xref: /openbmc/linux/drivers/dma/ste_dma40.c (revision 089a49b6)
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/of.h>
21 #include <linux/of_dma.h>
22 #include <linux/amba/bus.h>
23 #include <linux/regulator/consumer.h>
24 #include <linux/platform_data/dma-ste-dma40.h>
25 
26 #include "dmaengine.h"
27 #include "ste_dma40_ll.h"
28 
29 #define D40_NAME "dma40"
30 
31 #define D40_PHY_CHAN -1
32 
33 /* For masking out/in 2 bit channel positions */
34 #define D40_CHAN_POS(chan)  (2 * (chan / 2))
35 #define D40_CHAN_POS_MASK(chan) (0x3 << D40_CHAN_POS(chan))
36 
37 /* Maximum iterations taken before giving up suspending a channel */
38 #define D40_SUSPEND_MAX_IT 500
39 
40 /* Milliseconds */
41 #define DMA40_AUTOSUSPEND_DELAY	100
42 
43 /* Hardware requirement on LCLA alignment */
44 #define LCLA_ALIGNMENT 0x40000
45 
46 /* Max number of links per event group */
47 #define D40_LCLA_LINK_PER_EVENT_GRP 128
48 #define D40_LCLA_END D40_LCLA_LINK_PER_EVENT_GRP
49 
50 /* Max number of logical channels per physical channel */
51 #define D40_MAX_LOG_CHAN_PER_PHY 32
52 
53 /* Attempts before giving up to trying to get pages that are aligned */
54 #define MAX_LCLA_ALLOC_ATTEMPTS 256
55 
56 /* Bit markings for allocation map */
57 #define D40_ALLOC_FREE		BIT(31)
58 #define D40_ALLOC_PHY		BIT(30)
59 #define D40_ALLOC_LOG_FREE	0
60 
61 #define D40_MEMCPY_MAX_CHANS	8
62 
63 /* Reserved event lines for memcpy only. */
64 #define DB8500_DMA_MEMCPY_EV_0	51
65 #define DB8500_DMA_MEMCPY_EV_1	56
66 #define DB8500_DMA_MEMCPY_EV_2	57
67 #define DB8500_DMA_MEMCPY_EV_3	58
68 #define DB8500_DMA_MEMCPY_EV_4	59
69 #define DB8500_DMA_MEMCPY_EV_5	60
70 
71 static int dma40_memcpy_channels[] = {
72 	DB8500_DMA_MEMCPY_EV_0,
73 	DB8500_DMA_MEMCPY_EV_1,
74 	DB8500_DMA_MEMCPY_EV_2,
75 	DB8500_DMA_MEMCPY_EV_3,
76 	DB8500_DMA_MEMCPY_EV_4,
77 	DB8500_DMA_MEMCPY_EV_5,
78 };
79 
80 /* Default configuration for physcial memcpy */
81 static struct stedma40_chan_cfg dma40_memcpy_conf_phy = {
82 	.mode = STEDMA40_MODE_PHYSICAL,
83 	.dir = DMA_MEM_TO_MEM,
84 
85 	.src_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
86 	.src_info.psize = STEDMA40_PSIZE_PHY_1,
87 	.src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
88 
89 	.dst_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
90 	.dst_info.psize = STEDMA40_PSIZE_PHY_1,
91 	.dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
92 };
93 
94 /* Default configuration for logical memcpy */
95 static struct stedma40_chan_cfg dma40_memcpy_conf_log = {
96 	.mode = STEDMA40_MODE_LOGICAL,
97 	.dir = DMA_MEM_TO_MEM,
98 
99 	.src_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
100 	.src_info.psize = STEDMA40_PSIZE_LOG_1,
101 	.src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
102 
103 	.dst_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
104 	.dst_info.psize = STEDMA40_PSIZE_LOG_1,
105 	.dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
106 };
107 
108 /**
109  * enum 40_command - The different commands and/or statuses.
110  *
111  * @D40_DMA_STOP: DMA channel command STOP or status STOPPED,
112  * @D40_DMA_RUN: The DMA channel is RUNNING of the command RUN.
113  * @D40_DMA_SUSPEND_REQ: Request the DMA to SUSPEND as soon as possible.
114  * @D40_DMA_SUSPENDED: The DMA channel is SUSPENDED.
