xref: /openbmc/linux/drivers/dma/ste_dma40.c (revision 0246e77b4d374bb37aa08c3fcadad20012e85ea0)
1 /*
2  * driver/dma/ste_dma40.c
3  *
4  * Copyright (C) ST-Ericsson 2007-2010
5  * License terms: GNU General Public License (GPL) version 2
6  * Author: Per Friden <per.friden@stericsson.com>
7  * Author: Jonas Aaberg <jonas.aberg@stericsson.com>
8  *
9  */
10 
11 #include <linux/kernel.h>
12 #include <linux/slab.h>
13 #include <linux/dmaengine.h>
14 #include <linux/platform_device.h>
15 #include <linux/clk.h>
16 #include <linux/delay.h>
17 
18 #include <plat/ste_dma40.h>
19 
20 #include "ste_dma40_ll.h"
21 
22 #define D40_NAME "dma40"
23 
24 #define D40_PHY_CHAN -1
25 
26 /* For masking out/in 2 bit channel positions */
27 #define D40_CHAN_POS(chan)  (2 * (chan / 2))
28 #define D40_CHAN_POS_MASK(chan) (0x3 << D40_CHAN_POS(chan))
29 
30 /* Maximum iterations taken before giving up suspending a channel */
31 #define D40_SUSPEND_MAX_IT 500
32 
33 /* Hardware requirement on LCLA alignment */
34 #define LCLA_ALIGNMENT 0x40000
35 /* Attempts before giving up to trying to get pages that are aligned */
36 #define MAX_LCLA_ALLOC_ATTEMPTS 256
37 
38 /* Bit markings for allocation map */
39 #define D40_ALLOC_FREE		(1 << 31)
40 #define D40_ALLOC_PHY		(1 << 30)
41 #define D40_ALLOC_LOG_FREE	0
42 
43 /* Hardware designer of the block */
44 #define D40_PERIPHID2_DESIGNER 0x8
45 
46 /**
47  * enum 40_command - The different commands and/or statuses.
48  *
49  * @D40_DMA_STOP: DMA channel command STOP or status STOPPED,
50  * @D40_DMA_RUN: The DMA channel is RUNNING of the command RUN.
51  * @D40_DMA_SUSPEND_REQ: Request the DMA to SUSPEND as soon as possible.
52  * @D40_DMA_SUSPENDED: The DMA channel is SUSPENDED.
53  */
54 enum d40_command {
55 	D40_DMA_STOP		= 0,
56 	D40_DMA_RUN		= 1,
57 	D40_DMA_SUSPEND_REQ	= 2,
58 	D40_DMA_SUSPENDED	= 3
59 };
60 
61 /**
62  * struct d40_lli_pool - Structure for keeping LLIs in memory
63  *
64  * @base: Pointer to memory area when the pre_alloc_lli's are not large
65  * enough, IE bigger than the most common case, 1 dst and 1 src. NULL if
66  * pre_alloc_lli is used.
67  * @size: The size in bytes of the memory at base or the size of pre_alloc_lli.
68  * @pre_alloc_lli: Pre allocated area for the most common case of transfers,
69  * one buffer to one buffer.
70  */
71 struct d40_lli_pool {
72 	void	*base;
73 	int	 size;
74 	/* Space for dst and src, plus an extra for padding */
75 	u8	 pre_alloc_lli[3 * sizeof(struct d40_phy_lli)];
76 };
77 
78 /**
79  * struct d40_desc - A descriptor is one DMA job.
80  *
81  * @lli_phy: LLI settings for physical channel. Both src and dst=
82  * points into the lli_pool, to base if lli_len > 1 or to pre_alloc_lli if
83  * lli_len equals one.
84  * @lli_log: Same as above but for logical channels.
85  * @lli_pool: The pool with two entries pre-allocated.
86  * @lli_len: Number of llis of current descriptor.
87  * @lli_count: Number of transfered llis.
88  * @lli_tx_len: Max number of LLIs per transfer, there can be
89  * many transfer for one descriptor.
90  * @txd: DMA engine struct. Used for among other things for communication
91  * during a transfer.
92  * @node: List entry.
93  * @dir: The transfer direction of this job.
94  * @is_in_client_list: true if the client owns this descriptor.
95  *
96  * This descriptor is used for both logical and physical transfers.
97  */
98 
99 struct d40_desc {
100 	/* LLI physical */
101 	struct d40_phy_lli_bidir	 lli_phy;
102 	/* LLI logical */
103 	struct d40_log_lli_bidir	 lli_log;
104 
105 	struct d40_lli_pool		 lli_pool;
106 	int				 lli_len;
107 	int				 lli_count;
108 	u32				 lli_tx_len;
109 
110 	struct dma_async_tx_descriptor	 txd;
111 	struct list_head		 node;
112 
113 	enum dma_data_direction		 dir;
114 	bool				 is_in_client_list;
115 };
116 
117 /**
118  * struct d40_lcla_pool - LCLA pool settings and data.
119  *
120  * @base: The virtual address of LCLA. 18 bit aligned.
121  * @base_unaligned: The orignal kmalloc pointer, if kmalloc is used.
122  * This pointer is only there for clean-up on error.
123  * @pages: The number of pages needed for all physical channels.
124  * Only used later for clean-up on error
125  * @lock: Lock to protect the content in this struct.
126  * @alloc_map: Bitmap mapping between physical channel and LCLA entries.
127  * @num_blocks: The number of entries of alloc_map. Equals to the
128  * number of physical channels.
129  */
130 struct d40_lcla_pool {
131 	void		*base;
132 	void		*base_unaligned;
133 	int		 pages;
134 	spinlock_t	 lock;
135 	u32		*alloc_map;
136 	int		 num_blocks;
137 };
138 
139 /**
140  * struct d40_phy_res - struct for handling eventlines mapped to physical
141  * channels.
142  *
143  * @lock: A lock protection this entity.
144  * @num: The physical channel number of this entity.
145  * @allocated_src: Bit mapped to show which src event line's are mapped to
146  * this physical channel. Can also be free or physically allocated.
147  * @allocated_dst: Same as for src but is dst.
148  * allocated_dst and allocated_src uses the D40_ALLOC* defines as well as
149  * event line number. Both allocated_src and allocated_dst can not be
150  * allocated to a physical channel, since the interrupt handler has then
151  * no way of figure out which one the interrupt belongs to.
152  */
153 struct d40_phy_res {
154 	spinlock_t lock;
155 	int	   num;
156 	u32	   allocated_src;
157 	u32	   allocated_dst;
158 };
159 
160 struct d40_base;
161 
162 /**
163  * struct d40_chan - Struct that describes a channel.
164  *
165  * @lock: A spinlock to protect this struct.
166  * @log_num: The logical number, if any of this channel.
167  * @completed: Starts with 1, after first interrupt it is set to dma engine's
168  * current cookie.
169  * @pending_tx: The number of pending transfers. Used between interrupt handler
170  * and tasklet.
171  * @busy: Set to true when transfer is ongoing on this channel.
172  * @phy_chan: Pointer to physical channel which this instance runs on. If this
173  * point is NULL, then the channel is not allocated.
174  * @chan: DMA engine handle.
175  * @tasklet: Tasklet that gets scheduled from interrupt context to complete a
176  * transfer and call client callback.
177  * @client: Cliented owned descriptor list.
178  * @active: Active descriptor.
179  * @queue: Queued jobs.
180  * @dma_cfg: The client configuration of this dma channel.
181  * @base: Pointer to the device instance struct.
182  * @src_def_cfg: Default cfg register setting for src.
183  * @dst_def_cfg: Default cfg register setting for dst.
184  * @log_def: Default logical channel settings.
185  * @lcla: Space for one dst src pair for logical channel transfers.
186  * @lcpa: Pointer to dst and src lcpa settings.
187  *
188  * This struct can either "be" a logical or a physical channel.
189  */
190 struct d40_chan {
191 	spinlock_t			 lock;
192 	int				 log_num;
193 	/* ID of the most recent completed transfer */
194 	int				 completed;
195 	int				 pending_tx;
196 	bool				 busy;
197 	struct d40_phy_res		*phy_chan;
198 	struct dma_chan			 chan;
199 	struct tasklet_struct		 tasklet;
200 	struct list_head		 client;
201 	struct list_head		 active;
202 	struct list_head		 queue;
203 	struct stedma40_chan_cfg	 dma_cfg;
204 	struct d40_base			*base;
205 	/* Default register configurations */
206 	u32				 src_def_cfg;
207 	u32				 dst_def_cfg;
208 	struct d40_def_lcsp		 log_def;
209 	struct d40_lcla_elem		 lcla;
210 	struct d40_log_lli_full		*lcpa;
211 	/* Runtime reconfiguration */
212 	dma_addr_t			runtime_addr;
213 	enum dma_data_direction		runtime_direction;
214 };
215 
216 /**
217  * struct d40_base - The big global struct, one for each probe'd instance.
218  *
219  * @interrupt_lock: Lock used to make sure one interrupt is handle a time.
220  * @execmd_lock: Lock for execute command usage since several channels share
221  * the same physical register.
222  * @dev: The device structure.
223  * @virtbase: The virtual base address of the DMA's register.
224  * @rev: silicon revision detected.
225  * @clk: Pointer to the DMA clock structure.
226  * @phy_start: Physical memory start of the DMA registers.
227  * @phy_size: Size of the DMA register map.
228  * @irq: The IRQ number.
229  * @num_phy_chans: The number of physical channels. Read from HW. This
230  * is the number of available channels for this driver, not counting "Secure
231  * mode" allocated physical channels.
232  * @num_log_chans: The number of logical channels. Calculated from
233  * num_phy_chans.
234  * @dma_both: dma_device channels that can do both memcpy and slave transfers.
235  * @dma_slave: dma_device channels that can do only do slave transfers.
236  * @dma_memcpy: dma_device channels that can do only do memcpy transfers.
237  * @phy_chans: Room for all possible physical channels in system.
238  * @log_chans: Room for all possible logical channels in system.
239  * @lookup_log_chans: Used to map interrupt number to logical channel. Points
240  * to log_chans entries.
241  * @lookup_phy_chans: Used to map interrupt number to physical channel. Points
242  * to phy_chans entries.
243  * @plat_data: Pointer to provided platform_data which is the driver
244  * configuration.
245  * @phy_res: Vector containing all physical channels.
246  * @lcla_pool: lcla pool settings and data.
247  * @lcpa_base: The virtual mapped address of LCPA.
248  * @phy_lcpa: The physical address of the LCPA.
249  * @lcpa_size: The size of the LCPA area.
250  * @desc_slab: cache for descriptors.
251  */
252 struct d40_base {
253 	spinlock_t			 interrupt_lock;
254 	spinlock_t			 execmd_lock;
255 	struct device			 *dev;
256 	void __iomem			 *virtbase;
257 	u8				  rev:4;
258 	struct clk			 *clk;
259 	phys_addr_t			  phy_start;
260 	resource_size_t			  phy_size;
261 	int				  irq;
262 	int				  num_phy_chans;
263 	int				  num_log_chans;
264 	struct dma_device		  dma_both;
265 	struct dma_device		  dma_slave;
266 	struct dma_device		  dma_memcpy;
267 	struct d40_chan			 *phy_chans;
268 	struct d40_chan			 *log_chans;
269 	struct d40_chan			**lookup_log_chans;
270 	struct d40_chan			**lookup_phy_chans;
271 	struct stedma40_platform_data	 *plat_data;
272 	/* Physical half channels */
273 	struct d40_phy_res		 *phy_res;
274 	struct d40_lcla_pool		  lcla_pool;
275 	void				 *lcpa_base;
276 	dma_addr_t			  phy_lcpa;
277 	resource_size_t			  lcpa_size;
278 	struct kmem_cache		 *desc_slab;
279 };
280 
281 /**
282  * struct d40_interrupt_lookup - lookup table for interrupt handler
283  *
284  * @src: Interrupt mask register.
285  * @clr: Interrupt clear register.
286  * @is_error: true if this is an error interrupt.
287  * @offset: start delta in the lookup_log_chans in d40_base. If equals to
288  * D40_PHY_CHAN, the lookup_phy_chans shall be used instead.
289  */
290 struct d40_interrupt_lookup {
291 	u32 src;
292 	u32 clr;
293 	bool is_error;
294 	int offset;
295 };
296 
297 /**
298  * struct d40_reg_val - simple lookup struct
299  *
300  * @reg: The register.
301  * @val: The value that belongs to the register in reg.
