHAN_POS_MASK(chan) (0x3 << D40_CHAN_POS(chan))
34 
35 /* Maximum iterations taken before giving up suspending a channel */
36 #define D40_SUSPEND_MAX_IT 500
37 
38 /* Milliseconds */
39 #define DMA40_AUTOSUSPEND_DELAY	100
40 
41 /* Hardware requirement on LCLA alignment */
42 #define LCLA_ALIGNMENT 0x40000
43 
44 /* Max number of links per event group */
45 #define D40_LCLA_LINK_PER_EVENT_GRP 128
46 #define D40_LCLA_END D40_LCLA_LINK_PER_EVENT_GRP
47 
48 /* Attempts before giving up to trying to get pages that are aligned */
49 #define MAX_LCLA_ALLOC_ATTEMPTS 256
50 
51 /* Bit markings for allocation map */
52 #define D40_ALLOC_FREE		(1 << 31)
53 #define D40_ALLOC_PHY		(1 << 30)
54 #define D40_ALLOC_LOG_FREE	0
55 
56 /**
57  * enum 40_command - The different commands and/or statuses.
58  *
59  * @D40_DMA_STOP: DMA channel command STOP or status STOPPED,
60  * @D40_DMA_RUN: The DMA channel is RUNNING of the command RUN.
61  * @D40_DMA_SUSPEND_REQ: Request the DMA to SUSPEND as soon as possible.
62  * @D40_DMA_SUSPENDED: The DMA channel is SUSPENDED.
63  */
64 enum d40_command {
65 	D40_DMA_STOP		= 0,
66 	D40_DMA_RUN		= 1,
67 	D40_DMA_SUSPEND_REQ	= 2,
68 	D40_DMA_SUSPENDED	= 3
69 };
70 
71 /*
72  * enum d40_events - The different Event Enables for the event lines.
73  *
74  * @D40_DEACTIVATE_EVENTLINE: De-activate Event line, stopping the logical chan.
75  * @D40_ACTIVATE_EVENTLINE: Activate the Event line, to start a logical chan.
76  * @D40_SUSPEND_REQ_EVENTLINE: Requesting for suspending a event line.
77  * @D40_ROUND_EVENTLINE: Status check for event line.
78  */
79 
80 enum d40_events {
81 	D40_DEACTIVATE_EVENTLINE	= 0,
82 	D40_ACTIVATE_EVENTLINE		= 1,
83 	D40_SUSPEND_REQ_EVENTLINE	= 2,
84 	D40_ROUND_EVENTLINE		= 3
85 };
86 
87 /*
88  * These are the registers that has to be saved and later restored
89  * when the DMA hw is powered off.
90  * TODO: Add save/restore of D40_DREG_GCC on dma40 v3 or later, if that works.
91  */
92 static u32 d40_backup_regs[] = {
93 	D40_DREG_LCPA,
94 	D40_DREG_LCLA,
95 	D40_DREG_PRMSE,
96 	D40_DREG_PRMSO,
97 	D40_DREG_PRMOE,
98 	D40_DREG_PRMOO,
99 };
100 
101 #define BACKUP_REGS_SZ ARRAY_SIZE(d40_backup_regs)
102 
103 /* TODO: Check if all these registers have to be saved/restored on dma40 v3 */
104 static u32 d40_backup_regs_v3[] = {
105 	D40_DREG_PSEG1,
106 	D40_DREG_PSEG2,
107 	D40_DREG_PSEG3,
108 	D40_DREG_PSEG4,
109 	D40_DREG_PCEG1,
110 	D40_DREG_PCEG2,
111 	D40_DREG_PCEG3,
112 	D40_DREG_PCEG4,
113 	D40_DREG_RSEG1,
114 	D40_DREG_RSEG2,
115 	D40_DREG_RSEG3,
116 	D40_DREG_RSEG4,
117 	D40_DREG_RCEG1,
118 	D40_DREG_RCEG2,
119 	D40_DREG_RCEG3,
120 	D40_DREG_RCEG4,
121 };
122 
123 #define BACKUP_REGS_SZ_V3 ARRAY_SIZE(d40_backup_regs_v3)
124 
125 static u32 d40_backup_regs_chan[] = {
126 	D40_CHAN_REG_SSCFG,
127 	D40_CHAN_REG_SSELT,
128 	D40_CHAN_REG_SSPTR,
129 	D40_CHAN_REG_SSLNK,
130 	D40_CHAN_REG_SDCFG,
131 	D40_CHAN_REG_SDELT,
132 	D40_CHAN_REG_SDPTR,
133 	D40_CHAN_REG_SDLNK,
134 };
135 
136 /**
137  * struct d40_lli_pool - Structure for keeping LLIs in memory
138  *
139  * @base: Pointer to memory area when the pre_alloc_lli's are not large
140  * enough, IE bigger than the most common case, 1 dst and 1 src. NULL if
141  * pre_alloc_lli is used.
142  * @dma_addr: DMA address, if mapped
143  * @size: The size in bytes of the memory at base or the size of pre_alloc_lli.
144  * @pre_alloc_lli: Pre allocated area for the most common case of transfers,
145  * one buffer to one buffer.
146  */
147 struct d40_lli_pool {
148 	void	*base;
149 	int	 size;
150 	dma_addr_t	dma_addr;
151 	/* Space for dst and src, plus an extra for padding */
152 	u8	 pre_alloc_lli[3 * sizeof(struct d40_phy_lli)];
153 };
154 
155 /**
156  * struct d40_desc - A descriptor is one DMA job.
157  *
158  * @lli_phy: LLI settings for physical channel. Both src and dst=
159  * points into the lli_pool, to base if lli_len > 1 or to pre_alloc_lli if
160  * lli_len equals one.
161  * @lli_log: Same as above but for logical channels.
162  * @lli_pool: The pool with two entries pre-allocated.
163  * @lli_len: Number of llis of current descriptor.
164  * @lli_current: Number of transferred llis.
165  * @lcla_alloc: Number of LCLA entries allocated.
166  * @txd: DMA engine struct. Used for among other things for communication
167  * during a transfer.
168  * @node: List entry.
169  * @is_in_client_list: true if the client owns this descriptor.
170  * @cyclic: true if this is a cyclic job
171  *
172  * This descriptor is used for both logical and physical transfers.
173  */
174 struct d40_desc {
175 	/* LLI physical */
176 	struct d40_phy_lli_bidir	 lli_phy;
177 	/* LLI logical */
178 	struct d40_log_lli_bidir	 lli_log;
179 
180 	struct d40_lli_pool		 lli_pool;
181 	int				 lli_len;
182 	int				 lli_current;
183 	int				 lcla_alloc;
184 
185 	struct dma_async_tx_descriptor	 txd;
186 	struct list_head		 node;
187 
188 	bool				 is_in_client_list;
189 	bool				 cyclic;
190 };
191 
192 /**
193  * struct d40_lcla_pool - LCLA pool settings and data.
194  *
195  * @base: The virtual address of LCLA. 18 bit aligned.
196  * @base_unaligned: The orignal kmalloc pointer, if kmalloc is used.
197  * This pointer is only there for clean-up on error.
198  * @pages: The number of pages needed for all physical channels.
199  * Only used later for clean-up on error
200  * @lock: Lock to protect the content in this struct.
201  * @alloc_map: big map over which LCLA entry is own by which job.
202  */
203 struct d40_lcla_pool {
204 	void		*base;
205 	dma_addr_t	dma_addr;
206 	void		*base_unaligned;
207 	int		 pages;
208 	spinlock_t	 lock;
209 	struct d40_desc	**alloc_map;
210 };
211 
212 /**
213  * struct d40_phy_res - struct for handling eventlines mapped to physical
214  * channels.
215  *
216  * @lock: A lock protection this entity.
217  * @reserved: True if used by secure world or otherwise.
218  * @num: The physical channel number of this entity.
219  * @allocated_src: Bit mapped to show which src event line's are mapped to
220  * this physical channel. Can also be free or physically allocated.
221  * @allocated_dst: Same as for src but is dst.
222  * allocated_dst and allocated_src uses the D40_ALLOC* defines as well as
223  * event line number.
224  */
225 struct d40_phy_res {
226 	spinlock_t lock;
227 	bool	   reserved;
228 	int	   num;
229 	u32	   allocated_src;
230 	u32	   allocated_dst;
231 };
232 
233 struct d40_base;
234 
235 /**
236  * struct d40_chan - Struct that describes a channel.
237  *
238  * @lock: A spinlock to protect this struct.
239  * @log_num: The logical number, if any of this channel.
240  * @pending_tx: The number of pending transfers. Used between interrupt handler
241  * and tasklet.
242  * @busy: Set to true when transfer is ongoing on this channel.
243  * @phy_chan: Pointer to physical channel which this instance runs on. If this
244  * point is NULL, then the channel is not allocated.
245  * @chan: DMA engine handle.
246  * @tasklet: Tasklet that gets scheduled from interrupt context to complete a
247  * transfer and call client callback.
248  * @client: Cliented owned descriptor list.
249  * @pending_queue: Submitted jobs, to be issued by issue_pending()
250  * @active: Active descriptor.
251  * @queue: Queued jobs.
252  * @prepare_queue: Prepared jobs.
253  * @dma_cfg: The client configuration of this dma channel.
254  * @configured: whether the dma_cfg configuration is valid
255  * @base: Pointer to the device instance struct.
256  * @src_def_cfg: Default cfg register setting for src.
257  * @dst_def_cfg: Default cfg register setting for dst.
258  * @log_def: Default logical channel settings.
259  * @lcpa: Pointer to dst and src lcpa settings.
260  * @runtime_addr: runtime configured address.
261  * @runtime_direction: runtime configured direction.
262  *
263  * This struct can either "be" a logical or a physical channel.
264  */
265 struct d40_chan {
266 	spinlock_t			 lock;
267 	int				 log_num;
268 	int				 pending_tx;
269 	bool				 busy;
270 	struct d40_phy_res		*phy_chan;
271 	struct dma_chan			 chan;
272 	struct tasklet_struct		 tasklet;
273 	struct list_head		 client;
274 	struct list_head		 pending_queue;
275 	struct list_head		 active;
276 	struct list_head		 queue;
277 	struct list_head		 prepare_queue;
278 	struct stedma40_chan_cfg	 dma_cfg;
279 	bool				 configured;
280 	struct d40_base			*base;
281 	/* Default register configurations */
282 	u32				 src_def_cfg;
283 	u32				 dst_def_cfg;
284 	struct d40_def_lcsp		 log_def;
285 	struct d40_log_lli_full		*lcpa;
286 	/* Runtime reconfiguration */
287 	dma_addr_t			runtime_addr;
288 	enum dma_transfer_direction	runtime_direction;
289 };
290 
291 /**
292  * struct d40_base - The big global struct, one for each probe'd instance.
293  *
294  * @interrupt_lock: Lock used to make sure one interrupt is handle a time.
295  * @execmd_lock: Lock for execute command usage since several channels share
296  * the same physical register.
297  * @dev: The device structure.
298  * @virtbase: The virtual base address of the DMA's register.
299  * @rev: silicon revision detected.
300  * @clk: Pointer to the DMA clock structure.
301  * @phy_start: Physical memory start of the DMA registers.
302  * @phy_size: Size of the DMA register map.
303  * @irq: The IRQ number.
304  * @num_phy_chans: The number of physical channels. Read from HW. This
305  * is the number of available channels for this driver, not counting "Secure
306  * mode" allocated physical channels.
307  * @num_log_chans: The number of logical channels. Calculated from
308  * num_phy_chans.
309  * @dma_both: dma_device channels that can do both memcpy and slave transfers.
310  * @dma_slave: dma_device channels that can do only do slave transfers.
311  * @dma_memcpy: dma_device channels that can do only do memcpy transfers.
312  * @phy_chans: Room for all possible physical channels in system.
313  * @log_chans: Room for all possible logical channels in system.
314  * @lookup_log_chans: Used to map interrupt number to logical channel. Points
315  * to log_chans entries.
316  * @lookup_phy_chans: Used to map interrupt number to physical channel. Points
317  * to phy_chans entries.
318  * @plat_data: Pointer to provided platform_data which is the driver
319  * configuration.
320  * @lcpa_regulator: Pointer to hold the regulator for the esram bank for lcla.
321  * @phy_res: Vector containing all physical channels.
322  * @lcla_pool: lcla pool settings and data.
323  * @lcpa_base: The virtual mapped address of LCPA.
324  * @phy_lcpa: The physical address of the LCPA.
325  * @lcpa_size: The size of the LCPA area.
326  * @desc_slab: cache for descriptors.
327  * @reg_val_backup: Here the values of some hardware registers are stored
328  * before the DMA is powered off. They are restored when the power is back on.
329  * @reg_val_backup_v3: Backup of registers that only exits on dma40 v3 and
330  * later.
331  * @reg_val_backup_chan: Backup data for standard channel parameter registers.
332  * @gcc_pwr_off_mask: Mask to maintain the channels that can be turned off.
333  * @initialized: true if the dma has been initialized
334  */
335 struct d40_base {
336 	spinlock_t			 interrupt_lock;
337 	spinlock_t			 execmd_lock;
338 	struct device			 *dev;
339 	void __iomem			 *virtbase;
340 	u8				  rev:4;
341 	struct clk			 *clk;
342 	phys_addr_t			  phy_start;
343 	resource_size_t			  phy_size;
344 	int				  irq;
345 	int				  num_phy_chans;
346 	int				  num_log_chans;
347 	struct dma_device		  dma_both;
348 	struct dma_device		  dma_slave;
349 	struct dma_device		  dma_memcpy;
350 	struct d40_chan			 *phy_chans;
351 	struct d40_chan			 *log_chans;
352 	struct d40_chan			**lookup_log_chans;
353 	struct d40_chan			**lookup_phy_chans;
354 	struct stedma40_platform_data	 *plat_data;
355 	struct regulator		 *lcpa_regulator;
356 	/* Physical half channels */
357 	struct d40_phy_res		 *phy_res;
358 	struct d40_lcla_pool		  lcla_pool;
359 	void				 *lcpa_base;
360 	dma_addr_t			  phy_lcpa;
361 	resource_size_t			  lcpa_size;
362 	struct kmem_cache		 *desc_slab;
363 	u32				  reg_val_backup[BACKUP_REGS_SZ];
364 	u32				  reg_val_backup_v3[BACKUP_REGS_SZ_V3];
365 	u32				 *reg_val_backup_chan;
366 	u16				  gcc_pwr_off_mask;
367 	bool				  initialized;
368 };
369 
370 /**
371  * struct d40_interrupt_lookup - lookup table for interrupt handler
372  *
373  * @src: Interrupt mask register.
374  * @clr: Interrupt clear register.
375  * @is_error: true if this is an error interrupt.
376  * @offset: start delta in the lookup_log_chans in d40_base. If equals to
377  * D40_PHY_CHAN, the lookup_phy_chans shall be used instead.
378  */
379 struct d40_interrupt_lookup {
380 	u32 src;
381 	u32 clr;
382 	bool is_error;
383 	int offset;
384 };
385 
386 /**
387  * struct d40_reg_val - simple lookup struct
388  *
389  * @reg: The register.
390  * @val: The value that belongs to the register in reg.