115  */
116 enum d40_command {
117 	D40_DMA_STOP		= 0,
118 	D40_DMA_RUN		= 1,
119 	D40_DMA_SUSPEND_REQ	= 2,
120 	D40_DMA_SUSPENDED	= 3
121 };
122 
123 /*
124  * enum d40_events - The different Event Enables for the event lines.
125  *
126  * @D40_DEACTIVATE_EVENTLINE: De-activate Event line, stopping the logical chan.
127  * @D40_ACTIVATE_EVENTLINE: Activate the Event line, to start a logical chan.
128  * @D40_SUSPEND_REQ_EVENTLINE: Requesting for suspending a event line.
129  * @D40_ROUND_EVENTLINE: Status check for event line.
130  */
131 
132 enum d40_events {
133 	D40_DEACTIVATE_EVENTLINE	= 0,
134 	D40_ACTIVATE_EVENTLINE		= 1,
135 	D40_SUSPEND_REQ_EVENTLINE	= 2,
136 	D40_ROUND_EVENTLINE		= 3
137 };
138 
139 /*
140  * These are the registers that has to be saved and later restored
141  * when the DMA hw is powered off.
142  * TODO: Add save/restore of D40_DREG_GCC on dma40 v3 or later, if that works.
143  */
144 static u32 d40_backup_regs[] = {
145 	D40_DREG_LCPA,
146 	D40_DREG_LCLA,
147 	D40_DREG_PRMSE,
148 	D40_DREG_PRMSO,
149 	D40_DREG_PRMOE,
150 	D40_DREG_PRMOO,
151 };
152 
153 #define BACKUP_REGS_SZ ARRAY_SIZE(d40_backup_regs)
154 
155 /*
156  * since 9540 and 8540 has the same HW revision
157  * use v4a for 9540 or ealier
158  * use v4b for 8540 or later
159  * HW revision:
160  * DB8500ed has revision 0
161  * DB8500v1 has revision 2
162  * DB8500v2 has revision 3
163  * AP9540v1 has revision 4
164  * DB8540v1 has revision 4
165  * TODO: Check if all these registers have to be saved/restored on dma40 v4a
166  */
167 static u32 d40_backup_regs_v4a[] = {
168 	D40_DREG_PSEG1,
169 	D40_DREG_PSEG2,
170 	D40_DREG_PSEG3,
171 	D40_DREG_PSEG4,
172 	D40_DREG_PCEG1,
173 	D40_DREG_PCEG2,
174 	D40_DREG_PCEG3,
175 	D40_DREG_PCEG4,
176 	D40_DREG_RSEG1,
177 	D40_DREG_RSEG2,
178 	D40_DREG_RSEG3,
179 	D40_DREG_RSEG4,
180 	D40_DREG_RCEG1,
181 	D40_DREG_RCEG2,
182 	D40_DREG_RCEG3,
183 	D40_DREG_RCEG4,
184 };
185 
186 #define BACKUP_REGS_SZ_V4A ARRAY_SIZE(d40_backup_regs_v4a)
187 
188 static u32 d40_backup_regs_v4b[] = {
189 	D40_DREG_CPSEG1,
190 	D40_DREG_CPSEG2,
191 	D40_DREG_CPSEG3,
192 	D40_DREG_CPSEG4,
193 	D40_DREG_CPSEG5,
194 	D40_DREG_CPCEG1,
195 	D40_DREG_CPCEG2,
196 	D40_DREG_CPCEG3,
197 	D40_DREG_CPCEG4,
198 	D40_DREG_CPCEG5,
199 	D40_DREG_CRSEG1,
200 	D40_DREG_CRSEG2,
201 	D40_DREG_CRSEG3,
202 	D40_DREG_CRSEG4,
203 	D40_DREG_CRSEG5,
204 	D40_DREG_CRCEG1,
205 	D40_DREG_CRCEG2,
206 	D40_DREG_CRCEG3,
207 	D40_DREG_CRCEG4,
208 	D40_DREG_CRCEG5,
209 };
210 
211 #define BACKUP_REGS_SZ_V4B ARRAY_SIZE(d40_backup_regs_v4b)
212 
213 static u32 d40_backup_regs_chan[] = {
214 	D40_CHAN_REG_SSCFG,
215 	D40_CHAN_REG_SSELT,
216 	D40_CHAN_REG_SSPTR,
217 	D40_CHAN_REG_SSLNK,
218 	D40_CHAN_REG_SDCFG,
219 	D40_CHAN_REG_SDELT,
220 	D40_CHAN_REG_SDPTR,
221 	D40_CHAN_REG_SDLNK,
222 };
223 
224 #define BACKUP_REGS_SZ_MAX ((BACKUP_REGS_SZ_V4A > BACKUP_REGS_SZ_V4B) ? \
225 			     BACKUP_REGS_SZ_V4A : BACKUP_REGS_SZ_V4B)
226 
227 /**
228  * struct d40_interrupt_lookup - lookup table for interrupt handler
229  *
230  * @src: Interrupt mask register.