302  */
303 struct d40_reg_val {
304 	unsigned int reg;
305 	unsigned int val;
306 };
307 
308 static int d40_pool_lli_alloc(struct d40_desc *d40d,
309 			      int lli_len, bool is_log)
310 {
311 	u32 align;
312 	void *base;
313 
314 	if (is_log)
315 		align = sizeof(struct d40_log_lli);
316 	else
317 		align = sizeof(struct d40_phy_lli);
318 
319 	if (lli_len == 1) {
320 		base = d40d->lli_pool.pre_alloc_lli;
321 		d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
322 		d40d->lli_pool.base = NULL;
323 	} else {
324 		d40d->lli_pool.size = ALIGN(lli_len * 2 * align, align);
325 
326 		base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
327 		d40d->lli_pool.base = base;
328 
329 		if (d40d->lli_pool.base == NULL)
330 			return -ENOMEM;
331 	}
332 
333 	if (is_log) {
334 		d40d->lli_log.src = PTR_ALIGN((struct d40_log_lli *) base,
335 					      align);
336 		d40d->lli_log.dst = PTR_ALIGN(d40d->lli_log.src + lli_len,
337 					      align);
338 	} else {
339 		d40d->lli_phy.src = PTR_ALIGN((struct d40_phy_lli *)base,
340 					      align);
341 		d40d->lli_phy.dst = PTR_ALIGN(d40d->lli_phy.src + lli_len,
342 					      align);
343 
344 		d40d->lli_phy.src_addr = virt_to_phys(d40d->lli_phy.src);
345 		d40d->lli_phy.dst_addr = virt_to_phys(d40d->lli_phy.dst);
346 	}
347 
348 	return 0;
349 }
350 
351 static void d40_pool_lli_free(struct d40_desc *d40d)
352 {
353 	kfree(d40d->lli_pool.base);
354 	d40d->lli_pool.base = NULL;
355 	d40d->lli_pool.size = 0;
356 	d40d->lli_log.src = NULL;
357 	d40d->lli_log.dst = NULL;
358 	d40d->lli_phy.src = NULL;
359 	d40d->lli_phy.dst = NULL;
360 	d40d->lli_phy.src_addr = 0;
361 	d40d->lli_phy.dst_addr = 0;
362 }
363 
364 static dma_cookie_t d40_assign_cookie(struct d40_chan *d40c,
365 				      struct d40_desc *desc)
366 {
367 	dma_cookie_t cookie = d40c->chan.cookie;
368 
369 	if (++cookie < 0)
370 		cookie = 1;
371 
372 	d40c->chan.cookie = cookie;
373 	desc->txd.cookie = cookie;
374 
375 	return cookie;
376 }
377 
378 static void d40_desc_remove(struct d40_desc *d40d)
379 {
380 	list_del(&d40d->node);
381 }
382 
383 static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
384 {
385 	struct d40_desc *d;
386 	struct d40_desc *_d;
387 
388 	if (!list_empty(&d40c->client)) {
389 		list_for_each_entry_safe(d, _d, &d40c->client, node)
390 			if (async_tx_test_ack(&d->txd)) {
391 				d40_pool_lli_free(d);
392 				d40_desc_remove(d);
393 				break;
394 			}
395 	} else {
396 		d = kmem_cache_alloc(d40c->base->desc_slab, GFP_NOWAIT);
397 		if (d != NULL) {
398 			memset(d, 0, sizeof(struct d40_desc));
399 			INIT_LIST_HEAD(&d->node);
400 		}
401 	}
402 	return d;
403 }
404 
405 static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
406 {
407 	kmem_cache_free(d40c->base->desc_slab, d40d);
408 }
409 
410 static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
411 {
412 	list_add_tail(&desc->node, &d40c->active);
413 }
414 
415 static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
416 {
417 	struct d40_desc *d;
418 
419 	if (list_empty(&d40c->active))
420 		return NULL;
421 
422 	d = list_first_entry(&d40c->active,
423 			     struct d40_desc,
424 			     node);
425 	return d;
426 }
427 
428 static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
429 {
430 	list_add_tail(&desc->node, &d40c->queue);
431 }
432 
433 static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
434 {
435 	struct d40_desc *d;
436 
437 	if (list_empty(&d40c->queue))
438 		return NULL;
439 
440 	d = list_first_entry(&d40c->queue,
441 			     struct d40_desc,
442 			     node);
443 	return d;
444 }
445 
446 /* Support functions for logical channels */
447 
448 static int d40_lcla_id_get(struct d40_chan *d40c)
449 {
450 	int src_id = 0;
451 	int dst_id = 0;
452 	struct d40_log_lli *lcla_lidx_base =
453 		d40c->base->lcla_pool.base + d40c->phy_chan->num * 1024;
454 	int i;
455 	int lli_per_log = d40c->base->plat_data->llis_per_log;
456 	unsigned long flags;
457 
458 	if (d40c->lcla.src_id >= 0 && d40c->lcla.dst_id >= 0)
459 		return 0;
460 
461 	if (d40c->base->lcla_pool.num_blocks > 32)
462 		return -EINVAL;
463 
464 	spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
465 
466 	for (i = 0; i < d40c->base->lcla_pool.num_blocks; i++) {
467 		if (!(d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num] &
468 		      (0x1 << i))) {
469 			d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num] |=
470 				(0x1 << i);
471 			break;
472 		}
473 	}
474 	src_id = i;
475 	if (src_id >= d40c->base->lcla_pool.num_blocks)
476 		goto err;
477 
478 	for (; i < d40c->base->lcla_pool.num_blocks; i++) {
479 		if (!(d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num] &
480 		      (0x1 << i))) {
481 			d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num] |=
482 				(0x1 << i);
483 			break;
484 		}
485 	}
486 
487 	dst_id = i;
488 	if (dst_id == src_id)
489 		goto err;
490 
491 	d40c->lcla.src_id = src_id;
492 	d40c->lcla.dst_id = dst_id;
493 	d40c->lcla.dst = lcla_lidx_base + dst_id * lli_per_log + 1;
494 	d40c->lcla.src = lcla_lidx_base + src_id * lli_per_log + 1;
495 
496 	spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
497 	return 0;
498 err:
499 	spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
500 	return -EINVAL;
501 }
502 
503 
504 static int d40_channel_execute_command(struct d40_chan *d40c,
505 				       enum d40_command command)
506 {
507 	int status, i;
508 	void __iomem *active_reg;
509 	int ret = 0;
510 	unsigned long flags;
511 	u32 wmask;
512 
513 	spin_lock_irqsave(&d40c->base->execmd_lock, flags);
514 
515 	if (d40c->phy_chan->num % 2 == 0)
516 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
517 	else
518 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
519 
520 	if (command == D40_DMA_SUSPEND_REQ) {
521 		status = (readl(active_reg) &
522 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
523 			D40_CHAN_POS(d40c->phy_chan->num);
524 
525 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
526 			goto done;
527 	}
528 
529 	wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
530 	writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
531 	       active_reg);
532 
533 	if (command == D40_DMA_SUSPEND_REQ) {
534 
535 		for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
536 			status = (readl(active_reg) &
537 				  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
538 				D40_CHAN_POS(d40c->phy_chan->num);
539 
540 			cpu_relax();
541 			/*
542 			 * Reduce the number of bus accesses while
543 			 * waiting for the DMA to suspend.
544 			 */
545 			udelay(3);
546 
547 			if (status == D40_DMA_STOP ||
548 			    status == D40_DMA_SUSPENDED)
549 				break;
550 		}
551 
552 		if (i == D40_SUSPEND_MAX_IT) {
553 			dev_err(&d40c->chan.dev->device,
554 				"[%s]: unable to suspend the chl %d (log: %d) status %x\n",
555 				__func__, d40c->phy_chan->num, d40c->log_num,
556 				status);
557 			dump_stack();
558 			ret = -EBUSY;
559 		}
560 
561 	}
562 done:
563 	spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
564 	return ret;
565 }
566 
567 static void d40_term_all(struct d40_chan *d40c)
568 {
569 	struct d40_desc *d40d;
570 	unsigned long flags;
571 
572 	/* Release active descriptors */
573 	while ((d40d = d40_first_active_get(d40c))) {
574 		d40_desc_remove(d40d);
575 
576 		/* Return desc to free-list */
577 		d40_desc_free(d40c, d40d);
578 	}
579 
580 	/* Release queued descriptors waiting for transfer */
581 	while ((d40d = d40_first_queued(d40c))) {
582 		d40_desc_remove(d40d);
583 
584 		/* Return desc to free-list */
585 		d40_desc_free(d40c, d40d);
586 	}
587 
588 	spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
589 
590 	d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num] &=
591 		(~(0x1 << d40c->lcla.dst_id));
592 	d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num] &=
593 		(~(0x1 << d40c->lcla.src_id));
594 
595 	d40c->lcla.src_id = -1;
596 	d40c->lcla.dst_id = -1;
597 
598 	spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
599 
600 	d40c->pending_tx = 0;
601 	d40c->busy = false;
602 }
603 
604 static void d40_config_set_event(struct d40_chan *d40c, bool do_enable)
605 {
606 	u32 val;
607 	unsigned long flags;
608 
609 	/* Notice, that disable requires the physical channel to be stopped */
610 	if (do_enable)
611 		val = D40_ACTIVATE_EVENTLINE;
612 	else
613 		val = D40_DEACTIVATE_EVENTLINE;
614 
615 	spin_lock_irqsave(&d40c->phy_chan->lock, flags);
616 
617 	/* Enable event line connected to device (or memcpy) */
618 	if ((d40c->dma_cfg.dir ==  STEDMA40_PERIPH_TO_MEM) ||
619 	    (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH)) {
620 		u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
621 
622 		writel((val << D40_EVENTLINE_POS(event)) |
623 		       ~D40_EVENTLINE_MASK(event),
624 		       d40c->base->virtbase + D40_DREG_PCBASE +
625 		       d40c->phy_chan->num * D40_DREG_PCDELTA +
626 		       D40_CHAN_REG_SSLNK);
627 	}
628 	if (d40c->dma_cfg.dir !=  STEDMA40_PERIPH_TO_MEM) {
629 		u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
630 
631 		writel((val << D40_EVENTLINE_POS(event)) |
632 		       ~D40_EVENTLINE_MASK(event),
633 		       d40c->base->virtbase + D40_DREG_PCBASE +
634 		       d40c->phy_chan->num * D40_DREG_PCDELTA +
635 		       D40_CHAN_REG_SDLNK);
636 	}
637 
638 	spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
639 }
640 
641 static u32 d40_chan_has_events(struct d40_chan *d40c)
642 {
643 	u32 val;
644 
645 	val = readl(d40c->base->virtbase + D40_DREG_PCBASE +
646 		    d40c->phy_chan->num * D40_DREG_PCDELTA +
647 		    D40_CHAN_REG_SSLNK);
648 
649 	val |= readl(d40c->base->virtbase + D40_DREG_PCBASE +
650 		     d40c->phy_chan->num * D40_DREG_PCDELTA +
651 		     D40_CHAN_REG_SDLNK);
652 	return val;
653 }
654 
655 static void d40_config_write(struct d40_chan *d40c)
656 {
657 	u32 addr_base;
658 	u32 var;
659 
660 	/* Odd addresses are even addresses + 4 */
661 	addr_base = (d40c->phy_chan->num % 2) * 4;
662 	/* Setup channel mode to logical or physical */
663 	var = ((u32)(d40c->log_num != D40_PHY_CHAN) + 1) <<
664 		D40_CHAN_POS(d40c->phy_chan->num);
665 	writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
666 
667 	/* Setup operational mode option register */
668 	var = ((d40c->dma_cfg.channel_type >> STEDMA40_INFO_CH_MODE_OPT_POS) &
669 	       0x3) << D40_CHAN_POS(d40c->phy_chan->num);
670 
671 	writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
672 
673 	if (d40c->log_num != D40_PHY_CHAN) {
674 		/* Set default config for CFG reg */
675 		writel(d40c->src_def_cfg,
676 		       d40c->base->virtbase + D40_DREG_PCBASE +
677 		       d40c->phy_chan->num * D40_DREG_PCDELTA +
678 		       D40_CHAN_REG_SSCFG);
679 		writel(d40c->dst_def_cfg,
680 		       d40c->base->virtbase + D40_DREG_PCBASE +
681 		       d40c->phy_chan->num * D40_DREG_PCDELTA +
682 		       D40_CHAN_REG_SDCFG);
683 
684 		/* Set LIDX for lcla */
685 		writel((d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS) &
686 		       D40_SREG_ELEM_LOG_LIDX_MASK,
687 		       d40c->base->virtbase + D40_DREG_PCBASE +
688 		       d40c->phy_chan->num * D40_DREG_PCDELTA +
689 		       D40_CHAN_REG_SDELT);
690 
691 		writel((d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS) &
692 		       D40_SREG_ELEM_LOG_LIDX_MASK,
693 		       d40c->base->virtbase + D40_DREG_PCBASE +
694 		       d40c->phy_chan->num * D40_DREG_PCDELTA +
695 		       D40_CHAN_REG_SSELT);
696 
697 	}
698 }
699 
700 static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
701 {
702 	if (d40d->lli_phy.dst && d40d->lli_phy.src) {
703 		d40_phy_lli_write(d40c->base->virtbase,
704 				  d40c->phy_chan->num,
705 				  d40d->lli_phy.dst,
706 				  d40d->lli_phy.src);
707 	} else if (d40d->lli_log.dst && d40d->lli_log.src) {
708 		struct d40_log_lli *src = d40d->lli_log.src;
709 		struct d40_log_lli *dst = d40d->lli_log.dst;
710 		int s;
711 
712 		src += d40d->lli_count;
713 		dst += d40d->lli_count;
714 		s = d40_log_lli_write(d40c->lcpa,
715 				      d40c->lcla.src, d40c->lcla.dst,
716 				      dst, src,
717 				      d40c->base->plat_data->llis_per_log);
718 
719 		/* If s equals to zero, the job is not linked */
720 		if (s > 0) {
721 			(void) dma_map_single(d40c->base->dev, d40c->lcla.src,
722 					      s * sizeof(struct d40_log_lli),
723 					      DMA_TO_DEVICE);
724 			(void) dma_map_single(d40c->base->dev, d40c->lcla.dst,
725 					      s * sizeof(struct d40_log_lli),
726 					      DMA_TO_DEVICE);
727 		}
728 	}
729 	d40d->lli_count += d40d->lli_tx_len;
730 }
731 
732 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
733 {
734 	struct d40_chan *d40c = container_of(tx->chan,
735 					     struct d40_chan,
736 					     chan);
737 	struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
738 	unsigned long flags;
739 
740 	spin_lock_irqsave(&d40c->lock, flags);
741 
742 	tx->cookie = d40_assign_cookie(d40c, d40d);
743 
744 	d40_desc_queue(d40c, d40d);
745 
746 	spin_unlock_irqrestore(&d40c->lock, flags);
747 
748 	return tx->cookie;
749 }
750 
751 static int d40_start(struct d40_chan *d40c)
752 {
753 	if (d40c->base->rev == 0) {
754 		int err;
755 
756 		if (d40c->log_num != D40_PHY_CHAN) {
757 			err = d40_channel_execute_command(d40c,
758 							  D40_DMA_SUSPEND_REQ);
759 			if (err)
760 				return err;
761 		}
762 	}
763 
764 	if (d40c->log_num != D40_PHY_CHAN)
765 		d40_config_set_event(d40c, true);
766 
767 	return d40_channel_execute_command(d40c, D40_DMA_RUN);
768 }
769 
770 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
771 {
772 	struct d40_desc *d40d;
773 	int err;
774 
775 	/* Start queued jobs, if any */
776 	d40d = d40_first_queued(d40c);
777 
778 	if (d40d != NULL) {
779 		d40c->busy = true;
780 
781 		/* Remove from queue */
782 		d40_desc_remove(d40d);
783 
784 		/* Add to active queue */
785 		d40_desc_submit(d40c, d40d);
786 
787 		/* Initiate DMA job */
788 		d40_desc_load(d40c, d40d);
789 
790 		/* Start dma job */
791 		err = d40_start(d40c);
792 
793 		if (err)
794 			return NULL;
795 	}
796 
797 	return d40d;
798 }
799 
800 /* called from interrupt context */
801 static void dma_tc_handle(struct d40_chan *d40c)
802 {
803 	struct d40_desc *d40d;
804 
805 	if (!d40c->phy_chan)
806 		return;
807 
808 	/* Get first active entry from list */
809 	d40d = d40_first_active_get(d40c);
810 
811 	if (d40d == NULL)
812 		return;
813 
814 	if (d40d->lli_count < d40d->lli_len) {
815 
816 		d40_desc_load(d40c, d40d);
817 		/* Start dma job */
818 		(void) d40_start(d40c);
819 		return;
820 	}
821 
822 	if (d40_queue_start(d40c) == NULL)
823 		d40c->busy = false;
824 
825 	d40c->pending_tx++;
826 	tasklet_schedule(&d40c->tasklet);
827 
828 }
829 
830 static void dma_tasklet(unsigned long data)
831 {
832 	struct d40_chan *d40c = (struct d40_chan *) data;
833 	struct d40_desc *d40d_fin;
834 	unsigned long flags;
835 	dma_async_tx_callback callback;
836 	void *callback_param;
837 
838 	spin_lock_irqsave(&d40c->lock, flags);
839 
840 	/* Get first active entry from list */
841 	d40d_fin = d40_first_active_get(d40c);
842 
843 	if (d40d_fin == NULL)
844 		goto err;
845 
846 	d40c->completed = d40d_fin->txd.cookie;
847 
848 	/*
849 	 * If terminating a channel pending_tx is set to zero.