391  */
392 struct d40_reg_val {
393 	unsigned int reg;
394 	unsigned int val;
395 };
396 
397 static struct device *chan2dev(struct d40_chan *d40c)
398 {
399 	return &d40c->chan.dev->device;
400 }
401 
402 static bool chan_is_physical(struct d40_chan *chan)
403 {
404 	return chan->log_num == D40_PHY_CHAN;
405 }
406 
407 static bool chan_is_logical(struct d40_chan *chan)
408 {
409 	return !chan_is_physical(chan);
410 }
411 
412 static void __iomem *chan_base(struct d40_chan *chan)
413 {
414 	return chan->base->virtbase + D40_DREG_PCBASE +
415 	       chan->phy_chan->num * D40_DREG_PCDELTA;
416 }
417 
418 #define d40_err(dev, format, arg...)		\
419 	dev_err(dev, "[%s] " format, __func__, ## arg)
420 
421 #define chan_err(d40c, format, arg...)		\
422 	d40_err(chan2dev(d40c), format, ## arg)
423 
424 static int d40_pool_lli_alloc(struct d40_chan *d40c, struct d40_desc *d40d,
425 			      int lli_len)
426 {
427 	bool is_log = chan_is_logical(d40c);
428 	u32 align;
429 	void *base;
430 
431 	if (is_log)
432 		align = sizeof(struct d40_log_lli);
433 	else
434 		align = sizeof(struct d40_phy_lli);
435 
436 	if (lli_len == 1) {
437 		base = d40d->lli_pool.pre_alloc_lli;
438 		d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
439 		d40d->lli_pool.base = NULL;
440 	} else {
441 		d40d->lli_pool.size = lli_len * 2 * align;
442 
443 		base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
444 		d40d->lli_pool.base = base;
445 
446 		if (d40d->lli_pool.base == NULL)
447 			return -ENOMEM;
448 	}
449 
450 	if (is_log) {
451 		d40d->lli_log.src = PTR_ALIGN(base, align);
452 		d40d->lli_log.dst = d40d->lli_log.src + lli_len;
453 
454 		d40d->lli_pool.dma_addr = 0;
455 	} else {
456 		d40d->lli_phy.src = PTR_ALIGN(base, align);
457 		d40d->lli_phy.dst = d40d->lli_phy.src + lli_len;
458 
459 		d40d->lli_pool.dma_addr = dma_map_single(d40c->base->dev,
460 							 d40d->lli_phy.src,
461 							 d40d->lli_pool.size,
462 							 DMA_TO_DEVICE);
463 
464 		if (dma_mapping_error(d40c->base->dev,
465 				      d40d->lli_pool.dma_addr)) {
466 			kfree(d40d->lli_pool.base);
467 			d40d->lli_pool.base = NULL;
468 			d40d->lli_pool.dma_addr = 0;
469 			return -ENOMEM;
470 		}
471 	}
472 
473 	return 0;
474 }
475 
476 static void d40_pool_lli_free(struct d40_chan *d40c, struct d40_desc *d40d)
477 {
478 	if (d40d->lli_pool.dma_addr)
479 		dma_unmap_single(d40c->base->dev, d40d->lli_pool.dma_addr,
480 				 d40d->lli_pool.size, DMA_TO_DEVICE);
481 
482 	kfree(d40d->lli_pool.base);
483 	d40d->lli_pool.base = NULL;
484 	d40d->lli_pool.size = 0;
485 	d40d->lli_log.src = NULL;
486 	d40d->lli_log.dst = NULL;
487 	d40d->lli_phy.src = NULL;
488 	d40d->lli_phy.dst = NULL;
489 }
490 
491 static int d40_lcla_alloc_one(struct d40_chan *d40c,
492 			      struct d40_desc *d40d)
493 {
494 	unsigned long flags;
495 	int i;
496 	int ret = -EINVAL;
497 	int p;
498 
499 	spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
500 
501 	p = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP;
502 
503 	/*
504 	 * Allocate both src and dst at the same time, therefore the half
505 	 * start on 1 since 0 can't be used since zero is used as end marker.
506 	 */
507 	for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
508 		if (!d40c->base->lcla_pool.alloc_map[p + i]) {
509 			d40c->base->lcla_pool.alloc_map[p + i] = d40d;
510 			d40d->lcla_alloc++;
511 			ret = i;
512 			break;
513 		}
514 	}
515 
516 	spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
517 
518 	return ret;
519 }
520 
521 static int d40_lcla_free_all(struct d40_chan *d40c,
522 			     struct d40_desc *d40d)
523 {
524 	unsigned long flags;
525 	int i;
526 	int ret = -EINVAL;
527 
528 	if (chan_is_physical(d40c))
529 		return 0;
530 
531 	spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
532 
533 	for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
534 		if (d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num *
535 						    D40_LCLA_LINK_PER_EVENT_GRP + i] == d40d) {
536 			d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num *
537 							D40_LCLA_LINK_PER_EVENT_GRP + i] = NULL;
538 			d40d->lcla_alloc--;
539 			if (d40d->lcla_alloc == 0) {
540 				ret = 0;
541 				break;
542 			}
543 		}
544 	}
545 
546 	spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
547 
548 	return ret;
549 
550 }
551 
552 static void d40_desc_remove(struct d40_desc *d40d)
553 {
554 	list_del(&d40d->node);
555 }
556 
557 static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
558 {
559 	struct d40_desc *desc = NULL;
560 
561 	if (!list_empty(&d40c->client)) {
562 		struct d40_desc *d;
563 		struct d40_desc *_d;
564 
565 		list_for_each_entry_safe(d, _d, &d40c->client, node) {
566 			if (async_tx_test_ack(&d->txd)) {
567 				d40_desc_remove(d);
568 				desc = d;
569 				memset(desc, 0, sizeof(*desc));
570 				break;
571 			}
572 		}
573 	}
574 
575 	if (!desc)
576 		desc = kmem_cache_zalloc(d40c->base->desc_slab, GFP_NOWAIT);
577 
578 	if (desc)
579 		INIT_LIST_HEAD(&desc->node);
580 
581 	return desc;
582 }
583 
584 static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
585 {
586 
587 	d40_pool_lli_free(d40c, d40d);
588 	d40_lcla_free_all(d40c, d40d);
589 	kmem_cache_free(d40c->base->desc_slab, d40d);
590 }
591 
592 static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
593 {
594 	list_add_tail(&desc->node, &d40c->active);
595 }
596 
597 static void d40_phy_lli_load(struct d40_chan *chan, struct d40_desc *desc)
598 {
599 	struct d40_phy_lli *lli_dst = desc->lli_phy.dst;
600 	struct d40_phy_lli *lli_src = desc->lli_phy.src;
601 	void __iomem *base = chan_base(chan);
602 
603 	writel(lli_src->reg_cfg, base + D40_CHAN_REG_SSCFG);
604 	writel(lli_src->reg_elt, base + D40_CHAN_REG_SSELT);
605 	writel(lli_src->reg_ptr, base + D40_CHAN_REG_SSPTR);
606 	writel(lli_src->reg_lnk, base + D40_CHAN_REG_SSLNK);
607 
608 	writel(lli_dst->reg_cfg, base + D40_CHAN_REG_SDCFG);
609 	writel(lli_dst->reg_elt, base + D40_CHAN_REG_SDELT);
610 	writel(lli_dst->reg_ptr, base + D40_CHAN_REG_SDPTR);
611 	writel(lli_dst->reg_lnk, base + D40_CHAN_REG_SDLNK);
612 }
613 
614 static void d40_log_lli_to_lcxa(struct d40_chan *chan, struct d40_desc *desc)
615 {
616 	struct d40_lcla_pool *pool = &chan->base->lcla_pool;
617 	struct d40_log_lli_bidir *lli = &desc->lli_log;
618 	int lli_current = desc->lli_current;
619 	int lli_len = desc->lli_len;
620 	bool cyclic = desc->cyclic;
621 	int curr_lcla = -EINVAL;
622 	int first_lcla = 0;
623 	bool use_esram_lcla = chan->base->plat_data->use_esram_lcla;
624 	bool linkback;
625 
626 	/*
627 	 * We may have partially running cyclic transfers, in case we did't get
628 	 * enough LCLA entries.
629 	 */
630 	linkback = cyclic && lli_current == 0;
631 
632 	/*
633 	 * For linkback, we need one LCLA even with only one link, because we
634 	 * can't link back to the one in LCPA space
635 	 */
636 	if (linkback || (lli_len - lli_current > 1)) {
637 		curr_lcla = d40_lcla_alloc_one(chan, desc);
638 		first_lcla = curr_lcla;
639 	}
640 
641 	/*
642 	 * For linkback, we normally load the LCPA in the loop since we need to
643 	 * link it to the second LCLA and not the first.  However, if we
644 	 * couldn't even get a first LCLA, then we have to run in LCPA and
645 	 * reload manually.
646 	 */
647 	if (!linkback || curr_lcla == -EINVAL) {
648 		unsigned int flags = 0;
649 
650 		if (curr_lcla == -EINVAL)
651 			flags |= LLI_TERM_INT;
652 
653 		d40_log_lli_lcpa_write(chan->lcpa,
654 				       &lli->dst[lli_current],
655 				       &lli->src[lli_current],
656 				       curr_lcla,
657 				       flags);
658 		lli_current++;
659 	}
660 
661 	if (curr_lcla < 0)
662 		goto out;
663 
664 	for (; lli_current < lli_len; lli_current++) {
665 		unsigned int lcla_offset = chan->phy_chan->num * 1024 +
666 					   8 * curr_lcla * 2;
667 		struct d40_log_lli *lcla = pool->base + lcla_offset;
668 		unsigned int flags = 0;
669 		int next_lcla;
670 
671 		if (lli_current + 1 < lli_len)
672 			next_lcla = d40_lcla_alloc_one(chan, desc);
673 		else
674 			next_lcla = linkback ? first_lcla : -EINVAL;
675 
676 		if (cyclic || next_lcla == -EINVAL)
677 			flags |= LLI_TERM_INT;
678 
679 		if (linkback && curr_lcla == first_lcla) {
680 			/* First link goes in both LCPA and LCLA */
681 			d40_log_lli_lcpa_write(chan->lcpa,
682 					       &lli->dst[lli_current],
683 					       &lli->src[lli_current],
684 					       next_lcla, flags);
685 		}
686 
687 		/*
688 		 * One unused LCLA in the cyclic case if the very first
689 		 * next_lcla fails...
690 		 */
691 		d40_log_lli_lcla_write(lcla,
692 				       &lli->dst[lli_current],
693 				       &lli->src[lli_current],
694 				       next_lcla, flags);
695 
696 		/*
697 		 * Cache maintenance is not needed if lcla is
698 		 * mapped in esram
699 		 */
700 		if (!use_esram_lcla) {
701 			dma_sync_single_range_for_device(chan->base->dev,
702 						pool->dma_addr, lcla_offset,
703 						2 * sizeof(struct d40_log_lli),
704 						DMA_TO_DEVICE);
705 		}
706 		curr_lcla = next_lcla;
707 
708 		if (curr_lcla == -EINVAL || curr_lcla == first_lcla) {
709 			lli_current++;
710 			break;
711 		}
712 	}
713 
714 out:
715 	desc->lli_current = lli_current;
716 }
717 
718 static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
719 {
720 	if (chan_is_physical(d40c)) {
721 		d40_phy_lli_load(d40c, d40d);
722 		d40d->lli_current = d40d->lli_len;
723 	} else
724 		d40_log_lli_to_lcxa(d40c, d40d);
725 }
726 
727 static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
728 {
729 	struct d40_desc *d;
730 
731 	if (list_empty(&d40c->active))
732 		return NULL;
733 
734 	d = list_first_entry(&d40c->active,
735 			     struct d40_desc,
736 			     node);
737 	return d;
738 }
739 
740 /* remove desc from current queue and add it to the pending_queue */
741 static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
742 {
743 	d40_desc_remove(desc);
744 	desc->is_in_client_list = false;
745 	list_add_tail(&desc->node, &d40c->pending_queue);
746 }
747 
748 static struct d40_desc *d40_first_pending(struct d40_chan *d40c)
749 {
750 	struct d40_desc *d;
751 
752 	if (list_empty(&d40c->pending_queue))
753 		return NULL;
754 
755 	d = list_first_entry(&d40c->pending_queue,
756 			     struct d40_desc,
757 			     node);
758 	return d;
759 }
760 
761 static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
762 {
763 	struct d40_desc *d;
764 
765 	if (list_empty(&d40c->queue))
766 		return NULL;
767 
768 	d = list_first_entry(&d40c->queue,
769 			     struct d40_desc,
770 			     node);
771 	return d;
772 }
773 
774 static int d40_psize_2_burst_size(bool is_log, int psize)
775 {
776 	if (is_log) {
777 		if (psize == STEDMA40_PSIZE_LOG_1)
778 			return 1;
779 	} else {
780 		if (psize == STEDMA40_PSIZE_PHY_1)
781 			return 1;
782 	}
783 
784 	return 2 << psize;
785 }
786 
787 /*
788  * The dma only supports transmitting packages up to
789  * STEDMA40_MAX_SEG_SIZE << data_width. Calculate the total number of
790  * dma elements required to send the entire sg list
791  */
792 static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2)
793 {
794 	int dmalen;
795 	u32 max_w = max(data_width1, data_width2);
796 	u32 min_w = min(data_width1, data_width2);
797 	u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE << min_w, 1 << max_w);
798 
799 	if (seg_max > STEDMA40_MAX_SEG_SIZE)
800 		seg_max -= (1 << max_w);
801 
802 	if (!IS_ALIGNED(size, 1 << max_w))
803 		return -EINVAL;
804 
805 	if (size <= seg_max)
806 		dmalen = 1;
807 	else {
808 		dmalen = size / seg_max;
809 		if (dmalen * seg_max < size)
810 			dmalen++;
811 	}
812 	return dmalen;
813 }
814 
815 static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len,
816 			   u32 data_width1, u32 data_width2)
817 {
818 	struct scatterlist *sg;
819 	int i;
820 	int len = 0;
821 	int ret;
822 
823 	for_each_sg(sgl, sg, sg_len, i) {
824 		ret = d40_size_2_dmalen(sg_dma_len(sg),
825 					data_width1, data_width2);
826 		if (ret < 0)
827 			return ret;
828 		len += ret;
829 	}
830 	return len;
831 }
832 
833 
834 #ifdef CONFIG_PM
835 static void dma40_backup(void __iomem *baseaddr, u32 *backup,
836 			 u32 *regaddr, int num, bool save)
837 {
838 	int i;
839 
840 	for (i = 0; i < num; i++) {
841 		void __iomem *addr = baseaddr + regaddr[i];
842 
843 		if (save)
844 			backup[i] = readl_relaxed(addr);
845 		else
846 			writel_relaxed(backup[i], addr);
847 	}
848 }
849 
850 static void d40_save_restore_registers(struct d40_base *base, bool save)
851 {
852 	int i;
853 
854 	/* Save/Restore channel specific registers */
855 	for (i = 0; i < base->num_phy_chans; i++) {
856 		void __iomem *addr;
857 		int idx;
858 
859 		if (base->phy_res[i].reserved)
860 			continue;
861 
862 		addr = base->virtbase + D40_DREG_PCBASE + i * D40_DREG_PCDELTA;
863 		idx = i * ARRAY_SIZE(d40_backup_regs_chan);
864 
865 		dma40_backup(addr, &base->reg_val_backup_chan[idx],
866 			     d40_backup_regs_chan,
867 			     ARRAY_SIZE(d40_backup_regs_chan),
868 			     save);
869 	}
870 
871 	/* Save/Restore global registers */
872 	dma40_backup(base->virtbase, base->reg_val_backup,
873 		     d40_backup_regs, ARRAY_SIZE(d40_backup_regs),
874 		     save);
875 
876 	/* Save/Restore registers only existing on dma40 v3 and later */
877 	if (base->rev >= 3)
878 		dma40_backup(base->virtbase, base->reg_val_backup_v3,
879 			     d40_backup_regs_v3,
880 			     ARRAY_SIZE(d40_backup_regs_v3),
881 			     save);
882 }
883 #else
884 static void d40_save_restore_registers(struct d40_base *base, bool save)
885 {
886 }
887 #endif
888 
889 static int __d40_execute_command_phy(struct d40_chan *d40c,
890 				     enum d40_command command)
891 {
892 	u32 status;
893 	int i;
894 	void __iomem *active_reg;
895 	int ret = 0;
896 	unsigned long flags;
897 	u32 wmask;
898 
899 	if (command == D40_DMA_STOP) {
900 		ret = __d40_execute_command_phy(d40c, D40_DMA_SUSPEND_REQ);
901 		if (ret)
902 			return ret;
903 	}
904 
905 	spin_lock_irqsave(&d40c->base->execmd_lock, flags);
906 
907 	if (d40c->phy_chan->num % 2 == 0)
908 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
909 	else
910 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
911 
912 	if (command == D40_DMA_SUSPEND_REQ) {
913 		status = (readl(active_reg) &
914 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
915 			D40_CHAN_POS(d40c->phy_chan->num);
916 
917 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
918 			goto done;
919 	}
920 
921 	wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
922 	writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
923 	       active_reg);
924 
925 	if (command == D40_DMA_SUSPEND_REQ) {
926 
927 		for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
928 			status = (readl(active_reg) &
929 				  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
930 				D40_CHAN_POS(d40c->phy_chan->num);
931 
932 			cpu_relax();
933 			/*
934 			 * Reduce the number of bus accesses while
935 			 * waiting for the DMA to suspend.