231  * @clr: Interrupt clear register.
232  * @is_error: true if this is an error interrupt.
233  * @offset: start delta in the lookup_log_chans in d40_base. If equals to
234  * D40_PHY_CHAN, the lookup_phy_chans shall be used instead.
235  */
236 struct d40_interrupt_lookup {
237 	u32 src;
238 	u32 clr;
239 	bool is_error;
240 	int offset;
241 };
242 
243 
244 static struct d40_interrupt_lookup il_v4a[] = {
245 	{D40_DREG_LCTIS0, D40_DREG_LCICR0, false,  0},
246 	{D40_DREG_LCTIS1, D40_DREG_LCICR1, false, 32},
247 	{D40_DREG_LCTIS2, D40_DREG_LCICR2, false, 64},
248 	{D40_DREG_LCTIS3, D40_DREG_LCICR3, false, 96},
249 	{D40_DREG_LCEIS0, D40_DREG_LCICR0, true,   0},
250 	{D40_DREG_LCEIS1, D40_DREG_LCICR1, true,  32},
251 	{D40_DREG_LCEIS2, D40_DREG_LCICR2, true,  64},
252 	{D40_DREG_LCEIS3, D40_DREG_LCICR3, true,  96},
253 	{D40_DREG_PCTIS,  D40_DREG_PCICR,  false, D40_PHY_CHAN},
254 	{D40_DREG_PCEIS,  D40_DREG_PCICR,  true,  D40_PHY_CHAN},
255 };
256 
257 static struct d40_interrupt_lookup il_v4b[] = {
258 	{D40_DREG_CLCTIS1, D40_DREG_CLCICR1, false,  0},
259 	{D40_DREG_CLCTIS2, D40_DREG_CLCICR2, false, 32},
260 	{D40_DREG_CLCTIS3, D40_DREG_CLCICR3, false, 64},
261 	{D40_DREG_CLCTIS4, D40_DREG_CLCICR4, false, 96},
262 	{D40_DREG_CLCTIS5, D40_DREG_CLCICR5, false, 128},
263 	{D40_DREG_CLCEIS1, D40_DREG_CLCICR1, true,   0},
264 	{D40_DREG_CLCEIS2, D40_DREG_CLCICR2, true,  32},
265 	{D40_DREG_CLCEIS3, D40_DREG_CLCICR3, true,  64},
266 	{D40_DREG_CLCEIS4, D40_DREG_CLCICR4, true,  96},
267 	{D40_DREG_CLCEIS5, D40_DREG_CLCICR5, true,  128},
268 	{D40_DREG_CPCTIS,  D40_DREG_CPCICR,  false, D40_PHY_CHAN},
269 	{D40_DREG_CPCEIS,  D40_DREG_CPCICR,  true,  D40_PHY_CHAN},
270 };
271 
272 /**
273  * struct d40_reg_val - simple lookup struct
274  *
275  * @reg: The register.
276  * @val: The value that belongs to the register in reg.