850 	 * This prevents any finished active jobs to return to the client.
851 	 */
852 	if (d40c->pending_tx == 0) {
853 		spin_unlock_irqrestore(&d40c->lock, flags);
854 		return;
855 	}
856 
857 	/* Callback to client */
858 	callback = d40d_fin->txd.callback;
859 	callback_param = d40d_fin->txd.callback_param;
860 
861 	if (async_tx_test_ack(&d40d_fin->txd)) {
862 		d40_pool_lli_free(d40d_fin);
863 		d40_desc_remove(d40d_fin);
864 		/* Return desc to free-list */
865 		d40_desc_free(d40c, d40d_fin);
866 	} else {
867 		if (!d40d_fin->is_in_client_list) {
868 			d40_desc_remove(d40d_fin);
869 			list_add_tail(&d40d_fin->node, &d40c->client);
870 			d40d_fin->is_in_client_list = true;
871 		}
872 	}
873 
874 	d40c->pending_tx--;
875 
876 	if (d40c->pending_tx)
877 		tasklet_schedule(&d40c->tasklet);
878 
879 	spin_unlock_irqrestore(&d40c->lock, flags);
880 
881 	if (callback && (d40d_fin->txd.flags & DMA_PREP_INTERRUPT))
882 		callback(callback_param);
883 
884 	return;
885 
886  err:
887 	/* Rescue manouver if receiving double interrupts */
888 	if (d40c->pending_tx > 0)
889 		d40c->pending_tx--;
890 	spin_unlock_irqrestore(&d40c->lock, flags);
891 }
892 
893 static irqreturn_t d40_handle_interrupt(int irq, void *data)
894 {
895 	static const struct d40_interrupt_lookup il[] = {
896 		{D40_DREG_LCTIS0, D40_DREG_LCICR0, false,  0},
897 		{D40_DREG_LCTIS1, D40_DREG_LCICR1, false, 32},
898 		{D40_DREG_LCTIS2, D40_DREG_LCICR2, false, 64},
899 		{D40_DREG_LCTIS3, D40_DREG_LCICR3, false, 96},
900 		{D40_DREG_LCEIS0, D40_DREG_LCICR0, true,   0},
901 		{D40_DREG_LCEIS1, D40_DREG_LCICR1, true,  32},
902 		{D40_DREG_LCEIS2, D40_DREG_LCICR2, true,  64},
903 		{D40_DREG_LCEIS3, D40_DREG_LCICR3, true,  96},
904 		{D40_DREG_PCTIS,  D40_DREG_PCICR,  false, D40_PHY_CHAN},
905 		{D40_DREG_PCEIS,  D40_DREG_PCICR,  true,  D40_PHY_CHAN},
906 	};
907 
908 	int i;
909 	u32 regs[ARRAY_SIZE(il)];
910 	u32 idx;
911 	u32 row;
912 	long chan = -1;
913 	struct d40_chan *d40c;
914 	unsigned long flags;
915 	struct d40_base *base = data;
916 
917 	spin_lock_irqsave(&base->interrupt_lock, flags);
918 
919 	/* Read interrupt status of both logical and physical channels */
920 	for (i = 0; i < ARRAY_SIZE(il); i++)
921 		regs[i] = readl(base->virtbase + il[i].src);
922 
923 	for (;;) {
924 
925 		chan = find_next_bit((unsigned long *)regs,
926 				     BITS_PER_LONG * ARRAY_SIZE(il), chan + 1);
927 
928 		/* No more set bits found? */
929 		if (chan == BITS_PER_LONG * ARRAY_SIZE(il))
930 			break;
931 
932 		row = chan / BITS_PER_LONG;
933 		idx = chan & (BITS_PER_LONG - 1);
934 
935 		/* ACK interrupt */
936 		writel(1 << idx, base->virtbase + il[row].clr);
937 
938 		if (il[row].offset == D40_PHY_CHAN)
939 			d40c = base->lookup_phy_chans[idx];
940 		else
941 			d40c = base->lookup_log_chans[il[row].offset + idx];
942 		spin_lock(&d40c->lock);
943 
944 		if (!il[row].is_error)
945 			dma_tc_handle(d40c);
946 		else
947 			dev_err(base->dev,
948 				"[%s] IRQ chan: %ld offset %d idx %d\n",
949 				__func__, chan, il[row].offset, idx);
950 
951 		spin_unlock(&d40c->lock);
952 	}
953 
954 	spin_unlock_irqrestore(&base->interrupt_lock, flags);
955 
956 	return IRQ_HANDLED;
957 }
958 
959 
960 static int d40_validate_conf(struct d40_chan *d40c,
961 			     struct stedma40_chan_cfg *conf)
962 {
963 	int res = 0;
964 	u32 dst_event_group = D40_TYPE_TO_GROUP(conf->dst_dev_type);
965 	u32 src_event_group = D40_TYPE_TO_GROUP(conf->src_dev_type);
966 	bool is_log = (conf->channel_type & STEDMA40_CHANNEL_IN_OPER_MODE)
967 		== STEDMA40_CHANNEL_IN_LOG_MODE;
968 
969 	if (!conf->dir) {
970 		dev_err(&d40c->chan.dev->device, "[%s] Invalid direction.\n",
971 			__func__);
972 		res = -EINVAL;
973 	}
974 
975 	if (conf->dst_dev_type != STEDMA40_DEV_DST_MEMORY &&
976 	    d40c->base->plat_data->dev_tx[conf->dst_dev_type] == 0 &&
977 	    d40c->runtime_addr == 0) {
978 
979 		dev_err(&d40c->chan.dev->device,
980 			"[%s] Invalid TX channel address (%d)\n",
981 			__func__, conf->dst_dev_type);
982 		res = -EINVAL;
983 	}
984 
985 	if (conf->src_dev_type != STEDMA40_DEV_SRC_MEMORY &&
986 	    d40c->base->plat_data->dev_rx[conf->src_dev_type] == 0 &&
987 	    d40c->runtime_addr == 0) {
988 		dev_err(&d40c->chan.dev->device,
989 			"[%s] Invalid RX channel address (%d)\n",
990 			__func__, conf->src_dev_type);
991 		res = -EINVAL;
992 	}
993 
994 	if (conf->dir == STEDMA40_MEM_TO_PERIPH &&
995 	    dst_event_group == STEDMA40_DEV_DST_MEMORY) {
996 		dev_err(&d40c->chan.dev->device, "[%s] Invalid dst\n",
997 			__func__);
998 		res = -EINVAL;
999 	}
1000 
1001 	if (conf->dir == STEDMA40_PERIPH_TO_MEM &&
1002 	    src_event_group == STEDMA40_DEV_SRC_MEMORY) {
1003 		dev_err(&d40c->chan.dev->device, "[%s] Invalid src\n",
1004 			__func__);
1005 		res = -EINVAL;
1006 	}
1007 
1008 	if (src_event_group == STEDMA40_DEV_SRC_MEMORY &&
1009 	    dst_event_group == STEDMA40_DEV_DST_MEMORY && is_log) {
1010 		dev_err(&d40c->chan.dev->device,
1011 			"[%s] No event line\n", __func__);
1012 		res = -EINVAL;
1013 	}
1014 
1015 	if (conf->dir == STEDMA40_PERIPH_TO_PERIPH &&
1016 	    (src_event_group != dst_event_group)) {
1017 		dev_err(&d40c->chan.dev->device,
1018 			"[%s] Invalid event group\n", __func__);
1019 		res = -EINVAL;
1020 	}
1021 
1022 	if (conf->dir == STEDMA40_PERIPH_TO_PERIPH) {
1023 		/*
1024 		 * DMAC HW supports it. Will be added to this driver,
1025 		 * in case any dma client requires it.