936 			 */
937 			udelay(3);
938 
939 			if (status == D40_DMA_STOP ||
940 			    status == D40_DMA_SUSPENDED)
941 				break;
942 		}
943 
944 		if (i == D40_SUSPEND_MAX_IT) {
945 			chan_err(d40c,
946 				"unable to suspend the chl %d (log: %d) status %x\n",
947 				d40c->phy_chan->num, d40c->log_num,
948 				status);
949 			dump_stack();
950 			ret = -EBUSY;
951 		}
952 
953 	}
954 done:
955 	spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
956 	return ret;
957 }
958 
959 static void d40_term_all(struct d40_chan *d40c)
960 {
961 	struct d40_desc *d40d;
962 	struct d40_desc *_d;
963 
964 	/* Release active descriptors */
965 	while ((d40d = d40_first_active_get(d40c))) {
966 		d40_desc_remove(d40d);
967 		d40_desc_free(d40c, d40d);
968 	}
969 
970 	/* Release queued descriptors waiting for transfer */
971 	while ((d40d = d40_first_queued(d40c))) {
972 		d40_desc_remove(d40d);
973 		d40_desc_free(d40c, d40d);
974 	}
975 
976 	/* Release pending descriptors */
977 	while ((d40d = d40_first_pending(d40c))) {
978 		d40_desc_remove(d40d);
979 		d40_desc_free(d40c, d40d);
980 	}
981 
982 	/* Release client owned descriptors */
983 	if (!list_empty(&d40c->client))
984 		list_for_each_entry_safe(d40d, _d, &d40c->client, node) {
985 			d40_desc_remove(d40d);
986 			d40_desc_free(d40c, d40d);
987 		}
988 
989 	/* Release descriptors in prepare queue */
990 	if (!list_empty(&d40c->prepare_queue))
991 		list_for_each_entry_safe(d40d, _d,
992 					 &d40c->prepare_queue, node) {
993 			d40_desc_remove(d40d);
994 			d40_desc_free(d40c, d40d);
995 		}
996 
997 	d40c->pending_tx = 0;
998 }
999 
1000 static void __d40_config_set_event(struct d40_chan *d40c,
1001 				   enum d40_events event_type, u32 event,
1002 				   int reg)
1003 {
1004 	void __iomem *addr = chan_base(d40c) + reg;
1005 	int tries;
1006 	u32 status;
1007 
1008 	switch (event_type) {
1009 
1010 	case D40_DEACTIVATE_EVENTLINE:
1011 
1012 		writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
1013 		       | ~D40_EVENTLINE_MASK(event), addr);
1014 		break;
1015 
1016 	case D40_SUSPEND_REQ_EVENTLINE:
1017 		status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1018 			  D40_EVENTLINE_POS(event);
1019 
1020 		if (status == D40_DEACTIVATE_EVENTLINE ||
1021 		    status == D40_SUSPEND_REQ_EVENTLINE)
1022 			break;
1023 
1024 		writel((D40_SUSPEND_REQ_EVENTLINE << D40_EVENTLINE_POS(event))
1025 		       | ~D40_EVENTLINE_MASK(event), addr);
1026 
1027 		for (tries = 0 ; tries < D40_SUSPEND_MAX_IT; tries++) {
1028 
1029 			status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1030 				  D40_EVENTLINE_POS(event);
1031 
1032 			cpu_relax();
1033 			/*
1034 			 * Reduce the number of bus accesses while
1035 			 * waiting for the DMA to suspend.
1036 			 */
1037 			udelay(3);
1038 
1039 			if (status == D40_DEACTIVATE_EVENTLINE)
1040 				break;
1041 		}
1042 
1043 		if (tries == D40_SUSPEND_MAX_IT) {
1044 			chan_err(d40c,
1045 				"unable to stop the event_line chl %d (log: %d)"
1046 				"status %x\n", d40c->phy_chan->num,
1047 				 d40c->log_num, status);
1048 		}
1049 		break;
1050 
1051 	case D40_ACTIVATE_EVENTLINE:
1052 	/*
1053 	 * The hardware sometimes doesn't register the enable when src and dst
1054 	 * event lines are active on the same logical channel.  Retry to ensure
1055 	 * it does.  Usually only one retry is sufficient.
1056 	 */
1057 		tries = 100;
1058 		while (--tries) {
1059 			writel((D40_ACTIVATE_EVENTLINE <<
1060 				D40_EVENTLINE_POS(event)) |
1061 				~D40_EVENTLINE_MASK(event), addr);
1062 
1063 			if (readl(addr) & D40_EVENTLINE_MASK(event))
1064 				break;
1065 		}
1066 
1067 		if (tries != 99)
1068 			dev_dbg(chan2dev(d40c),
1069 				"[%s] workaround enable S%cLNK (%d tries)\n",
1070 				__func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D',
1071 				100 - tries);
1072 
1073 		WARN_ON(!tries);
1074 		break;
1075 
1076 	case D40_ROUND_EVENTLINE:
1077 		BUG();
1078 		break;
1079 
1080 	}
1081 }
1082 
1083 static void d40_config_set_event(struct d40_chan *d40c,
1084 				 enum d40_events event_type)
1085 {
1086 	/* Enable event line connected to device (or memcpy) */
1087 	if ((d40c->dma_cfg.dir ==  STEDMA40_PERIPH_TO_MEM) ||
1088 	    (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH)) {
1089 		u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
1090 
1091 		__d40_config_set_event(d40c, event_type, event,
1092 				       D40_CHAN_REG_SSLNK);
1093 	}
1094 
1095 	if (d40c->dma_cfg.dir !=  STEDMA40_PERIPH_TO_MEM) {
1096 		u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
1097 
1098 		__d40_config_set_event(d40c, event_type, event,
1099 				       D40_CHAN_REG_SDLNK);
1100 	}
1101 }
1102 
1103 static u32 d40_chan_has_events(struct d40_chan *d40c)
1104 {
1105 	void __iomem *chanbase = chan_base(d40c);
1106 	u32 val;
1107 
1108 	val = readl(chanbase + D40_CHAN_REG_SSLNK);
1109 	val |= readl(chanbase + D40_CHAN_REG_SDLNK);
1110 
1111 	return val;
1112 }
1113 
1114 static int
1115 __d40_execute_command_log(struct d40_chan *d40c, enum d40_command command)
1116 {
1117 	unsigned long flags;
1118 	int ret = 0;
1119 	u32 active_status;
1120 	void __iomem *active_reg;
1121 
1122 	if (d40c->phy_chan->num % 2 == 0)
1123 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1124 	else
1125 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1126 
1127 
1128 	spin_lock_irqsave(&d40c->phy_chan->lock, flags);
1129 
1130 	switch (command) {
1131 	case D40_DMA_STOP:
1132 	case D40_DMA_SUSPEND_REQ:
1133 
1134 		active_status = (readl(active_reg) &
1135 				 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1136 				 D40_CHAN_POS(d40c->phy_chan->num);
1137 
1138 		if (active_status == D40_DMA_RUN)
1139 			d40_config_set_event(d40c, D40_SUSPEND_REQ_EVENTLINE);
1140 		else
1141 			d40_config_set_event(d40c, D40_DEACTIVATE_EVENTLINE);
1142 
1143 		if (!d40_chan_has_events(d40c) && (command == D40_DMA_STOP))
1144 			ret = __d40_execute_command_phy(d40c, command);
1145 
1146 		break;
1147 
1148 	case D40_DMA_RUN:
1149 
1150 		d40_config_set_event(d40c, D40_ACTIVATE_EVENTLINE);
1151 		ret = __d40_execute_command_phy(d40c, command);
1152 		break;
1153 
1154 	case D40_DMA_SUSPENDED:
1155 		BUG();
1156 		break;
1157 	}
1158 
1159 	spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
1160 	return ret;
1161 }
1162 
1163 static int d40_channel_execute_command(struct d40_chan *d40c,
1164 				       enum d40_command command)
1165 {
1166 	if (chan_is_logical(d40c))
1167 		return __d40_execute_command_log(d40c, command);
1168 	else
1169 		return __d40_execute_command_phy(d40c, command);
1170 }
1171 
1172 static u32 d40_get_prmo(struct d40_chan *d40c)
1173 {
1174 	static const unsigned int phy_map[] = {
1175 		[STEDMA40_PCHAN_BASIC_MODE]
1176 			= D40_DREG_PRMO_PCHAN_BASIC,
1177 		[STEDMA40_PCHAN_MODULO_MODE]
1178 			= D40_DREG_PRMO_PCHAN_MODULO,
1179 		[STEDMA40_PCHAN_DOUBLE_DST_MODE]
1180 			= D40_DREG_PRMO_PCHAN_DOUBLE_DST,
1181 	};
1182 	static const unsigned int log_map[] = {
1183 		[STEDMA40_LCHAN_SRC_PHY_DST_LOG]
1184 			= D40_DREG_PRMO_LCHAN_SRC_PHY_DST_LOG,
1185 		[STEDMA40_LCHAN_SRC_LOG_DST_PHY]
1186 			= D40_DREG_PRMO_LCHAN_SRC_LOG_DST_PHY,
1187 		[STEDMA40_LCHAN_SRC_LOG_DST_LOG]
1188 			= D40_DREG_PRMO_LCHAN_SRC_LOG_DST_LOG,
1189 	};
1190 
1191 	if (chan_is_physical(d40c))
1192 		return phy_map[d40c->dma_cfg.mode_opt];
1193 	else
1194 		return log_map[d40c->dma_cfg.mode_opt];
1195 }
1196 
1197 static void d40_config_write(struct d40_chan *d40c)
1198 {
1199 	u32 addr_base;
1200 	u32 var;
1201 
1202 	/* Odd addresses are even addresses + 4 */
1203 	addr_base = (d40c->phy_chan->num % 2) * 4;
1204 	/* Setup channel mode to logical or physical */
1205 	var = ((u32)(chan_is_logical(d40c)) + 1) <<
1206 		D40_CHAN_POS(d40c->phy_chan->num);
1207 	writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
1208 
1209 	/* Setup operational mode option register */
1210 	var = d40_get_prmo(d40c) << D40_CHAN_POS(d40c->phy_chan->num);
1211 
1212 	writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
1213 
1214 	if (chan_is_logical(d40c)) {
1215 		int lidx = (d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS)
1216 			   & D40_SREG_ELEM_LOG_LIDX_MASK;
1217 		void __iomem *chanbase = chan_base(d40c);
1218 
1219 		/* Set default config for CFG reg */
1220 		writel(d40c->src_def_cfg, chanbase + D40_CHAN_REG_SSCFG);
1221 		writel(d40c->dst_def_cfg, chanbase + D40_CHAN_REG_SDCFG);
1222 
1223 		/* Set LIDX for lcla */
1224 		writel(lidx, chanbase + D40_CHAN_REG_SSELT);
1225 		writel(lidx, chanbase + D40_CHAN_REG_SDELT);
1226 
1227 		/* Clear LNK which will be used by d40_chan_has_events() */
1228 		writel(0, chanbase + D40_CHAN_REG_SSLNK);
1229 		writel(0, chanbase + D40_CHAN_REG_SDLNK);
1230 	}
1231 }
1232 
1233 static u32 d40_residue(struct d40_chan *d40c)
1234 {
1235 	u32 num_elt;
1236 
1237 	if (chan_is_logical(d40c))
1238 		num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
1239 			>> D40_MEM_LCSP2_ECNT_POS;
1240 	else {
1241 		u32 val = readl(chan_base(d40c) + D40_CHAN_REG_SDELT);
1242 		num_elt = (val & D40_SREG_ELEM_PHY_ECNT_MASK)
1243 			  >> D40_SREG_ELEM_PHY_ECNT_POS;
1244 	}
1245 
1246 	return num_elt * (1 << d40c->dma_cfg.dst_info.data_width);
1247 }
1248 
1249 static bool d40_tx_is_linked(struct d40_chan *d40c)
1250 {
1251 	bool is_link;
1252 
1253 	if (chan_is_logical(d40c))
1254 		is_link = readl(&d40c->lcpa->lcsp3) &  D40_MEM_LCSP3_DLOS_MASK;
1255 	else
1256 		is_link = readl(chan_base(d40c) + D40_CHAN_REG_SDLNK)
1257 			  & D40_SREG_LNK_PHYS_LNK_MASK;
1258 
1259 	return is_link;
1260 }
1261 
1262 static int d40_pause(struct d40_chan *d40c)
1263 {
1264 	int res = 0;
1265 	unsigned long flags;
1266 
1267 	if (!d40c->busy)
1268 		return 0;
1269 
1270 	pm_runtime_get_sync(d40c->base->dev);
1271 	spin_lock_irqsave(&d40c->lock, flags);
1272 
1273 	res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1274 
1275 	pm_runtime_mark_last_busy(d40c->base->dev);
1276 	pm_runtime_put_autosuspend(d40c->base->dev);
1277 	spin_unlock_irqrestore(&d40c->lock, flags);
1278 	return res;
1279 }
1280 
1281 static int d40_resume(struct d40_chan *d40c)
1282 {
1283 	int res = 0;
1284 	unsigned long flags;
1285 
1286 	if (!d40c->busy)
1287 		return 0;
1288 
1289 	spin_lock_irqsave(&d40c->lock, flags);
1290 	pm_runtime_get_sync(d40c->base->dev);
1291 
1292 	/* If bytes left to transfer or linked tx resume job */
1293 	if (d40_residue(d40c) || d40_tx_is_linked(d40c))
1294 		res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1295 
1296 	pm_runtime_mark_last_busy(d40c->base->dev);
1297 	pm_runtime_put_autosuspend(d40c->base->dev);
1298 	spin_unlock_irqrestore(&d40c->lock, flags);
1299 	return res;
1300 }
1301 
1302 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
1303 {
1304 	struct d40_chan *d40c = container_of(tx->chan,
1305 					     struct d40_chan,
1306 					     chan);
1307 	struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
1308 	unsigned long flags;
1309 	dma_cookie_t cookie;
1310 
1311 	spin_lock_irqsave(&d40c->lock, flags);
1312 	cookie = dma_cookie_assign(tx);
1313 	d40_desc_queue(d40c, d40d);
1314 	spin_unlock_irqrestore(&d40c->lock, flags);
1315 
1316 	return cookie;
1317 }
1318 
1319 static int d40_start(struct d40_chan *d40c)
1320 {
1321 	return d40_channel_execute_command(d40c, D40_DMA_RUN);
1322 }
1323 
1324 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
1325 {
1326 	struct d40_desc *d40d;
1327 	int err;
1328 
1329 	/* Start queued jobs, if any */
1330 	d40d = d40_first_queued(d40c);
1331 
1332 	if (d40d != NULL) {
1333 		if (!d40c->busy) {
1334 			d40c->busy = true;
1335 			pm_runtime_get_sync(d40c->base->dev);
1336 		}
1337 
1338 		/* Remove from queue */
1339 		d40_desc_remove(d40d);
1340 
1341 		/* Add to active queue */
1342 		d40_desc_submit(d40c, d40d);
1343 
1344 		/* Initiate DMA job */
1345 		d40_desc_load(d40c, d40d);
1346 
1347 		/* Start dma job */
1348 		err = d40_start(d40c);
1349 
1350 		if (err)
1351 			return NULL;
1352 	}
1353 
1354 	return d40d;
1355 }
1356 
1357 /* called from interrupt context */
1358 static void dma_tc_handle(struct d40_chan *d40c)
1359 {
1360 	struct d40_desc *d40d;
1361 
1362 	/* Get first active entry from list */
1363 	d40d = d40_first_active_get(d40c);
1364 
1365 	if (d40d == NULL)
1366 		return;
1367 
1368 	if (d40d->cyclic) {
1369 		/*
1370 		 * If this was a paritially loaded list, we need to reloaded
1371 		 * it, and only when the list is completed.  We need to check
1372 		 * for done because the interrupt will hit for every link, and
1373 		 * not just the last one.
1374 		 */
1375 		if (d40d->lli_current < d40d->lli_len
1376 		    && !d40_tx_is_linked(d40c)
1377 		    && !d40_residue(d40c)) {
1378 			d40_lcla_free_all(d40c, d40d);
1379 			d40_desc_load(d40c, d40d);
1380 			(void) d40_start(d40c);
1381 
1382 			if (d40d->lli_current == d40d->lli_len)
1383 				d40d->lli_current = 0;
1384 		}
1385 	} else {
1386 		d40_lcla_free_all(d40c, d40d);
1387 
1388 		if (d40d->lli_current < d40d->lli_len) {
1389 			d40_desc_load(d40c, d40d);
1390 			/* Start dma job */
1391 			(void) d40_start(d40c);
1392 			return;
1393 		}
1394 
1395 		if (d40_queue_start(d40c) == NULL)
1396 			d40c->busy = false;
1397 		pm_runtime_mark_last_busy(d40c->base->dev);
1398 		pm_runtime_put_autosuspend(d40c->base->dev);
1399 	}
1400 
1401 	d40c->pending_tx++;
1402 	tasklet_schedule(&d40c->tasklet);
1403 
1404 }
1405 
1406 static void dma_tasklet(unsigned long data)
1407 {
1408 	struct d40_chan *d40c = (struct d40_chan *) data;
1409 	struct d40_desc *d40d;
1410 	unsigned long flags;
1411 	dma_async_tx_callback callback;
1412 	void *callback_param;
1413 
1414 	spin_lock_irqsave(&d40c->lock, flags);
1415 
1416 	/* Get first active entry from list */
1417 	d40d = d40_first_active_get(d40c);
1418 	if (d40d == NULL)
1419 		goto err;
1420 
1421 	if (!d40d->cyclic)
1422 		dma_cookie_complete(&d40d->txd);
1423 
1424 	/*
1425 	 * If terminating a channel pending_tx is set to zero.
1426 	 * This prevents any finished active jobs to return to the client.
1427 	 */
1428 	if (d40c->pending_tx == 0) {
1429 		spin_unlock_irqrestore(&d40c->lock, flags);
1430 		return;
1431 	}
1432 
1433 	/* Callback to client */
1434 	callback = d40d->txd.callback;
1435 	callback_param = d40d->txd.callback_param;
1436 
1437 	if (!d40d->cyclic) {
1438 		if (async_tx_test_ack(&d40d->txd)) {
1439 			d40_desc_remove(d40d);
1440 			d40_desc_free(d40c, d40d);
1441 		} else {
1442 			if (!d40d->is_in_client_list) {
1443 				d40_desc_remove(d40d);
1444 				d40_lcla_free_all(d40c, d40d);
1445 				list_add_tail(&d40d->node, &d40c->client);
1446 				d40d->is_in_client_list = true;
1447 			}
1448 		}
1449 	}
1450 
1451 	d40c->pending_tx--;
1452 
1453 	if (d40c->pending_tx)
1454 		tasklet_schedule(&d40c->tasklet);
1455 
1456 	spin_unlock_irqrestore(&d40c->lock, flags);
1457 
1458 	if (callback && (d40d->txd.flags & DMA_PREP_INTERRUPT))
1459 		callback(callback_param);
1460 
1461 	return;
1462 
1463 err:
1464 	/* Rescue manouver if receiving double interrupts */
1465 	if (d40c->pending_tx > 0)
1466 		d40c->pending_tx--;
1467 	spin_unlock_irqrestore(&d40c->lock, flags);
1468 }
1469 
1470 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1471 {
1472 	static const struct d40_interrupt_lookup il[] = {
1473 		{D40_DREG_LCTIS0, D40_DREG_LCICR0, false,  0},
1474 		{D40_DREG_LCTIS1, D40_DREG_LCICR1, false, 32},
1475 		{D40_DREG_LCTIS2, D40_DREG_LCICR2, false, 64},
1476 		{D40_DREG_LCTIS3, D40_DREG_LCICR3, false, 96},
1477 		{D40_DREG_LCEIS0, D40_DREG_LCICR0, true,   0},
1478 		{D40_DREG_LCEIS1, D40_DREG_LCICR1, true,  32},
1479 		{D40_DREG_LCEIS2, D40_DREG_LCICR2, true,  64},
1480 		{D40_DREG_LCEIS3, D40_DREG_LCICR3, true,  96},
1481 		{D40_DREG_PCTIS,  D40_DREG_PCICR,  false, D40_PHY_CHAN},
1482 		{D40_DREG_PCEIS,  D40_DREG_PCICR,  true,  D40_PHY_CHAN},
1483 	};
1484 
1485 	int i;
1486 	u32 regs[ARRAY_SIZE(il)];
1487 	u32 idx;
1488 	u32 row;
1489 	long chan = -1;
1490 	struct d40_chan *d40c;
1491 	unsigned long flags;
1492 	struct d40_base *base = data;
1493 
1494 	spin_lock_irqsave(&base->interrupt_lock, flags);
1495 
1496 	/* Read interrupt status of both logical and physical channels */
1497 	for (i = 0; i < ARRAY_SIZE(il); i++)
1498 		regs[i] = readl(base->virtbase + il[i].src);
1499 
1500 	for (;;) {
1501 
1502 		chan = find_next_bit((unsigned long *)regs,
1503 				     BITS_PER_LONG * ARRAY_SIZE(il), chan + 1);
1504 
1505 		/* No more set bits found? */
1506 		if (chan == BITS_PER_LONG * ARRAY_SIZE(il))
1507 			break;
1508 
1509 		row = chan / BITS_PER_LONG;
1510 		idx = chan & (BITS_PER_LONG - 1);
1511 
1512 		/* ACK interrupt */
1513 		writel(1 << idx, base->virtbase + il[row].clr);
1514 
1515 		if (il[row].offset == D40_PHY_CHAN)
1516 			d40c = base->lookup_phy_chans[idx];
1517 		else
1518 			d40c = base->lookup_log_chans[il[row].offset + idx];
1519 		spin_lock(&d40c->lock);
1520 
1521 		if (!il[row].is_error)
1522 			dma_tc_handle(d40c);
1523 		else
1524 			d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1525 				chan, il[row].offset, idx);
1526 
1527 		spin_unlock(&d40c->lock);
1528 	}
1529 
1530 	spin_unlock_irqrestore(&base->interrupt_lock, flags);
1531 
1532 	return IRQ_HANDLED;
1533 }
1534 
1535 static int d40_validate_conf(struct d40_chan *d40c,
1536 			     struct stedma40_chan_cfg *conf)
1537 {
1538 	int res = 0;
1539 	u32 dst_event_group = D40_TYPE_TO_GROUP(conf->dst_dev_type);
1540 	u32 src_event_group = D40_TYPE_TO_GROUP(conf->src_dev_type);
1541 	bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1542 
1543 	if (!conf->dir) {
1544 		chan_err(d40c, "Invalid direction.\n");
1545 		res = -EINVAL;
1546 	}
1547 
1548 	if (conf->dst_dev_type != STEDMA40_DEV_DST_MEMORY &&
1549 	    d40c->base->plat_data->dev_tx[conf->dst_dev_type] == 0 &&
1550 	    d40c->runtime_addr == 0) {
1551 
1552 		chan_err(d40c, "Invalid TX channel address (%d)\n",
1553 			 conf->dst_dev_type);
1554 		res = -EINVAL;
1555 	}
1556 
1557 	if (conf->src_dev_type != STEDMA40_DEV_SRC_MEMORY &&
1558 	    d40c->base->plat_data->dev_rx[conf->src_dev_type] == 0 &&
1559 	    d40c->runtime_addr == 0) {
1560 		chan_err(d40c, "Invalid RX channel address (%d)\n",
1561 			conf->src_dev_type);
1562 		res = -EINVAL;
1563 	}
1564 
1565 	if (conf->dir == STEDMA40_MEM_TO_PERIPH &&
1566 	    dst_event_group == STEDMA40_DEV_DST_MEMORY) {
1567 		chan_err(d40c, "Invalid dst\n");
1568 		res = -EINVAL;
1569 	}
1570 
1571 	if (conf->dir == STEDMA40_PERIPH_TO_MEM &&
1572 	    src_event_group == STEDMA40_DEV_SRC_MEMORY) {
1573 		chan_err(d40c, "Invalid src\n");
1574 		res = -EINVAL;
1575 	}
1576 
1577 	if (src_event_group == STEDMA40_DEV_SRC_MEMORY &&
1578 	    dst_event_group == STEDMA40_DEV_DST_MEMORY && is_log) {
1579 		chan_err(d40c, "No event line\n");
1580 		res = -EINVAL;
1581 	}
1582 
1583 	if (conf->dir == STEDMA40_PERIPH_TO_PERIPH &&
1584 	    (src_event_group != dst_event_group)) {
1585 		chan_err(d40c, "Invalid event group\n");
1586 		res = -EINVAL;
1587 	}
1588 
1589 	if (conf->dir == STEDMA40_PERIPH_TO_PERIPH) {
1590 		/*
1591 		 * DMAC HW supports it. Will be added to this driver,
1592 		 * in case any dma client requires it.
1593 		 */
1594 		chan_err(d40c, "periph to periph not supported\n");
1595 		res = -EINVAL;
1596 	}
1597 
1598 	if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1599 	    (1 << conf->src_info.data_width) !=
1600 	    d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1601 	    (1 << conf->dst_info.data_width)) {
1602 		/*
1603 		 * The DMAC hardware only supports
1604 		 * src (burst x width) == dst (burst x width)
1605 		 */
1606 
1607 		chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1608 		res = -EINVAL;
1609 	}
1610 
1611 	return res;
1612 }
1613 
1614 static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1615 			       bool is_src, int log_event_line, bool is_log,
1616 			       bool *first_user)
1617 {
1618 	unsigned long flags;
1619 	spin_lock_irqsave(&phy->lock, flags);
1620 
1621 	*first_user = ((phy->allocated_src | phy->allocated_dst)
1622 			== D40_ALLOC_FREE);
1623 
1624 	if (!is_log) {
1625 		/* Physical interrupts are masked per physical full channel */
1626 		if (phy->allocated_src == D40_ALLOC_FREE &&
1627 		    phy->allocated_dst == D40_ALLOC_FREE) {
1628 			phy->allocated_dst = D40_ALLOC_PHY;
1629 			phy->allocated_src = D40_ALLOC_PHY;
1630 			goto found;
1631 		} else
1632 			goto not_found;
1633 	}
1634 
1635 	/* Logical channel */
1636 	if (is_src) {
1637 		if (phy->allocated_src == D40_ALLOC_PHY)
1638 			goto not_found;
1639 
1640 		if (phy->allocated_src == D40_ALLOC_FREE)
1641 			phy->allocated_src = D40_ALLOC_LOG_FREE;
1642 
1643 		if (!(phy->allocated_src & (1 << log_event_line))) {
1644 			phy->allocated_src |= 1 << log_event_line;
1645 			goto found;
1646 		} else
1647 			goto not_found;
1648 	} else {
1649 		if (phy->allocated_dst == D40_ALLOC_PHY)
1650 			goto not_found;
1651 
1652 		if (phy->allocated_dst == D40_ALLOC_FREE)
1653 			phy->allocated_dst = D40_ALLOC_LOG_FREE;
1654 
1655 		if (!(phy->allocated_dst & (1 << log_event_line))) {
1656 			phy->allocated_dst |= 1 << log_event_line;
1657 			goto found;
1658 		} else
1659 			goto not_found;
1660 	}
1661 
1662 not_found:
1663 	spin_unlock_irqrestore(&phy->lock, flags);
1664 	return false;
1665 found:
1666 	spin_unlock_irqrestore(&phy->lock, flags);
1667 	return true;
1668 }
1669 
1670 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1671 			       int log_event_line)
1672 {
1673 	unsigned long flags;
1674 	bool is_free = false;
1675 
1676 	spin_lock_irqsave(&phy->lock, flags);
1677 	if (!log_event_line) {
1678 		phy->allocated_dst = D40_ALLOC_FREE;
1679 		phy->allocated_src = D40_ALLOC_FREE;
1680 		is_free = true;
1681 		goto out;
1682 	}
1683 
1684 	/* Logical channel */
1685 	if (is_src) {
1686 		phy->allocated_src &= ~(1 << log_event_line);
1687 		if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1688 			phy->allocated_src = D40_ALLOC_FREE;
1689 	} else {
1690 		phy->allocated_dst &= ~(1 << log_event_line);
1691 		if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1692 			phy->allocated_dst = D40_ALLOC_FREE;
1693 	}
1694 
1695 	is_free = ((phy->allocated_src | phy->allocated_dst) ==
1696 		   D40_ALLOC_FREE);
1697 
1698 out:
1699 	spin_unlock_irqrestore(&phy->lock, flags);
1700 
1701 	return is_free;
1702 }
1703 
1704 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1705 {
1706 	int dev_type;
1707 	int event_group;
1708 	int event_line;
1709 	struct d40_phy_res *phys;
1710 	int i;
1711 	int j;
1712 	int log_num;
1713 	bool is_src;
1714 	bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1715 
1716 	phys = d40c->base->phy_res;
1717 
1718 	if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1719 		dev_type = d40c->dma_cfg.src_dev_type;
1720 		log_num = 2 * dev_type;
1721 		is_src = true;
1722 	} else if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1723 		   d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1724 		/* dst event lines are used for logical memcpy */
1725 		dev_type = d40c->dma_cfg.dst_dev_type;
1726 		log_num = 2 * dev_type + 1;
1727 		is_src = false;
1728 	} else
1729 		return -EINVAL;
1730 
1731 	event_group = D40_TYPE_TO_GROUP(dev_type);
1732 	event_line = D40_TYPE_TO_EVENT(dev_type);
1733 
1734 	if (!is_log) {
1735 		if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1736 			/* Find physical half channel */
1737 			for (i = 0; i < d40c->base->num_phy_chans; i++) {
1738 
1739 				if (d40_alloc_mask_set(&phys[i], is_src,
1740 						       0, is_log,
1741 						       first_phy_user))
1742 					goto found_phy;
1743 			}
1744 		} else
1745 			for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1746 				int phy_num = j  + event_group * 2;
1747 				for (i = phy_num; i < phy_num + 2; i++) {
1748 					if (d40_alloc_mask_set(&phys[i],
1749 							       is_src,
1750 							       0,
1751 							       is_log,
1752 							       first_phy_user))
1753 						goto found_phy;
1754 				}
1755 			}
1756 		return -EINVAL;
1757 found_phy:
1758 		d40c->phy_chan = &phys[i];
1759 		d40c->log_num = D40_PHY_CHAN;
1760 		goto out;
1761 	}
1762 	if (dev_type == -1)
1763 		return -EINVAL;
1764 
1765 	/* Find logical channel */
1766 	for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1767 		int phy_num = j + event_group * 2;
1768 
1769 		if (d40c->dma_cfg.use_fixed_channel) {
1770 			i = d40c->dma_cfg.phy_channel;
1771 
1772 			if ((i != phy_num) && (i != phy_num + 1)) {
1773 				dev_err(chan2dev(d40c),
1774 					"invalid fixed phy channel %d\n", i);
1775 				return -EINVAL;
1776 			}
1777 
1778 			if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1779 					       is_log, first_phy_user))
1780 				goto found_log;
1781 
1782 			dev_err(chan2dev(d40c),
1783 				"could not allocate fixed phy channel %d\n", i);
1784 			return -EINVAL;
1785 		}
1786 
1787 		/*
1788 		 * Spread logical channels across all available physical rather
1789 		 * than pack every logical channel at the first available phy
1790 		 * channels.