1026 		 */
1027 		dev_err(&d40c->chan.dev->device,
1028 			"[%s] periph to periph not supported\n",
1029 			__func__);
1030 		res = -EINVAL;
1031 	}
1032 
1033 	return res;
1034 }
1035 
1036 static bool d40_alloc_mask_set(struct d40_phy_res *phy, bool is_src,
1037 			       int log_event_line, bool is_log)
1038 {
1039 	unsigned long flags;
1040 	spin_lock_irqsave(&phy->lock, flags);
1041 	if (!is_log) {
1042 		/* Physical interrupts are masked per physical full channel */
1043 		if (phy->allocated_src == D40_ALLOC_FREE &&
1044 		    phy->allocated_dst == D40_ALLOC_FREE) {
1045 			phy->allocated_dst = D40_ALLOC_PHY;
1046 			phy->allocated_src = D40_ALLOC_PHY;
1047 			goto found;
1048 		} else
1049 			goto not_found;
1050 	}
1051 
1052 	/* Logical channel */
1053 	if (is_src) {
1054 		if (phy->allocated_src == D40_ALLOC_PHY)
1055 			goto not_found;
1056 
1057 		if (phy->allocated_src == D40_ALLOC_FREE)
1058 			phy->allocated_src = D40_ALLOC_LOG_FREE;
1059 
1060 		if (!(phy->allocated_src & (1 << log_event_line))) {
1061 			phy->allocated_src |= 1 << log_event_line;
1062 			goto found;
1063 		} else
1064 			goto not_found;
1065 	} else {
1066 		if (phy->allocated_dst == D40_ALLOC_PHY)
1067 			goto not_found;
1068 
1069 		if (phy->allocated_dst == D40_ALLOC_FREE)
1070 			phy->allocated_dst = D40_ALLOC_LOG_FREE;
1071 
1072 		if (!(phy->allocated_dst & (1 << log_event_line))) {
1073 			phy->allocated_dst |= 1 << log_event_line;
1074 			goto found;
1075 		} else
1076 			goto not_found;
1077 	}
1078 
1079 not_found:
1080 	spin_unlock_irqrestore(&phy->lock, flags);
1081 	return false;
1082 found:
1083 	spin_unlock_irqrestore(&phy->lock, flags);
1084 	return true;
1085 }
1086 
1087 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1088 			       int log_event_line)
1089 {
1090 	unsigned long flags;
1091 	bool is_free = false;
1092 
1093 	spin_lock_irqsave(&phy->lock, flags);
1094 	if (!log_event_line) {
1095 		/* Physical interrupts are masked per physical full channel */
1096 		phy->allocated_dst = D40_ALLOC_FREE;
1097 		phy->allocated_src = D40_ALLOC_FREE;
1098 		is_free = true;
1099 		goto out;
1100 	}
1101 
1102 	/* Logical channel */
1103 	if (is_src) {
1104 		phy->allocated_src &= ~(1 << log_event_line);
1105 		if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1106 			phy->allocated_src = D40_ALLOC_FREE;
1107 	} else {
1108 		phy->allocated_dst &= ~(1 << log_event_line);
1109 		if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1110 			phy->allocated_dst = D40_ALLOC_FREE;
1111 	}
1112 
1113 	is_free = ((phy->allocated_src | phy->allocated_dst) ==
1114 		   D40_ALLOC_FREE);
1115 
1116 out:
1117 	spin_unlock_irqrestore(&phy->lock, flags);
1118 
1119 	return is_free;
1120 }
1121 
1122 static int d40_allocate_channel(struct d40_chan *d40c)
1123 {
1124 	int dev_type;
1125 	int event_group;
1126 	int event_line;
1127 	struct d40_phy_res *phys;
1128 	int i;
1129 	int j;
1130 	int log_num;
1131 	bool is_src;
1132 	bool is_log = (d40c->dma_cfg.channel_type &
1133 		       STEDMA40_CHANNEL_IN_OPER_MODE)
1134 		== STEDMA40_CHANNEL_IN_LOG_MODE;
1135 
1136 
1137 	phys = d40c->base->phy_res;
1138 
1139 	if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1140 		dev_type = d40c->dma_cfg.src_dev_type;
1141 		log_num = 2 * dev_type;
1142 		is_src = true;
1143 	} else if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1144 		   d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1145 		/* dst event lines are used for logical memcpy */
1146 		dev_type = d40c->dma_cfg.dst_dev_type;
1147 		log_num = 2 * dev_type + 1;
1148 		is_src = false;
1149 	} else
1150 		return -EINVAL;
1151 
1152 	event_group = D40_TYPE_TO_GROUP(dev_type);
1153 	event_line = D40_TYPE_TO_EVENT(dev_type);
1154 
1155 	if (!is_log) {
1156 		if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1157 			/* Find physical half channel */
1158 			for (i = 0; i < d40c->base->num_phy_chans; i++) {
1159 
1160 				if (d40_alloc_mask_set(&phys[i], is_src,
1161 						       0, is_log))
1162 					goto found_phy;
1163 			}
1164 		} else
1165 			for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1166 				int phy_num = j  + event_group * 2;
1167 				for (i = phy_num; i < phy_num + 2; i++) {
1168 					if (d40_alloc_mask_set(&phys[i],
1169 							       is_src,
1170 							       0,
1171 							       is_log))
1172 						goto found_phy;
1173 				}
1174 			}
1175 		return -EINVAL;
1176 found_phy:
1177 		d40c->phy_chan = &phys[i];
1178 		d40c->log_num = D40_PHY_CHAN;
1179 		goto out;
1180 	}
1181 	if (dev_type == -1)
1182 		return -EINVAL;
1183 
1184 	/* Find logical channel */
1185 	for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1186 		int phy_num = j + event_group * 2;
1187 		/*
1188 		 * Spread logical channels across all available physical rather
1189 		 * than pack every logical channel at the first available phy
1190 		 * channels.
1191 		 */
1192 		if (is_src) {
1193 			for (i = phy_num; i < phy_num + 2; i++) {
1194 				if (d40_alloc_mask_set(&phys[i], is_src,
1195 						       event_line, is_log))
1196 					goto found_log;
1197 			}
1198 		} else {
1199 			for (i = phy_num + 1; i >= phy_num; i--) {
1200 				if (d40_alloc_mask_set(&phys[i], is_src,
1201 						       event_line, is_log))
1202 					goto found_log;
1203 			}
1204 		}
1205 	}
1206 	return -EINVAL;
1207 
1208 found_log:
1209 	d40c->phy_chan = &phys[i];
1210 	d40c->log_num = log_num;
1211 out:
1212 
1213 	if (is_log)
1214 		d40c->base->lookup_log_chans[d40c->log_num] = d40c;
1215 	else
1216 		d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
1217 
1218 	return 0;
1219 
1220 }
1221 
1222 static int d40_config_memcpy(struct d40_chan *d40c)
1223 {
1224 	dma_cap_mask_t cap = d40c->chan.device->cap_mask;
1225 
1226 	if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
1227 		d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_log;
1228 		d40c->dma_cfg.src_dev_type = STEDMA40_DEV_SRC_MEMORY;
1229 		d40c->dma_cfg.dst_dev_type = d40c->base->plat_data->
1230 			memcpy[d40c->chan.chan_id];
1231 
1232 	} else if (dma_has_cap(DMA_MEMCPY, cap) &&
1233 		   dma_has_cap(DMA_SLAVE, cap)) {
1234 		d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_phy;
1235 	} else {
1236 		dev_err(&d40c->chan.dev->device, "[%s] No memcpy\n",
1237 			__func__);
1238 		return -EINVAL;
1239 	}
1240 
1241 	return 0;
1242 }
1243 
1244 
1245 static int d40_free_dma(struct d40_chan *d40c)
1246 {
1247 
1248 	int res = 0;
1249 	u32 event;
1250 	struct d40_phy_res *phy = d40c->phy_chan;
1251 	bool is_src;
1252 	struct d40_desc *d;
1253 	struct d40_desc *_d;
1254 
1255 
1256 	/* Terminate all queued and active transfers */
1257 	d40_term_all(d40c);
1258 
1259 	/* Release client owned descriptors */
1260 	if (!list_empty(&d40c->client))
1261 		list_for_each_entry_safe(d, _d, &d40c->client, node) {
1262 			d40_pool_lli_free(d);
1263 			d40_desc_remove(d);
1264 			/* Return desc to free-list */
1265 			d40_desc_free(d40c, d);
1266 		}
1267 
1268 	if (phy == NULL) {
1269 		dev_err(&d40c->chan.dev->device, "[%s] phy == null\n",
1270 			__func__);
1271 		return -EINVAL;
1272 	}
1273 
1274 	if (phy->allocated_src == D40_ALLOC_FREE &&
1275 	    phy->allocated_dst == D40_ALLOC_FREE) {
1276 		dev_err(&d40c->chan.dev->device, "[%s] channel already free\n",
1277 			__func__);
1278 		return -EINVAL;
1279 	}
1280 
1281 	if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1282 	    d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1283 		event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
1284 		is_src = false;
1285 	} else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1286 		event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
1287 		is_src = true;
1288 	} else {
1289 		dev_err(&d40c->chan.dev->device,
1290 			"[%s] Unknown direction\n", __func__);
1291 		return -EINVAL;
1292 	}
1293 
1294 	res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1295 	if (res) {
1296 		dev_err(&d40c->chan.dev->device, "[%s] suspend failed\n",
1297 			__func__);
1298 		return res;
1299 	}
1300 
1301 	if (d40c->log_num != D40_PHY_CHAN) {
1302 		/* Release logical channel, deactivate the event line */
1303 
1304 		d40_config_set_event(d40c, false);
1305 		d40c->base->lookup_log_chans[d40c->log_num] = NULL;
1306 
1307 		/*
1308 		 * Check if there are more logical allocation
1309 		 * on this phy channel.
1310 		 */
1311 		if (!d40_alloc_mask_free(phy, is_src, event)) {
1312 			/* Resume the other logical channels if any */
1313 			if (d40_chan_has_events(d40c)) {
1314 				res = d40_channel_execute_command(d40c,
1315 								  D40_DMA_RUN);
1316 				if (res) {
1317 					dev_err(&d40c->chan.dev->device,
1318 						"[%s] Executing RUN command\n",
1319 						__func__);
1320 					return res;
1321 				}
1322 			}
1323 			return 0;
1324 		}
1325 	} else {
1326 		(void) d40_alloc_mask_free(phy, is_src, 0);
1327 	}
1328 
1329 	/* Release physical channel */
1330 	res = d40_channel_execute_command(d40c, D40_DMA_STOP);
1331 	if (res) {
1332 		dev_err(&d40c->chan.dev->device,
1333 			"[%s] Failed to stop channel\n", __func__);
1334 		return res;
1335 	}
1336 	d40c->phy_chan = NULL;
1337 	/* Invalidate channel type */
1338 	d40c->dma_cfg.channel_type = 0;
1339 	d40c->base->lookup_phy_chans[phy->num] = NULL;
1340 
1341 	return 0;
1342 }
1343 
1344 static int d40_pause(struct dma_chan *chan)
1345 {
1346 	struct d40_chan *d40c =
1347 		container_of(chan, struct d40_chan, chan);
1348 	int res;
1349 	unsigned long flags;
1350 
1351 	spin_lock_irqsave(&d40c->lock, flags);
1352 
1353 	res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1354 	if (res == 0) {
1355 		if (d40c->log_num != D40_PHY_CHAN) {
1356 			d40_config_set_event(d40c, false);
1357 			/* Resume the other logical channels if any */
1358 			if (d40_chan_has_events(d40c))
1359 				res = d40_channel_execute_command(d40c,
1360 								  D40_DMA_RUN);
1361 		}
1362 	}
1363 
1364 	spin_unlock_irqrestore(&d40c->lock, flags);
1365 	return res;
1366 }
1367 
1368 static bool d40_is_paused(struct d40_chan *d40c)
1369 {
1370 	bool is_paused = false;
1371 	unsigned long flags;
1372 	void __iomem *active_reg;
1373 	u32 status;
1374 	u32 event;
1375 
1376 	spin_lock_irqsave(&d40c->lock, flags);
1377 
1378 	if (d40c->log_num == D40_PHY_CHAN) {
1379 		if (d40c->phy_chan->num % 2 == 0)
1380 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1381 		else
1382 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1383 
1384 		status = (readl(active_reg) &
1385 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1386 			D40_CHAN_POS(d40c->phy_chan->num);
1387 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1388 			is_paused = true;
1389 
1390 		goto _exit;
1391 	}
1392 
1393 	if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1394 	    d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM)
1395 		event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
1396 	else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM)
1397 		event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
1398 	else {
1399 		dev_err(&d40c->chan.dev->device,
1400 			"[%s] Unknown direction\n", __func__);
1401 		goto _exit;
1402 	}
1403 	status = d40_chan_has_events(d40c);
1404 	status = (status & D40_EVENTLINE_MASK(event)) >>
1405 		D40_EVENTLINE_POS(event);
1406 
1407 	if (status != D40_DMA_RUN)
1408 		is_paused = true;
1409 _exit:
1410 	spin_unlock_irqrestore(&d40c->lock, flags);
1411 	return is_paused;
1412 
1413 }
1414 
1415 
1416 static bool d40_tx_is_linked(struct d40_chan *d40c)
1417 {
1418 	bool is_link;
1419 
1420 	if (d40c->log_num != D40_PHY_CHAN)
1421 		is_link = readl(&d40c->lcpa->lcsp3) &  D40_MEM_LCSP3_DLOS_MASK;
1422 	else
1423 		is_link = readl(d40c->base->virtbase + D40_DREG_PCBASE +
1424 				d40c->phy_chan->num * D40_DREG_PCDELTA +
1425 				D40_CHAN_REG_SDLNK) &
1426 			D40_SREG_LNK_PHYS_LNK_MASK;
1427 	return is_link;
1428 }
1429 
1430 static u32 d40_residue(struct d40_chan *d40c)
1431 {
1432 	u32 num_elt;
1433 
1434 	if (d40c->log_num != D40_PHY_CHAN)
1435 		num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
1436 			>> D40_MEM_LCSP2_ECNT_POS;
1437 	else
1438 		num_elt = (readl(d40c->base->virtbase + D40_DREG_PCBASE +
1439 				 d40c->phy_chan->num * D40_DREG_PCDELTA +
1440 				 D40_CHAN_REG_SDELT) &
1441 			   D40_SREG_ELEM_PHY_ECNT_MASK) >>
1442 			D40_SREG_ELEM_PHY_ECNT_POS;
1443 	return num_elt * (1 << d40c->dma_cfg.dst_info.data_width);
1444 }
1445 
1446 static int d40_resume(struct dma_chan *chan)
1447 {
1448 	struct d40_chan *d40c =
1449 		container_of(chan, struct d40_chan, chan);
1450 	int res = 0;
1451 	unsigned long flags;
1452 
1453 	spin_lock_irqsave(&d40c->lock, flags);
1454 
1455 	if (d40c->base->rev == 0)
1456 		if (d40c->log_num != D40_PHY_CHAN) {
1457 			res = d40_channel_execute_command(d40c,
1458 							  D40_DMA_SUSPEND_REQ);
1459 			goto no_suspend;
1460 		}
1461 
1462 	/* If bytes left to transfer or linked tx resume job */
1463 	if (d40_residue(d40c) || d40_tx_is_linked(d40c)) {
1464 		if (d40c->log_num != D40_PHY_CHAN)
1465 			d40_config_set_event(d40c, true);
1466 		res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1467 	}
1468 
1469 no_suspend:
1470 	spin_unlock_irqrestore(&d40c->lock, flags);
1471 	return res;
1472 }
1473 
1474 static u32 stedma40_residue(struct dma_chan *chan)
1475 {
1476 	struct d40_chan *d40c =
1477 		container_of(chan, struct d40_chan, chan);
1478 	u32 bytes_left;
1479 	unsigned long flags;
1480 
1481 	spin_lock_irqsave(&d40c->lock, flags);
1482 	bytes_left = d40_residue(d40c);
1483 	spin_unlock_irqrestore(&d40c->lock, flags);
1484 
1485 	return bytes_left;
1486 }
1487 
1488 /* Public DMA functions in addition to the DMA engine framework */
1489 
1490 int stedma40_set_psize(struct dma_chan *chan,
1491 		       int src_psize,
1492 		       int dst_psize)
1493 {
1494 	struct d40_chan *d40c =
1495 		container_of(chan, struct d40_chan, chan);
1496 	unsigned long flags;
1497 
1498 	spin_lock_irqsave(&d40c->lock, flags);
1499 
1500 	if (d40c->log_num != D40_PHY_CHAN) {
1501 		d40c->log_def.lcsp1 &= ~D40_MEM_LCSP1_SCFG_PSIZE_MASK;
1502 		d40c->log_def.lcsp3 &= ~D40_MEM_LCSP1_SCFG_PSIZE_MASK;
1503 		d40c->log_def.lcsp1 |= src_psize <<
1504 			D40_MEM_LCSP1_SCFG_PSIZE_POS;
1505 		d40c->log_def.lcsp3 |= dst_psize <<
1506 			D40_MEM_LCSP1_SCFG_PSIZE_POS;
1507 		goto out;
1508 	}
1509 
1510 	if (src_psize == STEDMA40_PSIZE_PHY_1)
1511 		d40c->src_def_cfg &= ~(1 << D40_SREG_CFG_PHY_PEN_POS);
1512 	else {
1513 		d40c->src_def_cfg |= 1 << D40_SREG_CFG_PHY_PEN_POS;
1514 		d40c->src_def_cfg &= ~(STEDMA40_PSIZE_PHY_16 <<
1515 				       D40_SREG_CFG_PSIZE_POS);
1516 		d40c->src_def_cfg |= src_psize << D40_SREG_CFG_PSIZE_POS;
1517 	}
1518 
1519 	if (dst_psize == STEDMA40_PSIZE_PHY_1)
1520 		d40c->dst_def_cfg &= ~(1 << D40_SREG_CFG_PHY_PEN_POS);
1521 	else {
1522 		d40c->dst_def_cfg |= 1 << D40_SREG_CFG_PHY_PEN_POS;
1523 		d40c->dst_def_cfg &= ~(STEDMA40_PSIZE_PHY_16 <<
1524 				       D40_SREG_CFG_PSIZE_POS);
1525 		d40c->dst_def_cfg |= dst_psize << D40_SREG_CFG_PSIZE_POS;
1526 	}
1527 out:
1528 	spin_unlock_irqrestore(&d40c->lock, flags);
1529 	return 0;
1530 }
1531 EXPORT_SYMBOL(stedma40_set_psize);
1532 
1533 struct dma_async_tx_descriptor *stedma40_memcpy_sg(struct dma_chan *chan,
1534 						   struct scatterlist *sgl_dst,
1535 						   struct scatterlist *sgl_src,
1536 						   unsigned int sgl_len,
1537 						   unsigned long dma_flags)
1538 {
1539 	int res;
1540 	struct d40_desc *d40d;
1541 	struct d40_chan *d40c = container_of(chan, struct d40_chan,
1542 					     chan);
1543 	unsigned long flags;
1544 
1545 	if (d40c->phy_chan == NULL) {
1546 		dev_err(&d40c->chan.dev->device,
1547 			"[%s] Unallocated channel.\n", __func__);
1548 		return ERR_PTR(-EINVAL);
1549 	}
1550 
1551 	spin_lock_irqsave(&d40c->lock, flags);
1552 	d40d = d40_desc_get(d40c);
1553 
1554 	if (d40d == NULL)
1555 		goto err;
1556 
1557 	d40d->lli_len = sgl_len;
1558 	d40d->lli_tx_len = d40d->lli_len;
1559 	d40d->txd.flags = dma_flags;
1560 
1561 	if (d40c->log_num != D40_PHY_CHAN) {
1562 		if (d40d->lli_len > d40c->base->plat_data->llis_per_log)
1563 			d40d->lli_tx_len = d40c->base->plat_data->llis_per_log;
1564 
1565 		if (sgl_len > 1)
1566 			/*
1567 			 * Check if there is space available in lcla. If not,
1568 			 * split list into 1-length and run only in lcpa
1569 			 * space.
1570 			 */
1571 			if (d40_lcla_id_get(d40c) != 0)
1572 				d40d->lli_tx_len = 1;
1573 
1574 		if (d40_pool_lli_alloc(d40d, sgl_len, true) < 0) {
1575 			dev_err(&d40c->chan.dev->device,
1576 				"[%s] Out of memory\n", __func__);
1577 			goto err;
1578 		}
1579 
1580 		(void) d40_log_sg_to_lli(d40c->lcla.src_id,
1581 					 sgl_src,
1582 					 sgl_len,
1583 					 d40d->lli_log.src,
1584 					 d40c->log_def.lcsp1,
1585 					 d40c->dma_cfg.src_info.data_width,
1586 					 d40d->lli_tx_len,
1587 					 d40c->base->plat_data->llis_per_log);
1588 
1589 		(void) d40_log_sg_to_lli(d40c->lcla.dst_id,
1590 					 sgl_dst,
1591 					 sgl_len,
1592 					 d40d->lli_log.dst,
1593 					 d40c->log_def.lcsp3,
1594 					 d40c->dma_cfg.dst_info.data_width,
1595 					 d40d->lli_tx_len,
1596 					 d40c->base->plat_data->llis_per_log);
1597 
1598 
1599 	} else {
1600 		if (d40_pool_lli_alloc(d40d, sgl_len, false) < 0) {
1601 			dev_err(&d40c->chan.dev->device,
1602 				"[%s] Out of memory\n", __func__);
1603 			goto err;
1604 		}
1605 
1606 		res = d40_phy_sg_to_lli(sgl_src,
1607 					sgl_len,
1608 					0,
1609 					d40d->lli_phy.src,
1610 					d40d->lli_phy.src_addr,
1611 					d40c->src_def_cfg,
1612 					d40c->dma_cfg.src_info.data_width,
1613 					d40c->dma_cfg.src_info.psize);
1614 
1615 		if (res < 0)
1616 			goto err;
1617 
1618 		res = d40_phy_sg_to_lli(sgl_dst,
1619 					sgl_len,
1620 					0,
1621 					d40d->lli_phy.dst,
1622 					d40d->lli_phy.dst_addr,
1623 					d40c->dst_def_cfg,
1624 					d40c->dma_cfg.dst_info.data_width,
1625 					d40c->dma_cfg.dst_info.psize);
1626 
1627 		if (res < 0)
1628 			goto err;
1629 
1630 		(void) dma_map_single(d40c->base->dev, d40d->lli_phy.src,
1631 				      d40d->lli_pool.size, DMA_TO_DEVICE);
1632 	}
1633 
1634 	dma_async_tx_descriptor_init(&d40d->txd, chan);
1635 
1636 	d40d->txd.tx_submit = d40_tx_submit;
1637 
1638 	spin_unlock_irqrestore(&d40c->lock, flags);
1639 
1640 	return &d40d->txd;
1641 err:
1642 	spin_unlock_irqrestore(&d40c->lock, flags);
1643 	return NULL;
1644 }
1645 EXPORT_SYMBOL(stedma40_memcpy_sg);
1646 
1647 bool stedma40_filter(struct dma_chan *chan, void *data)
1648 {
1649 	struct stedma40_chan_cfg *info = data;
1650 	struct d40_chan *d40c =
1651 		container_of(chan, struct d40_chan, chan);
1652 	int err;
1653 
1654 	if (data) {
1655 		err = d40_validate_conf(d40c, info);
1656 		if (!err)
1657 			d40c->dma_cfg = *info;
1658 	} else
1659 		err = d40_config_memcpy(d40c);
1660 
1661 	return err == 0;
1662 }
1663 EXPORT_SYMBOL(stedma40_filter);
1664 
1665 /* DMA ENGINE functions */
1666 static int d40_alloc_chan_resources(struct dma_chan *chan)
1667 {
1668 	int err;
1669 	unsigned long flags;
1670 	struct d40_chan *d40c =
1671 		container_of(chan, struct d40_chan, chan);
1672 	bool is_free_phy;
1673 	spin_lock_irqsave(&d40c->lock, flags);
1674 
1675 	d40c->completed = chan->cookie = 1;
1676 
1677 	/*
1678 	 * If no dma configuration is set (channel_type == 0)
1679 	 * use default configuration (memcpy)
1680 	 */
1681 	if (d40c->dma_cfg.channel_type == 0) {
1682 		err = d40_config_memcpy(d40c);
1683 		if (err) {
1684 			dev_err(&d40c->chan.dev->device,
1685 				"[%s] Failed to configure memcpy channel\n",
1686 				__func__);
1687 			goto fail;
1688 		}
1689 	}
1690 	is_free_phy = (d40c->phy_chan == NULL);
1691 
1692 	err = d40_allocate_channel(d40c);
1693 	if (err) {
1694 		dev_err(&d40c->chan.dev->device,
1695 			"[%s] Failed to allocate channel\n", __func__);
1696 		goto fail;
1697 	}
1698 
1699 	/* Fill in basic CFG register values */
1700 	d40_phy_cfg(&d40c->dma_cfg, &d40c->src_def_cfg,
1701 		    &d40c->dst_def_cfg, d40c->log_num != D40_PHY_CHAN);
1702 
1703 	if (d40c->log_num != D40_PHY_CHAN) {
1704 		d40_log_cfg(&d40c->dma_cfg,
1705 			    &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
1706 
1707 		if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM)
1708 			d40c->lcpa = d40c->base->lcpa_base +
1709 			  d40c->dma_cfg.src_dev_type * D40_LCPA_CHAN_SIZE;
1710 		else
1711 			d40c->lcpa = d40c->base->lcpa_base +
1712 			  d40c->dma_cfg.dst_dev_type *
1713 			  D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
1714 	}
1715 
1716 	/*
1717 	 * Only write channel configuration to the DMA if the physical
1718 	 * resource is free. In case of multiple logical channels
1719 	 * on the same physical resource, only the first write is necessary.
1720 	 */
1721 	if (is_free_phy)
1722 		d40_config_write(d40c);
1723 fail:
1724 	spin_unlock_irqrestore(&d40c->lock, flags);
1725 	return err;
1726 }
1727 
1728 static void d40_free_chan_resources(struct dma_chan *chan)
1729 {
1730 	struct d40_chan *d40c =
1731 		container_of(chan, struct d40_chan, chan);
1732 	int err;
1733 	unsigned long flags;
1734 
1735 	if (d40c->phy_chan == NULL) {
1736 		dev_err(&d40c->chan.dev->device,
1737 			"[%s] Cannot free unallocated channel\n", __func__);
1738 		return;
1739 	}
1740 
1741 
1742 	spin_lock_irqsave(&d40c->lock, flags);
1743 
1744 	err = d40_free_dma(d40c);
1745 
1746 	if (err)
1747 		dev_err(&d40c->chan.dev->device,
1748 			"[%s] Failed to free channel\n", __func__);
1749 	spin_unlock_irqrestore(&d40c->lock, flags);
1750 }
1751 
1752 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
1753 						       dma_addr_t dst,
1754 						       dma_addr_t src,
1755 						       size_t size,
1756 						       unsigned long dma_flags)
1757 {
1758 	struct d40_desc *d40d;
1759 	struct d40_chan *d40c = container_of(chan, struct d40_chan,
1760 					     chan);
1761 	unsigned long flags;
1762 	int err = 0;
1763 
1764 	if (d40c->phy_chan == NULL) {
1765 		dev_err(&d40c->chan.dev->device,
1766 			"[%s] Channel is not allocated.\n", __func__);
1767 		return ERR_PTR(-EINVAL);
1768 	}
1769 
1770 	spin_lock_irqsave(&d40c->lock, flags);
1771 	d40d = d40_desc_get(d40c);
1772 
1773 	if (d40d == NULL) {
1774 		dev_err(&d40c->chan.dev->device,
1775 			"[%s] Descriptor is NULL\n", __func__);
1776 		goto err;
1777 	}
1778 
1779 	d40d->txd.flags = dma_flags;
1780 
1781 	dma_async_tx_descriptor_init(&d40d->txd, chan);
1782 
1783 	d40d->txd.tx_submit = d40_tx_submit;
1784 
1785 	if (d40c->log_num != D40_PHY_CHAN) {
1786 
1787 		if (d40_pool_lli_alloc(d40d, 1, true) < 0) {
1788 			dev_err(&d40c->chan.dev->device,
1789 				"[%s] Out of memory\n", __func__);
1790 			goto err;
1791 		}
1792 		d40d->lli_len = 1;
1793 		d40d->lli_tx_len = 1;
1794 
1795 		d40_log_fill_lli(d40d->lli_log.src,
1796 				 src,
1797 				 size,
1798 				 0,
1799 				 d40c->log_def.lcsp1,
1800 				 d40c->dma_cfg.src_info.data_width,
1801 				 false, true);
1802 
1803 		d40_log_fill_lli(d40d->lli_log.dst,
1804 				 dst,
1805 				 size,
1806 				 0,
1807 				 d40c->log_def.lcsp3,
1808 				 d40c->dma_cfg.dst_info.data_width,
1809 				 true, true);
1810 
1811 	} else {
1812 
1813 		if (d40_pool_lli_alloc(d40d, 1, false) < 0) {
1814 			dev_err(&d40c->chan.dev->device,
1815 				"[%s] Out of memory\n", __func__);
1816 			goto err;
1817 		}
1818 
1819 		err = d40_phy_fill_lli(d40d->lli_phy.src,
1820 				       src,
1821 				       size,
1822 				       d40c->dma_cfg.src_info.psize,
1823 				       0,
1824 				       d40c->src_def_cfg,
1825 				       true,
1826 				       d40c->dma_cfg.src_info.data_width,
1827 				       false);
1828 		if (err)
1829 			goto err_fill_lli;
1830 
1831 		err = d40_phy_fill_lli(d40d->lli_phy.dst,
1832 				       dst,
1833 				       size,
1834 				       d40c->dma_cfg.dst_info.psize,
1835 				       0,
1836 				       d40c->dst_def_cfg,
1837 				       true,
1838 				       d40c->dma_cfg.dst_info.data_width,
1839 				       false);
1840 
1841 		if (err)
1842 			goto err_fill_lli;
1843 
1844 		(void) dma_map_single(d40c->base->dev, d40d->lli_phy.src,
1845 				      d40d->lli_pool.size, DMA_TO_DEVICE);
1846 	}
1847 
1848 	spin_unlock_irqrestore(&d40c->lock, flags);
1849 	return &d40d->txd;
1850 
1851 err_fill_lli:
1852 	dev_err(&d40c->chan.dev->device,
1853 		"[%s] Failed filling in PHY LLI\n", __func__);
1854 	d40_pool_lli_free(d40d);
1855 err:
1856 	spin_unlock_irqrestore(&d40c->lock, flags);
1857 	return NULL;
1858 }
1859 
1860 static int d40_prep_slave_sg_log(struct d40_desc *d40d,
1861 				 struct d40_chan *d40c,
1862 				 struct scatterlist *sgl,
1863 				 unsigned int sg_len,
1864 				 enum dma_data_direction direction,
1865 				 unsigned long dma_flags)
1866 {
1867 	dma_addr_t dev_addr = 0;
1868 	int total_size;
1869 
1870 	if (d40_pool_lli_alloc(d40d, sg_len, true) < 0) {
1871 		dev_err(&d40c->chan.dev->device,
1872 			"[%s] Out of memory\n", __func__);
1873 		return -ENOMEM;
1874 	}
1875 
1876 	d40d->lli_len = sg_len;
1877 	if (d40d->lli_len <= d40c->base->plat_data->llis_per_log)
1878 		d40d->lli_tx_len = d40d->lli_len;
1879 	else
1880 		d40d->lli_tx_len = d40c->base->plat_data->llis_per_log;
1881 
1882 	if (sg_len > 1)
1883 		/*
1884 		 * Check if there is space available in lcla.