1791 		 */
1792 		if (is_src) {
1793 			for (i = phy_num; i < phy_num + 2; i++) {
1794 				if (d40_alloc_mask_set(&phys[i], is_src,
1795 						       event_line, is_log,
1796 						       first_phy_user))
1797 					goto found_log;
1798 			}
1799 		} else {
1800 			for (i = phy_num + 1; i >= phy_num; i--) {
1801 				if (d40_alloc_mask_set(&phys[i], is_src,
1802 						       event_line, is_log,
1803 						       first_phy_user))
1804 					goto found_log;
1805 			}
1806 		}
1807 	}
1808 	return -EINVAL;
1809 
1810 found_log:
1811 	d40c->phy_chan = &phys[i];
1812 	d40c->log_num = log_num;
1813 out:
1814 
1815 	if (is_log)
1816 		d40c->base->lookup_log_chans[d40c->log_num] = d40c;
1817 	else
1818 		d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
1819 
1820 	return 0;
1821 
1822 }
1823 
1824 static int d40_config_memcpy(struct d40_chan *d40c)
1825 {
1826 	dma_cap_mask_t cap = d40c->chan.device->cap_mask;
1827 
1828 	if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
1829 		d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_log;
1830 		d40c->dma_cfg.src_dev_type = STEDMA40_DEV_SRC_MEMORY;
1831 		d40c->dma_cfg.dst_dev_type = d40c->base->plat_data->
1832 			memcpy[d40c->chan.chan_id];
1833 
1834 	} else if (dma_has_cap(DMA_MEMCPY, cap) &&
1835 		   dma_has_cap(DMA_SLAVE, cap)) {
1836 		d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_phy;
1837 	} else {
1838 		chan_err(d40c, "No memcpy\n");
1839 		return -EINVAL;
1840 	}
1841 
1842 	return 0;
1843 }
1844 
1845 static int d40_free_dma(struct d40_chan *d40c)
1846 {
1847 
1848 	int res = 0;
1849 	u32 event;
1850 	struct d40_phy_res *phy = d40c->phy_chan;
1851 	bool is_src;
1852 
1853 	/* Terminate all queued and active transfers */
1854 	d40_term_all(d40c);
1855 
1856 	if (phy == NULL) {
1857 		chan_err(d40c, "phy == null\n");
1858 		return -EINVAL;
1859 	}
1860 
1861 	if (phy->allocated_src == D40_ALLOC_FREE &&
1862 	    phy->allocated_dst == D40_ALLOC_FREE) {
1863 		chan_err(d40c, "channel already free\n");
1864 		return -EINVAL;
1865 	}
1866 
1867 	if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1868 	    d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1869 		event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
1870 		is_src = false;
1871 	} else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1872 		event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
1873 		is_src = true;
1874 	} else {
1875 		chan_err(d40c, "Unknown direction\n");
1876 		return -EINVAL;
1877 	}
1878 
1879 	pm_runtime_get_sync(d40c->base->dev);
1880 	res = d40_channel_execute_command(d40c, D40_DMA_STOP);
1881 	if (res) {
1882 		chan_err(d40c, "stop failed\n");
1883 		goto out;
1884 	}
1885 
1886 	d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
1887 
1888 	if (chan_is_logical(d40c))
1889 		d40c->base->lookup_log_chans[d40c->log_num] = NULL;
1890 	else
1891 		d40c->base->lookup_phy_chans[phy->num] = NULL;
1892 
1893 	if (d40c->busy) {
1894 		pm_runtime_mark_last_busy(d40c->base->dev);
1895 		pm_runtime_put_autosuspend(d40c->base->dev);
1896 	}
1897 
1898 	d40c->busy = false;
1899 	d40c->phy_chan = NULL;
1900 	d40c->configured = false;
1901 out:
1902 
1903 	pm_runtime_mark_last_busy(d40c->base->dev);
1904 	pm_runtime_put_autosuspend(d40c->base->dev);
1905 	return res;
1906 }
1907 
1908 static bool d40_is_paused(struct d40_chan *d40c)
1909 {
1910 	void __iomem *chanbase = chan_base(d40c);
1911 	bool is_paused = false;
1912 	unsigned long flags;
1913 	void __iomem *active_reg;
1914 	u32 status;
1915 	u32 event;
1916 
1917 	spin_lock_irqsave(&d40c->lock, flags);
1918 
1919 	if (chan_is_physical(d40c)) {
1920 		if (d40c->phy_chan->num % 2 == 0)
1921 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1922 		else
1923 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1924 
1925 		status = (readl(active_reg) &
1926 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1927 			D40_CHAN_POS(d40c->phy_chan->num);
1928 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1929 			is_paused = true;
1930 
1931 		goto _exit;
1932 	}
1933 
1934 	if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1935 	    d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1936 		event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
1937 		status = readl(chanbase + D40_CHAN_REG_SDLNK);
1938 	} else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1939 		event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
1940 		status = readl(chanbase + D40_CHAN_REG_SSLNK);
1941 	} else {
1942 		chan_err(d40c, "Unknown direction\n");
1943 		goto _exit;
1944 	}
1945 
1946 	status = (status & D40_EVENTLINE_MASK(event)) >>
1947 		D40_EVENTLINE_POS(event);
1948 
1949 	if (status != D40_DMA_RUN)
1950 		is_paused = true;
1951 _exit:
1952 	spin_unlock_irqrestore(&d40c->lock, flags);
1953 	return is_paused;
1954 
1955 }
1956 
1957 
1958 static u32 stedma40_residue(struct dma_chan *chan)
1959 {
1960 	struct d40_chan *d40c =
1961 		container_of(chan, struct d40_chan, chan);
1962 	u32 bytes_left;
1963 	unsigned long flags;
1964 
1965 	spin_lock_irqsave(&d40c->lock, flags);
1966 	bytes_left = d40_residue(d40c);
1967 	spin_unlock_irqrestore(&d40c->lock, flags);
1968 
1969 	return bytes_left;
1970 }
1971 
1972 static int
1973 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
1974 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
1975 		unsigned int sg_len, dma_addr_t src_dev_addr,
1976 		dma_addr_t dst_dev_addr)
1977 {
1978 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
1979 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
1980 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
1981 	int ret;
1982 
1983 	ret = d40_log_sg_to_lli(sg_src, sg_len,
1984 				src_dev_addr,
1985 				desc->lli_log.src,
1986 				chan->log_def.lcsp1,
1987 				src_info->data_width,
1988 				dst_info->data_width);
1989 
1990 	ret = d40_log_sg_to_lli(sg_dst, sg_len,
1991 				dst_dev_addr,
1992 				desc->lli_log.dst,
1993 				chan->log_def.lcsp3,
1994 				dst_info->data_width,
1995 				src_info->data_width);
1996 
1997 	return ret < 0 ? ret : 0;
1998 }
1999 
2000 static int
2001 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2002 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2003 		unsigned int sg_len, dma_addr_t src_dev_addr,
2004 		dma_addr_t dst_dev_addr)
2005 {
2006 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2007 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2008 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2009 	unsigned long flags = 0;
2010 	int ret;
2011 
2012 	if (desc->cyclic)
2013 		flags |= LLI_CYCLIC | LLI_TERM_INT;
2014 
2015 	ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2016 				desc->lli_phy.src,
2017 				virt_to_phys(desc->lli_phy.src),
2018 				chan->src_def_cfg,
2019 				src_info, dst_info, flags);
2020 
2021 	ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2022 				desc->lli_phy.dst,
2023 				virt_to_phys(desc->lli_phy.dst),
2024 				chan->dst_def_cfg,
2025 				dst_info, src_info, flags);
2026 
2027 	dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2028 				   desc->lli_pool.size, DMA_TO_DEVICE);
2029 
2030 	return ret < 0 ? ret : 0;
2031 }
2032 
2033 
2034 static struct d40_desc *
2035 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2036 	      unsigned int sg_len, unsigned long dma_flags)
2037 {
2038 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2039 	struct d40_desc *desc;
2040 	int ret;
2041 
2042 	desc = d40_desc_get(chan);
2043 	if (!desc)
2044 		return NULL;
2045 
2046 	desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2047 					cfg->dst_info.data_width);
2048 	if (desc->lli_len < 0) {
2049 		chan_err(chan, "Unaligned size\n");
2050 		goto err;
2051 	}
2052 
2053 	ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2054 	if (ret < 0) {
2055 		chan_err(chan, "Could not allocate lli\n");
2056 		goto err;
2057 	}
2058 
2059 
2060 	desc->lli_current = 0;
2061 	desc->txd.flags = dma_flags;
2062 	desc->txd.tx_submit = d40_tx_submit;
2063 
2064 	dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2065 
2066 	return desc;
2067 
2068 err:
2069 	d40_desc_free(chan, desc);
2070 	return NULL;
2071 }
2072 
2073 static dma_addr_t
2074 d40_get_dev_addr(struct d40_chan *chan, enum dma_transfer_direction direction)
2075 {
2076 	struct stedma40_platform_data *plat = chan->base->plat_data;
2077 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2078 	dma_addr_t addr = 0;
2079 
2080 	if (chan->runtime_addr)
2081 		return chan->runtime_addr;
2082 
2083 	if (direction == DMA_DEV_TO_MEM)
2084 		addr = plat->dev_rx[cfg->src_dev_type];
2085 	else if (direction == DMA_MEM_TO_DEV)
2086 		addr = plat->dev_tx[cfg->dst_dev_type];
2087 
2088 	return addr;
2089 }
2090 
2091 static struct dma_async_tx_descriptor *
2092 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2093 	    struct scatterlist *sg_dst, unsigned int sg_len,
2094 	    enum dma_transfer_direction direction, unsigned long dma_flags)
2095 {
2096 	struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2097 	dma_addr_t src_dev_addr = 0;
2098 	dma_addr_t dst_dev_addr = 0;
2099 	struct d40_desc *desc;
2100 	unsigned long flags;
2101 	int ret;
2102 
2103 	if (!chan->phy_chan) {
2104 		chan_err(chan, "Cannot prepare unallocated channel\n");
2105 		return NULL;
2106 	}
2107 
2108 
2109 	spin_lock_irqsave(&chan->lock, flags);
2110 
2111 	desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2112 	if (desc == NULL)
2113 		goto err;
2114 
2115 	if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2116 		desc->cyclic = true;
2117 
2118 	if (direction != DMA_TRANS_NONE) {
2119 		dma_addr_t dev_addr = d40_get_dev_addr(chan, direction);
2120 
2121 		if (direction == DMA_DEV_TO_MEM)
2122 			src_dev_addr = dev_addr;
2123 		else if (direction == DMA_MEM_TO_DEV)
2124 			dst_dev_addr = dev_addr;
2125 	}
2126 
2127 	if (chan_is_logical(chan))
2128 		ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2129 				      sg_len, src_dev_addr, dst_dev_addr);
2130 	else
2131 		ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2132 				      sg_len, src_dev_addr, dst_dev_addr);
2133 
2134 	if (ret) {
2135 		chan_err(chan, "Failed to prepare %s sg job: %d\n",
2136 			 chan_is_logical(chan) ? "log" : "phy", ret);
2137 		goto err;
2138 	}
2139 
2140 	/*
2141 	 * add descriptor to the prepare queue in order to be able
2142 	 * to free them later in terminate_all
2143 	 */
2144 	list_add_tail(&desc->node, &chan->prepare_queue);
2145 
2146 	spin_unlock_irqrestore(&chan->lock, flags);
2147 
2148 	return &desc->txd;
2149 
2150 err:
2151 	if (desc)
2152 		d40_desc_free(chan, desc);
2153 	spin_unlock_irqrestore(&chan->lock, flags);
2154 	return NULL;
2155 }
2156 
2157 bool stedma40_filter(struct dma_chan *chan, void *data)
2158 {
2159 	struct stedma40_chan_cfg *info = data;
2160 	struct d40_chan *d40c =
2161 		container_of(chan, struct d40_chan, chan);
2162 	int err;
2163 
2164 	if (data) {
2165 		err = d40_validate_conf(d40c, info);
2166 		if (!err)
2167 			d40c->dma_cfg = *info;
2168 	} else
2169 		err = d40_config_memcpy(d40c);
2170 
2171 	if (!err)
2172 		d40c->configured = true;
2173 
2174 	return err == 0;
2175 }
2176 EXPORT_SYMBOL(stedma40_filter);
2177 
2178 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2179 {
2180 	bool realtime = d40c->dma_cfg.realtime;
2181 	bool highprio = d40c->dma_cfg.high_priority;
2182 	u32 prioreg = highprio ? D40_DREG_PSEG1 : D40_DREG_PCEG1;
2183 	u32 rtreg = realtime ? D40_DREG_RSEG1 : D40_DREG_RCEG1;
2184 	u32 event = D40_TYPE_TO_EVENT(dev_type);
2185 	u32 group = D40_TYPE_TO_GROUP(dev_type);
2186 	u32 bit = 1 << event;
2187 
2188 	/* Destination event lines are stored in the upper halfword */
2189 	if (!src)
2190 		bit <<= 16;
2191 
2192 	writel(bit, d40c->base->virtbase + prioreg + group * 4);
2193 	writel(bit, d40c->base->virtbase + rtreg + group * 4);
2194 }
2195 
2196 static void d40_set_prio_realtime(struct d40_chan *d40c)
2197 {
2198 	if (d40c->base->rev < 3)
2199 		return;
2200 
2201 	if ((d40c->dma_cfg.dir ==  STEDMA40_PERIPH_TO_MEM) ||
2202 	    (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH))
2203 		__d40_set_prio_rt(d40c, d40c->dma_cfg.src_dev_type, true);
2204 
2205 	if ((d40c->dma_cfg.dir ==  STEDMA40_MEM_TO_PERIPH) ||
2206 	    (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH))
2207 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dst_dev_type, false);
2208 }
2209 
2210 /* DMA ENGINE functions */
2211 static int d40_alloc_chan_resources(struct dma_chan *chan)
2212 {
2213 	int err;
2214 	unsigned long flags;
2215 	struct d40_chan *d40c =
2216 		container_of(chan, struct d40_chan, chan);
2217 	bool is_free_phy;
2218 	spin_lock_irqsave(&d40c->lock, flags);
2219 
2220 	dma_cookie_init(chan);
2221 
2222 	/* If no dma configuration is set use default configuration (memcpy) */
2223 	if (!d40c->configured) {
2224 		err = d40_config_memcpy(d40c);
2225 		if (err) {
2226 			chan_err(d40c, "Failed to configure memcpy channel\n");
2227 			goto fail;
2228 		}
2229 	}
2230 
2231 	err = d40_allocate_channel(d40c, &is_free_phy);
2232 	if (err) {
2233 		chan_err(d40c, "Failed to allocate channel\n");
2234 		d40c->configured = false;
2235 		goto fail;
2236 	}
2237 
2238 	pm_runtime_get_sync(d40c->base->dev);
2239 	/* Fill in basic CFG register values */
2240 	d40_phy_cfg(&d40c->dma_cfg, &d40c->src_def_cfg,
2241 		    &d40c->dst_def_cfg, chan_is_logical(d40c));
2242 
2243 	d40_set_prio_realtime(d40c);
2244 
2245 	if (chan_is_logical(d40c)) {
2246 		d40_log_cfg(&d40c->dma_cfg,
2247 			    &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2248 
2249 		if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM)
2250 			d40c->lcpa = d40c->base->lcpa_base +
2251 			  d40c->dma_cfg.src_dev_type * D40_LCPA_CHAN_SIZE;
2252 		else
2253 			d40c->lcpa = d40c->base->lcpa_base +
2254 			  d40c->dma_cfg.dst_dev_type *
2255 			  D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2256 	}
2257 
2258 	dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2259 		 chan_is_logical(d40c) ? "logical" : "physical",
2260 		 d40c->phy_chan->num,
2261 		 d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2262 
2263 
2264 	/*
2265 	 * Only write channel configuration to the DMA if the physical
2266 	 * resource is free. In case of multiple logical channels
2267 	 * on the same physical resource, only the first write is necessary.
2268 	 */
2269 	if (is_free_phy)
2270 		d40_config_write(d40c);
2271 fail:
2272 	pm_runtime_mark_last_busy(d40c->base->dev);
2273 	pm_runtime_put_autosuspend(d40c->base->dev);
2274 	spin_unlock_irqrestore(&d40c->lock, flags);
2275 	return err;
2276 }
2277 
2278 static void d40_free_chan_resources(struct dma_chan *chan)
2279 {
2280 	struct d40_chan *d40c =
2281 		container_of(chan, struct d40_chan, chan);
2282 	int err;
2283 	unsigned long flags;
2284 
2285 	if (d40c->phy_chan == NULL) {
2286 		chan_err(d40c, "Cannot free unallocated channel\n");
2287 		return;
2288 	}
2289 
2290 
2291 	spin_lock_irqsave(&d40c->lock, flags);
2292 
2293 	err = d40_free_dma(d40c);
2294 
2295 	if (err)
2296 		chan_err(d40c, "Failed to free channel\n");
2297 	spin_unlock_irqrestore(&d40c->lock, flags);
2298 }
2299 
2300 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2301 						       dma_addr_t dst,
2302 						       dma_addr_t src,
2303 						       size_t size,
2304 						       unsigned long dma_flags)
2305 {
2306 	struct scatterlist dst_sg;
2307 	struct scatterlist src_sg;
2308 
2309 	sg_init_table(&dst_sg, 1);
2310 	sg_init_table(&src_sg, 1);
2311 
2312 	sg_dma_address(&dst_sg) = dst;
2313 	sg_dma_address(&src_sg) = src;
2314 
2315 	sg_dma_len(&dst_sg) = size;
2316 	sg_dma_len(&src_sg) = size;
2317 
2318 	return d40_prep_sg(chan, &src_sg, &dst_sg, 1, DMA_NONE, dma_flags);
2319 }
2320 
2321 static struct dma_async_tx_descriptor *
2322 d40_prep_memcpy_sg(struct dma_chan *chan,
2323 		   struct scatterlist *dst_sg, unsigned int dst_nents,
2324 		   struct scatterlist *src_sg, unsigned int src_nents,
2325 		   unsigned long dma_flags)
2326 {
2327 	if (dst_nents != src_nents)
2328 		return NULL;
2329 
2330 	return d40_prep_sg(chan, src_sg, dst_sg, src_nents, DMA_NONE, dma_flags);
2331 }
2332 
2333 static struct dma_async_tx_descriptor *d40_prep_slave_sg(struct dma_chan *chan,
2334 							 struct scatterlist *sgl,
2335 							 unsigned int sg_len,
2336 							 enum dma_transfer_direction direction,
2337 							 unsigned long dma_flags,
2338 							 void *context)
2339 {
2340 	if (direction != DMA_DEV_TO_MEM && direction != DMA_MEM_TO_DEV)
2341 		return NULL;
2342 
2343 	return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2344 }
2345 
2346 static struct dma_async_tx_descriptor *
2347 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2348 		     size_t buf_len, size_t period_len,
2349 		     enum dma_transfer_direction direction, unsigned long flags,
2350 		     void *context)
2351 {
2352 	unsigned int periods = buf_len / period_len;
2353 	struct dma_async_tx_descriptor *txd;
2354 	struct scatterlist *sg;
2355 	int i;
2356 
2357 	sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2358 	for (i = 0; i < periods; i++) {
2359 		sg_dma_address(&sg[i]) = dma_addr;
2360 		sg_dma_len(&sg[i]) = period_len;
2361 		dma_addr += period_len;
2362 	}
2363 
2364 	sg[periods].offset = 0;
2365 	sg_dma_len(&sg[periods]) = 0;
2366 	sg[periods].page_link =
2367 		((unsigned long)sg | 0x01) & ~0x02;
2368 
2369 	txd = d40_prep_sg(chan, sg, sg, periods, direction,
2370 			  DMA_PREP_INTERRUPT);
2371 
2372 	kfree(sg);
2373 
2374 	return txd;
2375 }
2376 
2377 static enum dma_status d40_tx_status(struct dma_chan *chan,
2378 				     dma_cookie_t cookie,
2379 				     struct dma_tx_state *txstate)
2380 {
2381 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2382 	enum dma_status ret;
2383 
2384 	if (d40c->phy_chan == NULL) {
2385 		chan_err(d40c, "Cannot read status of unallocated channel\n");
2386 		return -EINVAL;
2387 	}
2388 
2389 	ret = dma_cookie_status(chan, cookie, txstate);
2390 	if (ret != DMA_SUCCESS)
2391 		dma_set_residue(txstate, stedma40_residue(chan));
2392 
2393 	if (d40_is_paused(d40c))
2394 		ret = DMA_PAUSED;
2395 
2396 	return ret;
2397 }
2398 
2399 static void d40_issue_pending(struct dma_chan *chan)
2400 {
2401 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2402 	unsigned long flags;
2403 
2404 	if (d40c->phy_chan == NULL) {
2405 		chan_err(d40c, "Channel is not allocated!\n");
2406 		return;
2407 	}
2408 
2409 	spin_lock_irqsave(&d40c->lock, flags);
2410 
2411 	list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2412 
2413 	/* Busy means that queued jobs are already being processed */
2414 	if (!d40c->busy)
2415 		(void) d40_queue_start(d40c);
2416 
2417 	spin_unlock_irqrestore(&d40c->lock, flags);
2418 }
2419 
2420 static void d40_terminate_all(struct dma_chan *chan)
2421 {
2422 	unsigned long flags;
2423 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2424 	int ret;
2425 
2426 	spin_lock_irqsave(&d40c->lock, flags);
2427 
2428 	pm_runtime_get_sync(d40c->base->dev);
2429 	ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2430 	if (ret)
2431 		chan_err(d40c, "Failed to stop channel\n");
2432 
2433 	d40_term_all(d40c);
2434 	pm_runtime_mark_last_busy(d40c->base->dev);
2435 	pm_runtime_put_autosuspend(d40c->base->dev);
2436 	if (d40c->busy) {
2437 		pm_runtime_mark_last_busy(d40c->base->dev);
2438 		pm_runtime_put_autosuspend(d40c->base->dev);
2439 	}
2440 	d40c->busy = false;
2441 
2442 	spin_unlock_irqrestore(&d40c->lock, flags);
2443 }
2444 
2445 static int
2446 dma40_config_to_halfchannel(struct d40_chan *d40c,
2447 			    struct stedma40_half_channel_info *info,
2448 			    enum dma_slave_buswidth width,
2449 			    u32 maxburst)
2450 {
2451 	enum stedma40_periph_data_width addr_width;
2452 	int psize;
2453 
2454 	switch (width) {
2455 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
2456 		addr_width = STEDMA40_BYTE_WIDTH;
2457 		break;
2458 	case DMA_SLAVE_BUSWIDTH_2_BYTES:
2459 		addr_width = STEDMA40_HALFWORD_WIDTH;
2460 		break;
2461 	case DMA_SLAVE_BUSWIDTH_4_BYTES:
2462 		addr_width = STEDMA40_WORD_WIDTH;
2463 		break;
2464 	case DMA_SLAVE_BUSWIDTH_8_BYTES:
2465 		addr_width = STEDMA40_DOUBLEWORD_WIDTH;
2466 		break;
2467 	default:
2468 		dev_err(d40c->base->dev,
2469 			"illegal peripheral address width "
2470 			"requested (%d)\n",
2471 			width);
2472 		return -EINVAL;
2473 	}
2474 
2475 	if (chan_is_logical(d40c)) {
2476 		if (maxburst >= 16)
2477 			psize = STEDMA40_PSIZE_LOG_16;
2478 		else if (maxburst >= 8)
2479 			psize = STEDMA40_PSIZE_LOG_8;
2480 		else if (maxburst >= 4)
2481 			psize = STEDMA40_PSIZE_LOG_4;
2482 		else
2483 			psize = STEDMA40_PSIZE_LOG_1;
2484 	} else {
2485 		if (maxburst >= 16)
2486 			psize = STEDMA40_PSIZE_PHY_16;
2487 		else if (maxburst >= 8)
2488 			psize = STEDMA40_PSIZE_PHY_8;
2489 		else if (maxburst >= 4)
2490 			psize = STEDMA40_PSIZE_PHY_4;
2491 		else
2492 			psize = STEDMA40_PSIZE_PHY_1;
2493 	}
2494 
2495 	info->data_width = addr_width;
2496 	info->psize = psize;
2497 	info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2498 
2499 	return 0;
2500 }
2501 
2502 /* Runtime reconfiguration extension */
2503 static int d40_set_runtime_config(struct dma_chan *chan,
2504 				  struct dma_slave_config *config)
2505 {
2506 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2507 	struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2508 	enum dma_slave_buswidth src_addr_width, dst_addr_width;
2509 	dma_addr_t config_addr;
2510 	u32 src_maxburst, dst_maxburst;
2511 	int ret;
2512 
2513 	src_addr_width = config->src_addr_width;
2514 	src_maxburst = config->src_maxburst;
2515 	dst_addr_width = config->dst_addr_width;
2516 	dst_maxburst = config->dst_maxburst;
2517 
2518 	if (config->direction == DMA_DEV_TO_MEM) {
2519 		dma_addr_t dev_addr_rx =
2520 			d40c->base->plat_data->dev_rx[cfg->src_dev_type];
2521 
2522 		config_addr = config->src_addr;
2523 		if (dev_addr_rx)
2524 			dev_dbg(d40c->base->dev,
2525 				"channel has a pre-wired RX address %08x "
2526 				"overriding with %08x\n",
2527 				dev_addr_rx, config_addr);
2528 		if (cfg->dir != STEDMA40_PERIPH_TO_MEM)
2529 			dev_dbg(d40c->base->dev,
2530 				"channel was not configured for peripheral "
2531 				"to memory transfer (%d) overriding\n",
2532 				cfg->dir);
2533 		cfg->dir = STEDMA40_PERIPH_TO_MEM;
2534 
2535 		/* Configure the memory side */
2536 		if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2537 			dst_addr_width = src_addr_width;
2538 		if (dst_maxburst == 0)
2539 			dst_maxburst = src_maxburst;
2540 
2541 	} else if (config->direction == DMA_MEM_TO_DEV) {
2542 		dma_addr_t dev_addr_tx =
2543 			d40c->base->plat_data->dev_tx[cfg->dst_dev_type];
2544 
2545 		config_addr = config->dst_addr;
2546 		if (dev_addr_tx)
2547 			dev_dbg(d40c->base->dev,
2548 				"channel has a pre-wired TX address %08x "
2549 				"overriding with %08x\n",
2550 				dev_addr_tx, config_addr);
2551 		if (cfg->dir != STEDMA40_MEM_TO_PERIPH)
2552 			dev_dbg(d40c->base->dev,
2553 				"channel was not configured for memory "
2554 				"to peripheral transfer (%d) overriding\n",
2555 				cfg->dir);
2556 		cfg->dir = STEDMA40_MEM_TO_PERIPH;
2557 
2558 		/* Configure the memory side */
2559 		if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2560 			src_addr_width = dst_addr_width;
2561 		if (src_maxburst == 0)
2562 			src_maxburst = dst_maxburst;
2563 	} else {
2564 		dev_err(d40c->base->dev,
2565 			"unrecognized channel direction %d\n",
2566 			config->direction);
2567 		return -EINVAL;
2568 	}
2569 
2570 	if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2571 		dev_err(d40c->base->dev,
2572 			"src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2573 			src_maxburst,
2574 			src_addr_width,
2575 			dst_maxburst,
2576 			dst_addr_width);
2577 		return -EINVAL;
2578 	}
2579 
2580 	ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2581 					  src_addr_width,
2582 					  src_maxburst);
2583 	if (ret)
2584 		return ret;
2585 
2586 	ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2587 					  dst_addr_width,
2588 					  dst_maxburst);
2589 	if (ret)
2590 		return ret;
2591 
2592 	/* Fill in register values */
2593 	if (chan_is_logical(d40c))