1885 		 * If not, split list into 1-length and run only
1886 		 * in lcpa space.
1887 		 */
1888 		if (d40_lcla_id_get(d40c) != 0)
1889 			d40d->lli_tx_len = 1;
1890 
1891 	if (direction == DMA_FROM_DEVICE)
1892 		if (d40c->runtime_addr)
1893 			dev_addr = d40c->runtime_addr;
1894 		else
1895 			dev_addr = d40c->base->plat_data->dev_rx[d40c->dma_cfg.src_dev_type];
1896 	else if (direction == DMA_TO_DEVICE)
1897 		if (d40c->runtime_addr)
1898 			dev_addr = d40c->runtime_addr;
1899 		else
1900 			dev_addr = d40c->base->plat_data->dev_tx[d40c->dma_cfg.dst_dev_type];
1901 
1902 	else
1903 		return -EINVAL;
1904 
1905 	total_size = d40_log_sg_to_dev(&d40c->lcla,
1906 				       sgl, sg_len,
1907 				       &d40d->lli_log,
1908 				       &d40c->log_def,
1909 				       d40c->dma_cfg.src_info.data_width,
1910 				       d40c->dma_cfg.dst_info.data_width,
1911 				       direction,
1912 				       dev_addr, d40d->lli_tx_len,
1913 				       d40c->base->plat_data->llis_per_log);
1914 
1915 	if (total_size < 0)
1916 		return -EINVAL;
1917 
1918 	return 0;
1919 }
1920 
1921 static int d40_prep_slave_sg_phy(struct d40_desc *d40d,
1922 				 struct d40_chan *d40c,
1923 				 struct scatterlist *sgl,
1924 				 unsigned int sgl_len,
1925 				 enum dma_data_direction direction,
1926 				 unsigned long dma_flags)
1927 {
1928 	dma_addr_t src_dev_addr;
1929 	dma_addr_t dst_dev_addr;
1930 	int res;
1931 
1932 	if (d40_pool_lli_alloc(d40d, sgl_len, false) < 0) {
1933 		dev_err(&d40c->chan.dev->device,
1934 			"[%s] Out of memory\n", __func__);
1935 		return -ENOMEM;
1936 	}
1937 
1938 	d40d->lli_len = sgl_len;
1939 	d40d->lli_tx_len = sgl_len;
1940 
1941 	if (direction == DMA_FROM_DEVICE) {
1942 		dst_dev_addr = 0;
1943 		if (d40c->runtime_addr)
1944 			src_dev_addr = d40c->runtime_addr;
1945 		else
1946 			src_dev_addr = d40c->base->plat_data->dev_rx[d40c->dma_cfg.src_dev_type];
1947 	} else if (direction == DMA_TO_DEVICE) {
1948 		if (d40c->runtime_addr)
1949 			dst_dev_addr = d40c->runtime_addr;
1950 		else
1951 			dst_dev_addr = d40c->base->plat_data->dev_tx[d40c->dma_cfg.dst_dev_type];
1952 		src_dev_addr = 0;
1953 	} else
1954 		return -EINVAL;
1955 
1956 	res = d40_phy_sg_to_lli(sgl,
1957 				sgl_len,
1958 				src_dev_addr,
1959 				d40d->lli_phy.src,
1960 				d40d->lli_phy.src_addr,
1961 				d40c->src_def_cfg,
1962 				d40c->dma_cfg.src_info.data_width,
1963 				d40c->dma_cfg.src_info.psize);
1964 	if (res < 0)
1965 		return res;
1966 
1967 	res = d40_phy_sg_to_lli(sgl,
1968 				sgl_len,
1969 				dst_dev_addr,
1970 				d40d->lli_phy.dst,
1971 				d40d->lli_phy.dst_addr,
1972 				d40c->dst_def_cfg,
1973 				d40c->dma_cfg.dst_info.data_width,
1974 				d40c->dma_cfg.dst_info.psize);
1975 	if (res < 0)
1976 		return res;
1977 
1978 	(void) dma_map_single(d40c->base->dev, d40d->lli_phy.src,
1979 			      d40d->lli_pool.size, DMA_TO_DEVICE);
1980 	return 0;
1981 }
1982 
1983 static struct dma_async_tx_descriptor *d40_prep_slave_sg(struct dma_chan *chan,
1984 							 struct scatterlist *sgl,
1985 							 unsigned int sg_len,
1986 							 enum dma_data_direction direction,
1987 							 unsigned long dma_flags)
1988 {
1989 	struct d40_desc *d40d;
1990 	struct d40_chan *d40c = container_of(chan, struct d40_chan,
1991 					     chan);
1992 	unsigned long flags;
1993 	int err;
1994 
1995 	if (d40c->phy_chan == NULL) {
1996 		dev_err(&d40c->chan.dev->device,
1997 			"[%s] Cannot prepare unallocated channel\n", __func__);
1998 		return ERR_PTR(-EINVAL);
1999 	}
2000 
2001 	if (d40c->dma_cfg.pre_transfer)
2002 		d40c->dma_cfg.pre_transfer(chan,
2003 					   d40c->dma_cfg.pre_transfer_data,
2004 					   sg_dma_len(sgl));
2005 
2006 	spin_lock_irqsave(&d40c->lock, flags);
2007 	d40d = d40_desc_get(d40c);
2008 	spin_unlock_irqrestore(&d40c->lock, flags);
2009 
2010 	if (d40d == NULL)
2011 		return NULL;
2012 
2013 	if (d40c->log_num != D40_PHY_CHAN)
2014 		err = d40_prep_slave_sg_log(d40d, d40c, sgl, sg_len,
2015 					    direction, dma_flags);
2016 	else
2017 		err = d40_prep_slave_sg_phy(d40d, d40c, sgl, sg_len,
2018 					    direction, dma_flags);
2019 	if (err) {
2020 		dev_err(&d40c->chan.dev->device,
2021 			"[%s] Failed to prepare %s slave sg job: %d\n",
2022 			__func__,
2023 			d40c->log_num != D40_PHY_CHAN ? "log" : "phy", err);
2024 		return NULL;
2025 	}
2026 
2027 	d40d->txd.flags = dma_flags;
2028 
2029 	dma_async_tx_descriptor_init(&d40d->txd, chan);
2030 
2031 	d40d->txd.tx_submit = d40_tx_submit;
2032 
2033 	return &d40d->txd;
2034 }
2035 
2036 static enum dma_status d40_tx_status(struct dma_chan *chan,
2037 				     dma_cookie_t cookie,
2038 				     struct dma_tx_state *txstate)
2039 {
2040 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2041 	dma_cookie_t last_used;
2042 	dma_cookie_t last_complete;
2043 	int ret;
2044 
2045 	if (d40c->phy_chan == NULL) {
2046 		dev_err(&d40c->chan.dev->device,
2047 			"[%s] Cannot read status of unallocated channel\n",
2048 			__func__);
2049 		return -EINVAL;
2050 	}
2051 
2052 	last_complete = d40c->completed;
2053 	last_used = chan->cookie;
2054 
2055 	if (d40_is_paused(d40c))
2056 		ret = DMA_PAUSED;
2057 	else
2058 		ret = dma_async_is_complete(cookie, last_complete, last_used);
2059 
2060 	dma_set_tx_state(txstate, last_complete, last_used,
2061 			 stedma40_residue(chan));
2062 
2063 	return ret;
2064 }
2065 
2066 static void d40_issue_pending(struct dma_chan *chan)
2067 {
2068 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2069 	unsigned long flags;
2070 
2071 	if (d40c->phy_chan == NULL) {
2072 		dev_err(&d40c->chan.dev->device,
2073 			"[%s] Channel is not allocated!\n", __func__);
2074 		return;
2075 	}
2076 
2077 	spin_lock_irqsave(&d40c->lock, flags);
2078 
2079 	/* Busy means that pending jobs are already being processed */
2080 	if (!d40c->busy)
2081 		(void) d40_queue_start(d40c);
2082 
2083 	spin_unlock_irqrestore(&d40c->lock, flags);
2084 }
2085 
2086 /* Runtime reconfiguration extension */
2087 static void d40_set_runtime_config(struct dma_chan *chan,
2088 			       struct dma_slave_config *config)
2089 {
2090 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2091 	struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2092 	enum dma_slave_buswidth config_addr_width;
2093 	dma_addr_t config_addr;
2094 	u32 config_maxburst;
2095 	enum stedma40_periph_data_width addr_width;
2096 	int psize;
2097 
2098 	if (config->direction == DMA_FROM_DEVICE) {
2099 		dma_addr_t dev_addr_rx =
2100 			d40c->base->plat_data->dev_rx[cfg->src_dev_type];
2101 
2102 		config_addr = config->src_addr;
2103 		if (dev_addr_rx)
2104 			dev_dbg(d40c->base->dev,
2105 				"channel has a pre-wired RX address %08x "
2106 				"overriding with %08x\n",
2107 				dev_addr_rx, config_addr);
2108 		if (cfg->dir != STEDMA40_PERIPH_TO_MEM)
2109 			dev_dbg(d40c->base->dev,
2110 				"channel was not configured for peripheral "
2111 				"to memory transfer (%d) overriding\n",
2112 				cfg->dir);
2113 		cfg->dir = STEDMA40_PERIPH_TO_MEM;
2114 
2115 		config_addr_width = config->src_addr_width;
2116 		config_maxburst = config->src_maxburst;
2117 
2118 	} else if (config->direction == DMA_TO_DEVICE) {
2119 		dma_addr_t dev_addr_tx =
2120 			d40c->base->plat_data->dev_tx[cfg->dst_dev_type];
2121 
2122 		config_addr = config->dst_addr;
2123 		if (dev_addr_tx)
2124 			dev_dbg(d40c->base->dev,
2125 				"channel has a pre-wired TX address %08x "
2126 				"overriding with %08x\n",
2127 				dev_addr_tx, config_addr);
2128 		if (cfg->dir != STEDMA40_MEM_TO_PERIPH)
2129 			dev_dbg(d40c->base->dev,
2130 				"channel was not configured for memory "
2131 				"to peripheral transfer (%d) overriding\n",
2132 				cfg->dir);
2133 		cfg->dir = STEDMA40_MEM_TO_PERIPH;
2134 
2135 		config_addr_width = config->dst_addr_width;
2136 		config_maxburst = config->dst_maxburst;
2137 
2138 	} else {
2139 		dev_err(d40c->base->dev,
2140 			"unrecognized channel direction %d\n",
2141 			config->direction);
2142 		return;
2143 	}
2144 
2145 	switch (config_addr_width) {
2146 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
2147 		addr_width = STEDMA40_BYTE_WIDTH;
2148 		break;
2149 	case DMA_SLAVE_BUSWIDTH_2_BYTES:
2150 		addr_width = STEDMA40_HALFWORD_WIDTH;
2151 		break;
2152 	case DMA_SLAVE_BUSWIDTH_4_BYTES:
2153 		addr_width = STEDMA40_WORD_WIDTH;
2154 		break;
2155 	case DMA_SLAVE_BUSWIDTH_8_BYTES:
2156 		addr_width = STEDMA40_DOUBLEWORD_WIDTH;
2157 		break;
2158 	default:
2159 		dev_err(d40c->base->dev,
2160 			"illegal peripheral address width "
2161 			"requested (%d)\n",
2162 			config->src_addr_width);
2163 		return;
2164 	}
2165 
2166 	if (config_maxburst >= 16)
2167 		psize = STEDMA40_PSIZE_LOG_16;
2168 	else if (config_maxburst >= 8)
2169 		psize = STEDMA40_PSIZE_LOG_8;
2170 	else if (config_maxburst >= 4)
2171 		psize = STEDMA40_PSIZE_LOG_4;
2172 	else
2173 		psize = STEDMA40_PSIZE_LOG_1;
2174 
2175 	/* Set up all the endpoint configs */
2176 	cfg->src_info.data_width = addr_width;
2177 	cfg->src_info.psize = psize;
2178 	cfg->src_info.endianess = STEDMA40_LITTLE_ENDIAN;
2179 	cfg->src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2180 	cfg->dst_info.data_width = addr_width;
2181 	cfg->dst_info.psize = psize;
2182 	cfg->dst_info.endianess = STEDMA40_LITTLE_ENDIAN;
2183 	cfg->dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2184 
2185 	/* These settings will take precedence later */
2186 	d40c->runtime_addr = config_addr;
2187 	d40c->runtime_direction = config->direction;
2188 	dev_dbg(d40c->base->dev,
2189 		"configured channel %s for %s, data width %d, "
2190 		"maxburst %d bytes, LE, no flow control\n",
2191 		dma_chan_name(chan),
2192 		(config->direction == DMA_FROM_DEVICE) ? "RX" : "TX",
2193 		config_addr_width,
2194 		config_maxburst);
2195 }
2196 
2197 static int d40_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
2198 		       unsigned long arg)
2199 {
2200 	unsigned long flags;
2201 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2202 
2203 	if (d40c->phy_chan == NULL) {
2204 		dev_err(&d40c->chan.dev->device,
2205 			"[%s] Channel is not allocated!\n", __func__);
2206 		return -EINVAL;