2594 		d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2595 	else
2596 		d40_phy_cfg(cfg, &d40c->src_def_cfg,
2597 			    &d40c->dst_def_cfg, false);
2598 
2599 	/* These settings will take precedence later */
2600 	d40c->runtime_addr = config_addr;
2601 	d40c->runtime_direction = config->direction;
2602 	dev_dbg(d40c->base->dev,
2603 		"configured channel %s for %s, data width %d/%d, "
2604 		"maxburst %d/%d elements, LE, no flow control\n",
2605 		dma_chan_name(chan),
2606 		(config->direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2607 		src_addr_width, dst_addr_width,
2608 		src_maxburst, dst_maxburst);
2609 
2610 	return 0;
2611 }
2612 
2613 static int d40_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
2614 		       unsigned long arg)
2615 {
2616 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2617 
2618 	if (d40c->phy_chan == NULL) {
2619 		chan_err(d40c, "Channel is not allocated!\n");
2620 		return -EINVAL;
2621 	}
2622 
2623 	switch (cmd) {
2624 	case DMA_TERMINATE_ALL:
2625 		d40_terminate_all(chan);
2626 		return 0;
2627 	case DMA_PAUSE:
2628 		return d40_pause(d40c);
2629 	case DMA_RESUME:
2630 		return d40_resume(d40c);
2631 	case DMA_SLAVE_CONFIG:
2632 		return d40_set_runtime_config(chan,
2633 			(struct dma_slave_config *) arg);
2634 	default:
2635 		break;
2636 	}
2637 
2638 	/* Other commands are unimplemented */
2639 	return -ENXIO;
2640 }
2641 
2642 /* Initialization functions */
2643 
2644 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2645 				 struct d40_chan *chans, int offset,
2646 				 int num_chans)
2647 {
2648 	int i = 0;
2649 	struct d40_chan *d40c;
2650 
2651 	INIT_LIST_HEAD(&dma->channels);
2652 
2653 	for (i = offset; i < offset + num_chans; i++) {
2654 		d40c = &chans[i];
2655 		d40c->base = base;
2656 		d40c->chan.device = dma;
2657 
2658 		spin_lock_init(&d40c->lock);
2659 
2660 		d40c->log_num = D40_PHY_CHAN;
2661 
2662 		INIT_LIST_HEAD(&d40c->active);
2663 		INIT_LIST_HEAD(&d40c->queue);
2664 		INIT_LIST_HEAD(&d40c->pending_queue);
2665 		INIT_LIST_HEAD(&d40c->client);
2666 		INIT_LIST_HEAD(&d40c->prepare_queue);
2667 
2668 		tasklet_init(&d40c->tasklet, dma_tasklet,
2669 			     (unsigned long) d40c);
2670 
2671 		list_add_tail(&d40c->chan.device_node,
2672 			      &dma->channels);
2673 	}
2674 }
2675 
2676 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2677 {
2678 	if (dma_has_cap(DMA_SLAVE, dev->cap_mask))
2679 		dev->device_prep_slave_sg = d40_prep_slave_sg;
2680 
2681 	if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2682 		dev->device_prep_dma_memcpy = d40_prep_memcpy;
2683 
2684 		/*
2685 		 * This controller can only access address at even
2686 		 * 32bit boundaries, i.e. 2^2
2687 		 */
2688 		dev->copy_align = 2;
2689 	}
2690 
2691 	if (dma_has_cap(DMA_SG, dev->cap_mask))
2692 		dev->device_prep_dma_sg = d40_prep_memcpy_sg;
2693 
2694 	if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2695 		dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2696 
2697 	dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2698 	dev->device_free_chan_resources = d40_free_chan_resources;
2699 	dev->device_issue_pending = d40_issue_pending;
2700 	dev->device_tx_status = d40_tx_status;
2701 	dev->device_control = d40_control;
2702 	dev->dev = base->dev;
2703 }
2704 
2705 static int __init d40_dmaengine_init(struct d40_base *base,
2706 				     int num_reserved_chans)
2707 {
2708 	int err ;
2709 
2710 	d40_chan_init(base, &base->dma_slave, base->log_chans,
2711 		      0, base->num_log_chans);
2712 
2713 	dma_cap_zero(base->dma_slave.cap_mask);
2714 	dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2715 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2716 
2717 	d40_ops_init(base, &base->dma_slave);
2718 
2719 	err = dma_async_device_register(&base->dma_slave);
2720 
2721 	if (err) {
2722 		d40_err(base->dev, "Failed to register slave channels\n");
2723 		goto failure1;
2724 	}
2725 
2726 	d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2727 		      base->num_log_chans, base->plat_data->memcpy_len);
2728 
2729 	dma_cap_zero(base->dma_memcpy.cap_mask);
2730 	dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2731 	dma_cap_set(DMA_SG, base->dma_memcpy.cap_mask);
2732 
2733 	d40_ops_init(base, &base->dma_memcpy);
2734 
2735 	err = dma_async_device_register(&base->dma_memcpy);
2736 
2737 	if (err) {
2738 		d40_err(base->dev,
2739 			"Failed to regsiter memcpy only channels\n");
2740 		goto failure2;
2741 	}
2742 
2743 	d40_chan_init(base, &base->dma_both, base->phy_chans,
2744 		      0, num_reserved_chans);
2745 
2746 	dma_cap_zero(base->dma_both.cap_mask);
2747 	dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2748 	dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2749 	dma_cap_set(DMA_SG, base->dma_both.cap_mask);
2750 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2751 
2752 	d40_ops_init(base, &base->dma_both);
2753 	err = dma_async_device_register(&base->dma_both);
2754 
2755 	if (err) {
2756 		d40_err(base->dev,
2757 			"Failed to register logical and physical capable channels\n");
2758 		goto failure3;
2759 	}
2760 	return 0;
2761 failure3:
2762 	dma_async_device_unregister(&base->dma_memcpy);
2763 failure2:
2764 	dma_async_device_unregister(&base->dma_slave);
2765 failure1:
2766 	return err;
2767 }
2768 
2769 /* Suspend resume functionality */
2770 #ifdef CONFIG_PM
2771 static int dma40_pm_suspend(struct device *dev)
2772 {
2773 	struct platform_device *pdev = to_platform_device(dev);
2774 	struct d40_base *base = platform_get_drvdata(pdev);
2775 	int ret = 0;
2776 	if (!pm_runtime_suspended(dev))
2777 		return -EBUSY;
2778 
2779 	if (base->lcpa_regulator)
2780 		ret = regulator_disable(base->lcpa_regulator);
2781 	return ret;
2782 }
2783 
2784 static int dma40_runtime_suspend(struct device *dev)
2785 {
2786 	struct platform_device *pdev = to_platform_device(dev);
2787 	struct d40_base *base = platform_get_drvdata(pdev);
2788 
2789 	d40_save_restore_registers(base, true);
2790 
2791 	/* Don't disable/enable clocks for v1 due to HW bugs */
2792 	if (base->rev != 1)
2793 		writel_relaxed(base->gcc_pwr_off_mask,
2794 			       base->virtbase + D40_DREG_GCC);
2795 
2796 	return 0;
2797 }
2798 
2799 static int dma40_runtime_resume(struct device *dev)
2800 {
2801 	struct platform_device *pdev = to_platform_device(dev);
2802 	struct d40_base *base = platform_get_drvdata(pdev);
2803 
2804 	if (base->initialized)
2805 		d40_save_restore_registers(base, false);
2806 
2807 	writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
2808 		       base->virtbase + D40_DREG_GCC);
2809 	return 0;
2810 }
2811 
2812 static int dma40_resume(struct device *dev)
2813 {
2814 	struct platform_device *pdev = to_platform_device(dev);
2815 	struct d40_base *base = platform_get_drvdata(pdev);
2816 	int ret = 0;
2817 
2818 	if (base->lcpa_regulator)
2819 		ret = regulator_enable(base->lcpa_regulator);
2820 
2821 	return ret;
2822 }
2823 
2824 static const struct dev_pm_ops dma40_pm_ops = {
2825 	.suspend		= dma40_pm_suspend,
2826 	.runtime_suspend	= dma40_runtime_suspend,
2827 	.runtime_resume		= dma40_runtime_resume,
2828 	.resume			= dma40_resume,
2829 };
2830 #define DMA40_PM_OPS	(&dma40_pm_ops)
2831 #else
2832 #define DMA40_PM_OPS	NULL
2833 #endif
2834 
2835 /* Initialization functions. */
2836 
2837 static int __init d40_phy_res_init(struct d40_base *base)
2838 {
2839 	int i;
2840 	int num_phy_chans_avail = 0;
2841 	u32 val[2];
2842 	int odd_even_bit = -2;
2843 	int gcc = D40_DREG_GCC_ENA;
2844 
2845 	val[0] = readl(base->virtbase + D40_DREG_PRSME);
2846 	val[1] = readl(base->virtbase + D40_DREG_PRSMO);
2847 
2848 	for (i = 0; i < base->num_phy_chans; i++) {
2849 		base->phy_res[i].num = i;
2850 		odd_even_bit += 2 * ((i % 2) == 0);
2851 		if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
2852 			/* Mark security only channels as occupied */
2853 			base->phy_res[i].allocated_src = D40_ALLOC_PHY;
2854 			base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
2855 			base->phy_res[i].reserved = true;
2856 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
2857 						       D40_DREG_GCC_SRC);
2858 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
2859 						       D40_DREG_GCC_DST);
2860 
2861 
2862 		} else {
2863 			base->phy_res[i].allocated_src = D40_ALLOC_FREE;
2864 			base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
2865 			base->phy_res[i].reserved = false;
2866 			num_phy_chans_avail++;
2867 		}
2868 		spin_lock_init(&base->phy_res[i].lock);
2869 	}
2870 
2871 	/* Mark disabled channels as occupied */
2872 	for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
2873 		int chan = base->plat_data->disabled_channels[i];
2874 
2875 		base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
2876 		base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
2877 		base->phy_res[chan].reserved = true;
2878 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
2879 					       D40_DREG_GCC_SRC);
2880 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
2881 					       D40_DREG_GCC_DST);
2882 		num_phy_chans_avail--;
2883 	}
2884 
2885 	dev_info(base->dev, "%d of %d physical DMA channels available\n",
2886 		 num_phy_chans_avail, base->num_phy_chans);
2887 
2888 	/* Verify settings extended vs standard */
2889 	val[0] = readl(base->virtbase + D40_DREG_PRTYP);
2890 
2891 	for (i = 0; i < base->num_phy_chans; i++) {
2892 
2893 		if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
2894 		    (val[0] & 0x3) != 1)
2895 			dev_info(base->dev,
2896 				 "[%s] INFO: channel %d is misconfigured (%d)\n",
2897 				 __func__, i, val[0] & 0x3);
2898 
2899 		val[0] = val[0] >> 2;
2900 	}
2901 
2902 	/*
2903 	 * To keep things simple, Enable all clocks initially.
2904 	 * The clocks will get managed later post channel allocation.
2905 	 * The clocks for the event lines on which reserved channels exists
2906 	 * are not managed here.
2907 	 */
2908 	writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
2909 	base->gcc_pwr_off_mask = gcc;
2910 
2911 	return num_phy_chans_avail;
2912 }
2913 
2914 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
2915 {
2916 	struct stedma40_platform_data *plat_data;
2917 	struct clk *clk = NULL;
2918 	void __iomem *virtbase = NULL;
2919 	struct resource *res = NULL;
2920 	struct d40_base *base = NULL;
2921 	int num_log_chans = 0;
2922 	int num_phy_chans;
2923 	int clk_ret = -EINVAL;
2924 	int i;
2925 	u32 pid;
2926 	u32 cid;
2927 	u8 rev;
2928 
2929 	clk = clk_get(&pdev->dev, NULL);
2930 	if (IS_ERR(clk)) {
2931 		d40_err(&pdev->dev, "No matching clock found\n");
2932 		goto failure;
2933 	}
2934 
2935 	clk_ret = clk_prepare_enable(clk);
2936 	if (clk_ret) {
2937 		d40_err(&pdev->dev, "Failed to prepare/enable clock\n");
2938 		goto failure;
2939 	}
2940 
2941 	/* Get IO for DMAC base address */
2942 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
2943 	if (!res)
2944 		goto failure;
2945 
2946 	if (request_mem_region(res->start, resource_size(res),
2947 			       D40_NAME " I/O base") == NULL)
2948 		goto failure;
2949 
2950 	virtbase = ioremap(res->start, resource_size(res));
2951 	if (!virtbase)
2952 		goto failure;
2953 
2954 	/* This is just a regular AMBA PrimeCell ID actually */
2955 	for (pid = 0, i = 0; i < 4; i++)
2956 		pid |= (readl(virtbase + resource_size(res) - 0x20 + 4 * i)
2957 			& 255) << (i * 8);
2958 	for (cid = 0, i = 0; i < 4; i++)
2959 		cid |= (readl(virtbase + resource_size(res) - 0x10 + 4 * i)
2960 			& 255) << (i * 8);
2961 
2962 	if (cid != AMBA_CID) {
2963 		d40_err(&pdev->dev, "Unknown hardware! No PrimeCell ID\n");
2964 		goto failure;
2965 	}
2966 	if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
2967 		d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
2968 			AMBA_MANF_BITS(pid),
2969 			AMBA_VENDOR_ST);
2970 		goto failure;
2971 	}
2972 	/*
2973 	 * HW revision:
2974 	 * DB8500ed has revision 0
2975 	 * ? has revision 1
2976 	 * DB8500v1 has revision 2
2977 	 * DB8500v2 has revision 3
2978 	 */
2979 	rev = AMBA_REV_BITS(pid);
2980 
2981 	/* The number of physical channels on this HW */
2982 	num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
2983 
2984 	dev_info(&pdev->dev, "hardware revision: %d @ 0x%x\n",
2985 		 rev, res->start);
2986 
2987 	if (rev < 2) {
2988 		d40_err(&pdev->dev, "hardware revision: %d is not supported",
2989 			rev);
2990 		goto failure;
2991 	}
2992 
2993 	plat_data = pdev->dev.platform_data;
2994 
2995 	/* Count the number of logical channels in use */
2996 	for (i = 0; i < plat_data->dev_len; i++)
2997 		if (plat_data->dev_rx[i] != 0)
2998 			num_log_chans++;
2999 
3000 	for (i = 0; i < plat_data->dev_len; i++)
3001 		if (plat_data->dev_tx[i] != 0)
3002 			num_log_chans++;
3003 
3004 	base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
3005 		       (num_phy_chans + num_log_chans + plat_data->memcpy_len) *
3006 		       sizeof(struct d40_chan), GFP_KERNEL);
3007 
3008 	if (base == NULL) {
3009 		d40_err(&pdev->dev, "Out of memory\n");
3010 		goto failure;
3011 	}
3012 
3013 	base->rev = rev;
3014 	base->clk = clk;
3015 	base->num_phy_chans = num_phy_chans;
3016 	base->num_log_chans = num_log_chans;
3017 	base->phy_start = res->start;
3018 	base->phy_size = resource_size(res);
3019 	base->virtbase = virtbase;
3020 	base->plat_data = plat_data;
3021 	base->dev = &pdev->dev;
3022 	base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3023 	base->log_chans = &base->phy_chans[num_phy_chans];
3024 
3025 	base->phy_res = kzalloc(num_phy_chans * sizeof(struct d40_phy_res),
3026 				GFP_KERNEL);
3027 	if (!base->phy_res)
3028 		goto failure;
3029 
3030 	base->lookup_phy_chans = kzalloc(num_phy_chans *
3031 					 sizeof(struct d40_chan *),
3032 					 GFP_KERNEL);
3033 	if (!base->lookup_phy_chans)
3034 		goto failure;
3035 
3036 	if (num_log_chans + plat_data->memcpy_len) {
3037 		/*
3038 		 * The max number of logical channels are event lines for all
3039 		 * src devices and dst devices
3040 		 */
3041 		base->lookup_log_chans = kzalloc(plat_data->dev_len * 2 *
3042 						 sizeof(struct d40_chan *),
3043 						 GFP_KERNEL);
3044 		if (!base->lookup_log_chans)
3045 			goto failure;
3046 	}
3047 
3048 	base->reg_val_backup_chan = kmalloc(base->num_phy_chans *
3049 					    sizeof(d40_backup_regs_chan),
3050 					    GFP_KERNEL);
3051 	if (!base->reg_val_backup_chan)
3052 		goto failure;
3053 
3054 	base->lcla_pool.alloc_map =
3055 		kzalloc(num_phy_chans * sizeof(struct d40_desc *)
3056 			* D40_LCLA_LINK_PER_EVENT_GRP, GFP_KERNEL);
3057 	if (!base->lcla_pool.alloc_map)
3058 		goto failure;
3059 
3060 	base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3061 					    0, SLAB_HWCACHE_ALIGN,
3062 					    NULL);
3063 	if (base->desc_slab == NULL)
3064 		goto failure;
3065 
3066 	return base;
3067 
3068 failure:
3069 	if (!clk_ret)
3070 		clk_disable_unprepare(clk);
3071 	if (!IS_ERR(clk))
3072 		clk_put(clk);
3073 	if (virtbase)
3074 		iounmap(virtbase);
3075 	if (res)
3076 		release_mem_region(res->start,
3077 				   resource_size(res));
3078 	if (virtbase)
3079 		iounmap(virtbase);
3080 
3081 	if (base) {
3082 		kfree(base->lcla_pool.alloc_map);
3083 		kfree(base->reg_val_backup_chan);
3084 		kfree(base->lookup_log_chans);
3085 		kfree(base->lookup_phy_chans);
3086 		kfree(base->phy_res);
3087 		kfree(base);
3088 	}
3089 
3090 	return NULL;
3091 }
3092 
3093 static void __init d40_hw_init(struct d40_base *base)
3094 {
3095 
3096 	static struct d40_reg_val dma_init_reg[] = {
3097 		/* Clock every part of the DMA block from start */
3098 		{ .reg = D40_DREG_GCC,    .val = D40_DREG_GCC_ENABLE_ALL},
3099 
3100 		/* Interrupts on all logical channels */
3101 		{ .reg = D40_DREG_LCMIS0, .val = 0xFFFFFFFF},
3102 		{ .reg = D40_DREG_LCMIS1, .val = 0xFFFFFFFF},
3103 		{ .reg = D40_DREG_LCMIS2, .val = 0xFFFFFFFF},
3104 		{ .reg = D40_DREG_LCMIS3, .val = 0xFFFFFFFF},
3105 		{ .reg = D40_DREG_LCICR0, .val = 0xFFFFFFFF},
3106 		{ .reg = D40_DREG_LCICR1, .val = 0xFFFFFFFF},
3107 		{ .reg = D40_DREG_LCICR2, .val = 0xFFFFFFFF},
3108 		{ .reg = D40_DREG_LCICR3, .val = 0xFFFFFFFF},
3109 		{ .reg = D40_DREG_LCTIS0, .val = 0xFFFFFFFF},
3110 		{ .reg = D40_DREG_LCTIS1, .val = 0xFFFFFFFF},
3111 		{ .reg = D40_DREG_LCTIS2, .val = 0xFFFFFFFF},
3112 		{ .reg = D40_DREG_LCTIS3, .val = 0xFFFFFFFF}
3113 	};
3114 	int i;
3115 	u32 prmseo[2] = {0, 0};
3116 	u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3117 	u32 pcmis = 0;
3118 	u32 pcicr = 0;
3119 
3120 	for (i = 0; i < ARRAY_SIZE(dma_init_reg); i++)
3121 		writel(dma_init_reg[i].val,
3122 		       base->virtbase + dma_init_reg[i].reg);
3123 
3124 	/* Configure all our dma channels to default settings */
3125 	for (i = 0; i < base->num_phy_chans; i++) {
3126 
3127 		activeo[i % 2] = activeo[i % 2] << 2;
3128 
3129 		if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3130 		    == D40_ALLOC_PHY) {
3131 			activeo[i % 2] |= 3;
3132 			continue;
3133 		}
3134 
3135 		/* Enable interrupt # */
3136 		pcmis = (pcmis << 1) | 1;
3137 
3138 		/* Clear interrupt # */
3139 		pcicr = (pcicr << 1) | 1;
3140 
3141 		/* Set channel to physical mode */
3142 		prmseo[i % 2] = prmseo[i % 2] << 2;
3143 		prmseo[i % 2] |= 1;
3144 
3145 	}
3146 
3147 	writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3148 	writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3149 	writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3150 	writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3151 
3152 	/* Write which interrupt to enable */
3153 	writel(pcmis, base->virtbase + D40_DREG_PCMIS);
3154 
3155 	/* Write which interrupt to clear */
3156 	writel(pcicr, base->virtbase + D40_DREG_PCICR);
3157 
3158 }
3159 
3160 static int __init d40_lcla_allocate(struct d40_base *base)
3161 {
3162 	struct d40_lcla_pool *pool = &base->lcla_pool;
3163 	unsigned long *page_list;
3164 	int i, j;
3165 	int ret = 0;
3166 
3167 	/*
3168 	 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3169 	 * To full fill this hardware requirement without wasting 256 kb
3170 	 * we allocate pages until we get an aligned one.
3171 	 */
3172 	page_list = kmalloc(sizeof(unsigned long) * MAX_LCLA_ALLOC_ATTEMPTS,
3173 			    GFP_KERNEL);
3174 
3175 	if (!page_list) {
3176 		ret = -ENOMEM;
3177 		goto failure;
3178 	}
3179 
3180 	/* Calculating how many pages that are required */
3181 	base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3182 
3183 	for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3184 		page_list[i] = __get_free_pages(GFP_KERNEL,
3185 						base->lcla_pool.pages);
3186 		if (!page_list[i]) {
3187 
3188 			d40_err(base->dev, "Failed to allocate %d pages.\n",
3189 				base->lcla_pool.pages);
3190 
3191 			for (j = 0; j < i; j++)
3192 				free_pages(page_list[j], base->lcla_pool.pages);
3193 			goto failure;
3194 		}
3195 
3196 		if ((virt_to_phys((void *)page_list[i]) &
3197 		     (LCLA_ALIGNMENT - 1)) == 0)
3198 			break;
3199 	}
3200 
3201 	for (j = 0; j < i; j++)
3202 		free_pages(page_list[j], base->lcla_pool.pages);
3203 
3204 	if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3205 		base->lcla_pool.base = (void *)page_list[i];
3206 	} else {
3207 		/*
3208 		 * After many attempts and no succees with finding the correct
3209 		 * alignment, try with allocating a big buffer.
3210 		 */
3211 		dev_warn(base->dev,
3212 			 "[%s] Failed to get %d pages @ 18 bit align.\n",
3213 			 __func__, base->lcla_pool.pages);
3214 		base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3215 							 base->num_phy_chans +
3216 							 LCLA_ALIGNMENT,
3217 							 GFP_KERNEL);
3218 		if (!base->lcla_pool.base_unaligned) {
3219 			ret = -ENOMEM;
3220 			goto failure;
3221 		}
3222 
3223 		base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3224 						 LCLA_ALIGNMENT);
3225 	}
3226 
3227 	pool->dma_addr = dma_map_single(base->dev, pool->base,
3228 					SZ_1K * base->num_phy_chans,
3229 					DMA_TO_DEVICE);
3230 	if (dma_mapping_error(base->dev, pool->dma_addr)) {
3231 		pool->dma_addr = 0;
3232 		ret = -ENOMEM;
3233 		goto failure;
3234 	}
3235 
3236 	writel(virt_to_phys(base->lcla_pool.base),
3237 	       base->virtbase + D40_DREG_LCLA);
3238 failure:
3239 	kfree(page_list);
3240 	return ret;
3241 }
3242 
3243 static int __init d40_probe(struct platform_device *pdev)
3244 {
3245 	int err;
3246 	int ret = -ENOENT;
3247 	struct d40_base *base;
3248 	struct resource *res = NULL;
3249 	int num_reserved_chans;
3250 	u32 val;
3251 
3252 	base = d40_hw_detect_init(pdev);
3253 
3254 	if (!base)
3255 		goto failure;
3256 
3257 	num_reserved_chans = d40_phy_res_init(base);
3258 
3259 	platform_set_drvdata(pdev, base);
3260 
3261 	spin_lock_init(&base->interrupt_lock);
3262 	spin_lock_init(&base->execmd_lock);
3263 
3264 	/* Get IO for logical channel parameter address */
3265 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
3266 	if (!res) {
3267 		ret = -ENOENT;
3268 		d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
3269 		goto failure;
3270 	}
3271 	base->lcpa_size = resource_size(res);
3272 	base->phy_lcpa = res->start;
3273 
3274 	if (request_mem_region(res->start, resource_size(res),
3275 			       D40_NAME " I/O lcpa") == NULL) {
3276 		ret = -EBUSY;
3277 		d40_err(&pdev->dev,
3278 			"Failed to request LCPA region 0x%x-0x%x\n",
3279 			res->start, res->end);
3280 		goto failure;
3281 	}
3282 
3283 	/* We make use of ESRAM memory for this. */
3284 	val = readl(base->virtbase + D40_DREG_LCPA);
3285 	if (res->start != val && val != 0) {
3286 		dev_warn(&pdev->dev,
3287 			 "[%s] Mismatch LCPA dma 0x%x, def 0x%x\n",
3288 			 __func__, val, res->start);
3289 	} else
3290 		writel(res->start, base->virtbase + D40_DREG_LCPA);
3291 
3292 	base->lcpa_base = ioremap(res->start, resource_size(res));
3293 	if (!base->lcpa_base) {
3294 		ret = -ENOMEM;
3295 		d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
3296 		goto failure;
3297 	}
3298 	/* If lcla has to be located in ESRAM we don't need to allocate */
3299 	if (base->plat_data->use_esram_lcla) {
3300 		res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3301 							"lcla_esram");
3302 		if (!res) {
3303 			ret = -ENOENT;
3304 			d40_err(&pdev->dev,
3305 				"No \"lcla_esram\" memory resource\n");
3306 			goto failure;
3307 		}
3308 		base->lcla_pool.base = ioremap(res->start,
3309 						resource_size(res));
3310 		if (!base->lcla_pool.base) {
3311 			ret = -ENOMEM;
3312 			d40_err(&pdev->dev, "Failed to ioremap LCLA region\n");
3313 			goto failure;
3314 		}
3315 		writel(res->start, base->virtbase + D40_DREG_LCLA);
3316 
3317 	} else {
3318 		ret = d40_lcla_allocate(base);
3319 		if (ret) {
3320 			d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
3321 			goto failure;
3322 		}
3323 	}
3324 
3325 	spin_lock_init(&base->lcla_pool.lock);
3326 
3327 	base->irq = platform_get_irq(pdev, 0);
3328 
3329 	ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3330 	if (ret) {
3331 		d40_err(&pdev->dev, "No IRQ defined\n");
3332 		goto failure;
3333 	}
3334 
3335 	pm_runtime_irq_safe(base->dev);
3336 	pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3337 	pm_runtime_use_autosuspend(base->dev);
3338 	pm_runtime_enable(base->dev);
3339 	pm_runtime_resume(base->dev);
3340 
3341 	if (base->plat_data->use_esram_lcla) {
3342 
3343 		base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3344 		if (IS_ERR(base->lcpa_regulator)) {
3345 			d40_err(&pdev->dev, "Failed to get lcpa_regulator\n");
3346 			base->lcpa_regulator = NULL;
3347 			goto failure;
3348 		}
3349 
3350 		ret = regulator_enable(base->lcpa_regulator);
3351 		if (ret) {
3352 			d40_err(&pdev->dev,
3353 				"Failed to enable lcpa_regulator\n");
3354 			regulator_put(base->lcpa_regulator);
3355 			base->lcpa_regulator = NULL;
3356 			goto failure;
3357 		}
3358 	}
3359 
3360 	base->initialized = true;
3361 	err = d40_dmaengine_init(base, num_reserved_chans);
3362 	if (err)
3363 		goto failure;
3364 
3365 	d40_hw_init(base);
3366 
3367 	dev_info(base->dev, "initialized\n");
3368 	return 0;
3369 
3370 failure:
3371 	if (base) {
3372 		if (base->desc_slab)
3373 			kmem_cache_destroy(base->desc_slab);
3374 		if (base->virtbase)
3375 			iounmap(base->virtbase);
3376 
3377 		if (base->lcla_pool.base && base->plat_data->use_esram_lcla) {
3378 			iounmap(base->lcla_pool.base);
3379 			base->lcla_pool.base = NULL;
3380 		}
3381 
3382 		if (base->lcla_pool.dma_addr)
3383 			dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3384 					 SZ_1K * base->num_phy_chans,
3385 					 DMA_TO_DEVICE);
3386 
3387 		if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3388 			free_pages((unsigned long)base->lcla_pool.base,
3389 				   base->lcla_pool.pages);
3390 
3391 		kfree(base->lcla_pool.base_unaligned);
3392 
3393 		if (base->phy_lcpa)
3394 			release_mem_region(base->phy_lcpa,
3395 					   base->lcpa_size);
3396 		if (base->phy_start)
3397 			release_mem_region(base->phy_start,
3398 					   base->phy_size);
3399 		if (base->clk) {
3400 			clk_disable(base->clk);
3401 			clk_put(base->clk);
3402 		}
3403 
3404 		if (base->lcpa_regulator) {
3405 			regulator_disable(base->lcpa_regulator);
3406 			regulator_put(base->lcpa_regulator);
3407 		}
3408 
3409 		kfree(base->lcla_pool.alloc_map);
3410 		kfree(base->lookup_log_chans);
3411 		kfree(base->lookup_phy_chans);
3412 		kfree(base->phy_res);
3413 		kfree(base);
3414 	}
3415 
3416 	d40_err(&pdev->dev, "probe failed\n");
3417 	return ret;
3418 }
3419 
3420 static struct platform_driver d40_driver = {
3421 	.driver = {
3422 		.owner = THIS_MODULE,
3423 		.name  = D40_NAME,
3424 		.pm = DMA40_PM_OPS,
3425 	},
3426 };
3427 
3428 static int __init stedma40_init(void)
3429 {
3430 	return platform_driver_probe(&d40_driver, d40_probe);
3431 }
3432 subsys_initcall(stedma40_init);
3433