2207 	}
2208 
2209 	switch (cmd) {
2210 	case DMA_TERMINATE_ALL:
2211 		spin_lock_irqsave(&d40c->lock, flags);
2212 		d40_term_all(d40c);
2213 		spin_unlock_irqrestore(&d40c->lock, flags);
2214 		return 0;
2215 	case DMA_PAUSE:
2216 		return d40_pause(chan);
2217 	case DMA_RESUME:
2218 		return d40_resume(chan);
2219 	case DMA_SLAVE_CONFIG:
2220 		d40_set_runtime_config(chan,
2221 			(struct dma_slave_config *) arg);
2222 		return 0;
2223 	default:
2224 		break;
2225 	}
2226 
2227 	/* Other commands are unimplemented */
2228 	return -ENXIO;
2229 }
2230 
2231 /* Initialization functions */
2232 
2233 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2234 				 struct d40_chan *chans, int offset,
2235 				 int num_chans)
2236 {
2237 	int i = 0;
2238 	struct d40_chan *d40c;
2239 
2240 	INIT_LIST_HEAD(&dma->channels);
2241 
2242 	for (i = offset; i < offset + num_chans; i++) {
2243 		d40c = &chans[i];
2244 		d40c->base = base;
2245 		d40c->chan.device = dma;
2246 
2247 		/* Invalidate lcla element */
2248 		d40c->lcla.src_id = -1;
2249 		d40c->lcla.dst_id = -1;
2250 
2251 		spin_lock_init(&d40c->lock);
2252 
2253 		d40c->log_num = D40_PHY_CHAN;
2254 
2255 		INIT_LIST_HEAD(&d40c->active);
2256 		INIT_LIST_HEAD(&d40c->queue);
2257 		INIT_LIST_HEAD(&d40c->client);
2258 
2259 		tasklet_init(&d40c->tasklet, dma_tasklet,
2260 			     (unsigned long) d40c);
2261 
2262 		list_add_tail(&d40c->chan.device_node,
2263 			      &dma->channels);
2264 	}
2265 }
2266 
2267 static int __init d40_dmaengine_init(struct d40_base *base,
2268 				     int num_reserved_chans)
2269 {
2270 	int err ;
2271 
2272 	d40_chan_init(base, &base->dma_slave, base->log_chans,
2273 		      0, base->num_log_chans);
2274 
2275 	dma_cap_zero(base->dma_slave.cap_mask);
2276 	dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2277 
2278 	base->dma_slave.device_alloc_chan_resources = d40_alloc_chan_resources;
2279 	base->dma_slave.device_free_chan_resources = d40_free_chan_resources;
2280 	base->dma_slave.device_prep_dma_memcpy = d40_prep_memcpy;
2281 	base->dma_slave.device_prep_slave_sg = d40_prep_slave_sg;
2282 	base->dma_slave.device_tx_status = d40_tx_status;
2283 	base->dma_slave.device_issue_pending = d40_issue_pending;
2284 	base->dma_slave.device_control = d40_control;
2285 	base->dma_slave.dev = base->dev;
2286 
2287 	err = dma_async_device_register(&base->dma_slave);
2288 
2289 	if (err) {
2290 		dev_err(base->dev,
2291 			"[%s] Failed to register slave channels\n",
2292 			__func__);
2293 		goto failure1;
2294 	}
2295 
2296 	d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2297 		      base->num_log_chans, base->plat_data->memcpy_len);
2298 
2299 	dma_cap_zero(base->dma_memcpy.cap_mask);
2300 	dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2301 
2302 	base->dma_memcpy.device_alloc_chan_resources = d40_alloc_chan_resources;
2303 	base->dma_memcpy.device_free_chan_resources = d40_free_chan_resources;
2304 	base->dma_memcpy.device_prep_dma_memcpy = d40_prep_memcpy;
2305 	base->dma_memcpy.device_prep_slave_sg = d40_prep_slave_sg;
2306 	base->dma_memcpy.device_tx_status = d40_tx_status;
2307 	base->dma_memcpy.device_issue_pending = d40_issue_pending;
2308 	base->dma_memcpy.device_control = d40_control;
2309 	base->dma_memcpy.dev = base->dev;
2310 	/*
2311 	 * This controller can only access address at even
2312 	 * 32bit boundaries, i.e. 2^2
2313 	 */
2314 	base->dma_memcpy.copy_align = 2;
2315 
2316 	err = dma_async_device_register(&base->dma_memcpy);
2317 
2318 	if (err) {
2319 		dev_err(base->dev,
2320 			"[%s] Failed to regsiter memcpy only channels\n",
2321 			__func__);
2322 		goto failure2;
2323 	}
2324 
2325 	d40_chan_init(base, &base->dma_both, base->phy_chans,
2326 		      0, num_reserved_chans);
2327 
2328 	dma_cap_zero(base->dma_both.cap_mask);
2329 	dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2330 	dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2331 
2332 	base->dma_both.device_alloc_chan_resources = d40_alloc_chan_resources;
2333 	base->dma_both.device_free_chan_resources = d40_free_chan_resources;
2334 	base->dma_both.device_prep_dma_memcpy = d40_prep_memcpy;
2335 	base->dma_both.device_prep_slave_sg = d40_prep_slave_sg;
2336 	base->dma_both.device_tx_status = d40_tx_status;
2337 	base->dma_both.device_issue_pending = d40_issue_pending;
2338 	base->dma_both.device_control = d40_control;
2339 	base->dma_both.dev = base->dev;
2340 	base->dma_both.copy_align = 2;
2341 	err = dma_async_device_register(&base->dma_both);
2342 
2343 	if (err) {
2344 		dev_err(base->dev,
2345 			"[%s] Failed to register logical and physical capable channels\n",
2346 			__func__);
2347 		goto failure3;
2348 	}
2349 	return 0;
2350 failure3:
2351 	dma_async_device_unregister(&base->dma_memcpy);
2352 failure2:
2353 	dma_async_device_unregister(&base->dma_slave);
2354 failure1:
2355 	return err;
2356 }
2357 
2358 /* Initialization functions. */
2359 
2360 static int __init d40_phy_res_init(struct d40_base *base)
2361 {
2362 	int i;
2363 	int num_phy_chans_avail = 0;
2364 	u32 val[2];
2365 	int odd_even_bit = -2;
2366 
2367 	val[0] = readl(base->virtbase + D40_DREG_PRSME);
2368 	val[1] = readl(base->virtbase + D40_DREG_PRSMO);
2369 
2370 	for (i = 0; i < base->num_phy_chans; i++) {
2371 		base->phy_res[i].num = i;
2372 		odd_even_bit += 2 * ((i % 2) == 0);
2373 		if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
2374 			/* Mark security only channels as occupied */
2375 			base->phy_res[i].allocated_src = D40_ALLOC_PHY;
2376 			base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
2377 		} else {
2378 			base->phy_res[i].allocated_src = D40_ALLOC_FREE;
2379 			base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
2380 			num_phy_chans_avail++;
2381 		}
2382 		spin_lock_init(&base->phy_res[i].lock);
2383 	}
2384 
2385 	/* Mark disabled channels as occupied */
2386 	for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
2387 			base->phy_res[i].allocated_src = D40_ALLOC_PHY;
2388 			base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
2389 			num_phy_chans_avail--;
2390 	}
2391 
2392 	dev_info(base->dev, "%d of %d physical DMA channels available\n",
2393 		 num_phy_chans_avail, base->num_phy_chans);
2394 
2395 	/* Verify settings extended vs standard */
2396 	val[0] = readl(base->virtbase + D40_DREG_PRTYP);
2397 
2398 	for (i = 0; i < base->num_phy_chans; i++) {
2399 
2400 		if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
2401 		    (val[0] & 0x3) != 1)
2402 			dev_info(base->dev,
2403 				 "[%s] INFO: channel %d is misconfigured (%d)\n",
2404 				 __func__, i, val[0] & 0x3);
2405 
2406 		val[0] = val[0] >> 2;
2407 	}
2408 
2409 	return num_phy_chans_avail;
2410 }
2411 
2412 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
2413 {
2414 	static const struct d40_reg_val dma_id_regs[] = {
2415 		/* Peripheral Id */
2416 		{ .reg = D40_DREG_PERIPHID0, .val = 0x0040},
2417 		{ .reg = D40_DREG_PERIPHID1, .val = 0x0000},
2418 		/*
2419 		 * D40_DREG_PERIPHID2 Depends on HW revision:
2420 		 *  MOP500/HREF ED has 0x0008,
2421 		 *  ? has 0x0018,
2422 		 *  HREF V1 has 0x0028
2423 		 */
2424 		{ .reg = D40_DREG_PERIPHID3, .val = 0x0000},
2425 
2426 		/* PCell Id */
2427 		{ .reg = D40_DREG_CELLID0, .val = 0x000d},
2428 		{ .reg = D40_DREG_CELLID1, .val = 0x00f0},
2429 		{ .reg = D40_DREG_CELLID2, .val = 0x0005},
2430 		{ .reg = D40_DREG_CELLID3, .val = 0x00b1}
2431 	};
2432 	struct stedma40_platform_data *plat_data;
2433 	struct clk *clk = NULL;
2434 	void __iomem *virtbase = NULL;
2435 	struct resource *res = NULL;
2436 	struct d40_base *base = NULL;
2437 	int num_log_chans = 0;
2438 	int num_phy_chans;
2439 	int i;
2440 	u32 val;
2441 
2442 	clk = clk_get(&pdev->dev, NULL);
2443 
2444 	if (IS_ERR(clk)) {
2445 		dev_err(&pdev->dev, "[%s] No matching clock found\n",
2446 			__func__);
2447 		goto failure;
2448 	}
2449 
2450 	clk_enable(clk);
2451 
2452 	/* Get IO for DMAC base address */
2453 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
2454 	if (!res)
2455 		goto failure;
2456 
2457 	if (request_mem_region(res->start, resource_size(res),
2458 			       D40_NAME " I/O base") == NULL)
2459 		goto failure;
2460 
2461 	virtbase = ioremap(res->start, resource_size(res));
2462 	if (!virtbase)
2463 		goto failure;
2464 
2465 	/* HW version check */
2466 	for (i = 0; i < ARRAY_SIZE(dma_id_regs); i++) {
2467 		if (dma_id_regs[i].val !=
2468 		    readl(virtbase + dma_id_regs[i].reg)) {
2469 			dev_err(&pdev->dev,
2470 				"[%s] Unknown hardware! Expected 0x%x at 0x%x but got 0x%x\n",
2471 				__func__,
2472 				dma_id_regs[i].val,
2473 				dma_id_regs[i].reg,
2474 				readl(virtbase + dma_id_regs[i].reg));
2475 			goto failure;
2476 		}
2477 	}
2478 
2479 	/* Get silicon revision */
2480 	val = readl(virtbase + D40_DREG_PERIPHID2);
2481 
2482 	if ((val & 0xf) != D40_PERIPHID2_DESIGNER) {
2483 		dev_err(&pdev->dev,
2484 			"[%s] Unknown designer! Got %x wanted %x\n",
2485 			__func__, val & 0xf, D40_PERIPHID2_DESIGNER);
2486 		goto failure;
2487 	}
2488 
2489 	/* The number of physical channels on this HW */
2490 	num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
2491 
2492 	dev_info(&pdev->dev, "hardware revision: %d @ 0x%x\n",
2493 		 (val >> 4) & 0xf, res->start);
2494 
2495 	plat_data = pdev->dev.platform_data;
2496 
2497 	/* Count the number of logical channels in use */
2498 	for (i = 0; i < plat_data->dev_len; i++)
2499 		if (plat_data->dev_rx[i] != 0)
2500 			num_log_chans++;
2501 
2502 	for (i = 0; i < plat_data->dev_len; i++)
2503 		if (plat_data->dev_tx[i] != 0)
2504 			num_log_chans++;
2505 
2506 	base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
2507 		       (num_phy_chans + num_log_chans + plat_data->memcpy_len) *
2508 		       sizeof(struct d40_chan), GFP_KERNEL);
2509 
2510 	if (base == NULL) {
2511 		dev_err(&pdev->dev, "[%s] Out of memory\n", __func__);
2512 		goto failure;
2513 	}
2514 
2515 	base->rev = (val >> 4) & 0xf;
2516 	base->clk = clk;
2517 	base->num_phy_chans = num_phy_chans;
2518 	base->num_log_chans = num_log_chans;
2519 	base->phy_start = res->start;
2520 	base->phy_size = resource_size(res);
2521 	base->virtbase = virtbase;
2522 	base->plat_data = plat_data;
2523 	base->dev = &pdev->dev;
2524 	base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
2525 	base->log_chans = &base->phy_chans[num_phy_chans];
2526 
2527 	base->phy_res = kzalloc(num_phy_chans * sizeof(struct d40_phy_res),
2528 				GFP_KERNEL);
2529 	if (!base->phy_res)
2530 		goto failure;
2531 
2532 	base->lookup_phy_chans = kzalloc(num_phy_chans *
2533 					 sizeof(struct d40_chan *),
2534 					 GFP_KERNEL);
2535 	if (!base->lookup_phy_chans)
2536 		goto failure;
2537 
2538 	if (num_log_chans + plat_data->memcpy_len) {
2539 		/*
2540 		 * The max number of logical channels are event lines for all
2541 		 * src devices and dst devices
2542 		 */
2543 		base->lookup_log_chans = kzalloc(plat_data->dev_len * 2 *
2544 						 sizeof(struct d40_chan *),
2545 						 GFP_KERNEL);
2546 		if (!base->lookup_log_chans)
2547 			goto failure;
2548 	}
2549 	base->lcla_pool.alloc_map = kzalloc(num_phy_chans * sizeof(u32),
2550 					    GFP_KERNEL);
2551 	if (!base->lcla_pool.alloc_map)
2552 		goto failure;
2553 
2554 	base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
2555 					    0, SLAB_HWCACHE_ALIGN,
2556 					    NULL);
2557 	if (base->desc_slab == NULL)
2558 		goto failure;
2559 
2560 	return base;
2561 
2562 failure:
2563 	if (clk) {
2564 		clk_disable(clk);
2565 		clk_put(clk);
2566 	}
2567 	if (virtbase)
2568 		iounmap(virtbase);
2569 	if (res)
2570 		release_mem_region(res->start,
2571 				   resource_size(res));
2572 	if (virtbase)
2573 		iounmap(virtbase);
2574 
2575 	if (base) {
2576 		kfree(base->lcla_pool.alloc_map);
2577 		kfree(base->lookup_log_chans);
2578 		kfree(base->lookup_phy_chans);
2579 		kfree(base->phy_res);
2580 		kfree(base);
2581 	}
2582 
2583 	return NULL;
2584 }
2585 
2586 static void __init d40_hw_init(struct d40_base *base)
2587 {
2588 
2589 	static const struct d40_reg_val dma_init_reg[] = {
2590 		/* Clock every part of the DMA block from start */
2591 		{ .reg = D40_DREG_GCC,    .val = 0x0000ff01},
2592 
2593 		/* Interrupts on all logical channels */
2594 		{ .reg = D40_DREG_LCMIS0, .val = 0xFFFFFFFF},
2595 		{ .reg = D40_DREG_LCMIS1, .val = 0xFFFFFFFF},
2596 		{ .reg = D40_DREG_LCMIS2, .val = 0xFFFFFFFF},
2597 		{ .reg = D40_DREG_LCMIS3, .val = 0xFFFFFFFF},
2598 		{ .reg = D40_DREG_LCICR0, .val = 0xFFFFFFFF},
2599 		{ .reg = D40_DREG_LCICR1, .val = 0xFFFFFFFF},
2600 		{ .reg = D40_DREG_LCICR2, .val = 0xFFFFFFFF},
2601 		{ .reg = D40_DREG_LCICR3, .val = 0xFFFFFFFF},
2602 		{ .reg = D40_DREG_LCTIS0, .val = 0xFFFFFFFF},
2603 		{ .reg = D40_DREG_LCTIS1, .val = 0xFFFFFFFF},
2604 		{ .reg = D40_DREG_LCTIS2, .val = 0xFFFFFFFF},
2605 		{ .reg = D40_DREG_LCTIS3, .val = 0xFFFFFFFF}
2606 	};
2607 	int i;
2608 	u32 prmseo[2] = {0, 0};
2609 	u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
2610 	u32 pcmis = 0;
2611 	u32 pcicr = 0;
2612 
2613 	for (i = 0; i < ARRAY_SIZE(dma_init_reg); i++)
2614 		writel(dma_init_reg[i].val,
2615 		       base->virtbase + dma_init_reg[i].reg);
2616 
2617 	/* Configure all our dma channels to default settings */
2618 	for (i = 0; i < base->num_phy_chans; i++) {
2619 
2620 		activeo[i % 2] = activeo[i % 2] << 2;
2621 
2622 		if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
2623 		    == D40_ALLOC_PHY) {
2624 			activeo[i % 2] |= 3;
2625 			continue;
2626 		}
2627 
2628 		/* Enable interrupt # */
2629 		pcmis = (pcmis << 1) | 1;
2630 
2631 		/* Clear interrupt # */
2632 		pcicr = (pcicr << 1) | 1;
2633 
2634 		/* Set channel to physical mode */
2635 		prmseo[i % 2] = prmseo[i % 2] << 2;
2636 		prmseo[i % 2] |= 1;
2637 
2638 	}
2639 
2640 	writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
2641 	writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
2642 	writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
2643 	writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
2644 
2645 	/* Write which interrupt to enable */
2646 	writel(pcmis, base->virtbase + D40_DREG_PCMIS);
2647 
2648 	/* Write which interrupt to clear */
2649 	writel(pcicr, base->virtbase + D40_DREG_PCICR);
2650 
2651 }
2652 
2653 static int __init d40_lcla_allocate(struct d40_base *base)
2654 {
2655 	unsigned long *page_list;
2656 	int i, j;
2657 	int ret = 0;
2658 
2659 	/*
2660 	 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
2661 	 * To full fill this hardware requirement without wasting 256 kb
2662 	 * we allocate pages until we get an aligned one.
2663 	 */
2664 	page_list = kmalloc(sizeof(unsigned long) * MAX_LCLA_ALLOC_ATTEMPTS,
2665 			    GFP_KERNEL);
2666 
2667 	if (!page_list) {
2668 		ret = -ENOMEM;
2669 		goto failure;
2670 	}
2671 
2672 	/* Calculating how many pages that are required */
2673 	base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
2674 
2675 	for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
2676 		page_list[i] = __get_free_pages(GFP_KERNEL,
2677 						base->lcla_pool.pages);
2678 		if (!page_list[i]) {
2679 
2680 			dev_err(base->dev,
2681 				"[%s] Failed to allocate %d pages.\n",
2682 				__func__, base->lcla_pool.pages);
2683 
2684 			for (j = 0; j < i; j++)
2685 				free_pages(page_list[j], base->lcla_pool.pages);
2686 			goto failure;
2687 		}
2688 
2689 		if ((virt_to_phys((void *)page_list[i]) &
2690 		     (LCLA_ALIGNMENT - 1)) == 0)
2691 			break;
2692 	}
2693 
2694 	for (j = 0; j < i; j++)
2695 		free_pages(page_list[j], base->lcla_pool.pages);
2696 
2697 	if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
2698 		base->lcla_pool.base = (void *)page_list[i];
2699 	} else {
2700 		/* After many attempts, no succees with finding the correct
2701 		 * alignment try with allocating a big buffer */
2702 		dev_warn(base->dev,
2703 			 "[%s] Failed to get %d pages @ 18 bit align.\n",
2704 			 __func__, base->lcla_pool.pages);
2705 		base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
2706 							 base->num_phy_chans +
2707 							 LCLA_ALIGNMENT,
2708 							 GFP_KERNEL);
2709 		if (!base->lcla_pool.base_unaligned) {
2710 			ret = -ENOMEM;
2711 			goto failure;
2712 		}
2713 
2714 		base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
2715 						 LCLA_ALIGNMENT);
2716 	}
2717 
2718 	writel(virt_to_phys(base->lcla_pool.base),
2719 	       base->virtbase + D40_DREG_LCLA);
2720 failure:
2721 	kfree(page_list);
2722 	return ret;
2723 }
2724 
2725 static int __init d40_probe(struct platform_device *pdev)
2726 {
2727 	int err;
2728 	int ret = -ENOENT;
2729 	struct d40_base *base;
2730 	struct resource *res = NULL;
2731 	int num_reserved_chans;
2732 	u32 val;
2733 
2734 	base = d40_hw_detect_init(pdev);
2735 
2736 	if (!base)
2737 		goto failure;
2738 
2739 	num_reserved_chans = d40_phy_res_init(base);
2740 
2741 	platform_set_drvdata(pdev, base);
2742 
2743 	spin_lock_init(&base->interrupt_lock);
2744 	spin_lock_init(&base->execmd_lock);
2745 
2746 	/* Get IO for logical channel parameter address */
2747 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
2748 	if (!res) {
2749 		ret = -ENOENT;
2750 		dev_err(&pdev->dev,
2751 			"[%s] No \"lcpa\" memory resource\n",
2752 			__func__);
2753 		goto failure;
2754 	}
2755 	base->lcpa_size = resource_size(res);
2756 	base->phy_lcpa = res->start;
2757 
2758 	if (request_mem_region(res->start, resource_size(res),
2759 			       D40_NAME " I/O lcpa") == NULL) {
2760 		ret = -EBUSY;
2761 		dev_err(&pdev->dev,
2762 			"[%s] Failed to request LCPA region 0x%x-0x%x\n",
2763 			__func__, res->start, res->end);
2764 		goto failure;
2765 	}
2766 
2767 	/* We make use of ESRAM memory for this. */
2768 	val = readl(base->virtbase + D40_DREG_LCPA);
2769 	if (res->start != val && val != 0) {
2770 		dev_warn(&pdev->dev,
2771 			 "[%s] Mismatch LCPA dma 0x%x, def 0x%x\n",
2772 			 __func__, val, res->start);
2773 	} else
2774 		writel(res->start, base->virtbase + D40_DREG_LCPA);
2775 
2776 	base->lcpa_base = ioremap(res->start, resource_size(res));
2777 	if (!base->lcpa_base) {
2778 		ret = -ENOMEM;
2779 		dev_err(&pdev->dev,
2780 			"[%s] Failed to ioremap LCPA region\n",
2781 			__func__);
2782 		goto failure;
2783 	}
2784 
2785 	ret = d40_lcla_allocate(base);
2786 	if (ret) {
2787 		dev_err(&pdev->dev, "[%s] Failed to allocate LCLA area\n",
2788 			__func__);
2789 		goto failure;
2790 	}
2791 
2792 	spin_lock_init(&base->lcla_pool.lock);
2793 
2794 	base->lcla_pool.num_blocks = base->num_phy_chans;
2795 
2796 	base->irq = platform_get_irq(pdev, 0);
2797 
2798 	ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
2799 
2800 	if (ret) {
2801 		dev_err(&pdev->dev, "[%s] No IRQ defined\n", __func__);
2802 		goto failure;
2803 	}
2804 
2805 	err = d40_dmaengine_init(base, num_reserved_chans);
2806 	if (err)
2807 		goto failure;
2808 
2809 	d40_hw_init(base);
2810 
2811 	dev_info(base->dev, "initialized\n");
2812 	return 0;
2813 
2814 failure:
2815 	if (base) {
2816 		if (base->desc_slab)
2817 			kmem_cache_destroy(base->desc_slab);
2818 		if (base->virtbase)
2819 			iounmap(base->virtbase);
2820 		if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
2821 			free_pages((unsigned long)base->lcla_pool.base,
2822 				   base->lcla_pool.pages);
2823 		if (base->lcla_pool.base_unaligned)
2824 			kfree(base->lcla_pool.base_unaligned);
2825 		if (base->phy_lcpa)
2826 			release_mem_region(base->phy_lcpa,
2827 					   base->lcpa_size);
2828 		if (base->phy_start)
2829 			release_mem_region(base->phy_start,
2830 					   base->phy_size);
2831 		if (base->clk) {
2832 			clk_disable(base->clk);
2833 			clk_put(base->clk);
2834 		}
2835 
2836 		kfree(base->lcla_pool.alloc_map);
2837 		kfree(base->lookup_log_chans);
2838 		kfree(base->lookup_phy_chans);
2839 		kfree(base->phy_res);
2840 		kfree(base);
2841 	}
2842 
2843 	dev_err(&pdev->dev, "[%s] probe failed\n", __func__);
2844 	return ret;
2845 }
2846 
2847 static struct platform_driver d40_driver = {
2848 	.driver = {
2849 		.owner = THIS_MODULE,
2850 		.name  = D40_NAME,
2851 	},
2852 };
2853 
2854 int __init stedma40_init(void)
2855 {
2856 	return platform_driver_probe(&d40_driver, d40_probe);
2857 }
2858 arch_initcall(stedma40_init);
2859