277  */
278 struct d40_reg_val {
279 	unsigned int reg;
280 	unsigned int val;
281 };
282 
283 static __initdata struct d40_reg_val dma_init_reg_v4a[] = {
284 	/* Clock every part of the DMA block from start */
285 	{ .reg = D40_DREG_GCC,    .val = D40_DREG_GCC_ENABLE_ALL},
286 
287 	/* Interrupts on all logical channels */
288 	{ .reg = D40_DREG_LCMIS0, .val = 0xFFFFFFFF},
289 	{ .reg = D40_DREG_LCMIS1, .val = 0xFFFFFFFF},
290 	{ .reg = D40_DREG_LCMIS2, .val = 0xFFFFFFFF},
291 	{ .reg = D40_DREG_LCMIS3, .val = 0xFFFFFFFF},
292 	{ .reg = D40_DREG_LCICR0, .val = 0xFFFFFFFF},
293 	{ .reg = D40_DREG_LCICR1, .val = 0xFFFFFFFF},
294 	{ .reg = D40_DREG_LCICR2, .val = 0xFFFFFFFF},
295 	{ .reg = D40_DREG_LCICR3, .val = 0xFFFFFFFF},
296 	{ .reg = D40_DREG_LCTIS0, .val = 0xFFFFFFFF},
297 	{ .reg = D40_DREG_LCTIS1, .val = 0xFFFFFFFF},
298 	{ .reg = D40_DREG_LCTIS2, .val = 0xFFFFFFFF},
299 	{ .reg = D40_DREG_LCTIS3, .val = 0xFFFFFFFF}
300 };
301 static __initdata struct d40_reg_val dma_init_reg_v4b[] = {
302 	/* Clock every part of the DMA block from start */
303 	{ .reg = D40_DREG_GCC,    .val = D40_DREG_GCC_ENABLE_ALL},
304 
305 	/* Interrupts on all logical channels */
306 	{ .reg = D40_DREG_CLCMIS1, .val = 0xFFFFFFFF},
307 	{ .reg = D40_DREG_CLCMIS2, .val = 0xFFFFFFFF},
308 	{ .reg = D40_DREG_CLCMIS3, .val = 0xFFFFFFFF},
309 	{ .reg = D40_DREG_CLCMIS4, .val = 0xFFFFFFFF},
310 	{ .reg = D40_DREG_CLCMIS5, .val = 0xFFFFFFFF},
311 	{ .reg = D40_DREG_CLCICR1, .val = 0xFFFFFFFF},
312 	{ .reg = D40_DREG_CLCICR2, .val = 0xFFFFFFFF},
313 	{ .reg = D40_DREG_CLCICR3, .val = 0xFFFFFFFF},
314 	{ .reg = D40_DREG_CLCICR4, .val = 0xFFFFFFFF},
315 	{ .reg = D40_DREG_CLCICR5, .val = 0xFFFFFFFF},
316 	{ .reg = D40_DREG_CLCTIS1, .val = 0xFFFFFFFF},
317 	{ .reg = D40_DREG_CLCTIS2, .val = 0xFFFFFFFF},
318 	{ .reg = D40_DREG_CLCTIS3, .val = 0xFFFFFFFF},
319 	{ .reg = D40_DREG_CLCTIS4, .val = 0xFFFFFFFF},
320 	{ .reg = D40_DREG_CLCTIS5, .val = 0xFFFFFFFF}
321 };
322 
323 /**
324  * struct d40_lli_pool - Structure for keeping LLIs in memory
325  *
326  * @base: Pointer to memory area when the pre_alloc_lli's are not large
327  * enough, IE bigger than the most common case, 1 dst and 1 src. NULL if
328  * pre_alloc_lli is used.
329  * @dma_addr: DMA address, if mapped
330  * @size: The size in bytes of the memory at base or the size of pre_alloc_lli.
331  * @pre_alloc_lli: Pre allocated area for the most common case of transfers,
332  * one buffer to one buffer.
333  */
334 struct d40_lli_pool {
335 	void	*base;
336 	int	 size;
337 	dma_addr_t	dma_addr;
338 	/* Space for dst and src, plus an extra for padding */
339 	u8	 pre_alloc_lli[3 * sizeof(struct d40_phy_lli)];
340 };
341 
342 /**
343  * struct d40_desc - A descriptor is one DMA job.
344  *
345  * @lli_phy: LLI settings for physical channel. Both src and dst=
346  * points into the lli_pool, to base if lli_len > 1 or to pre_alloc_lli if
347  * lli_len equals one.
348  * @lli_log: Same as above but for logical channels.
349  * @lli_pool: The pool with two entries pre-allocated.
350  * @lli_len: Number of llis of current descriptor.
351  * @lli_current: Number of transferred llis.
352  * @lcla_alloc: Number of LCLA entries allocated.
353  * @txd: DMA engine struct. Used for among other things for communication
354  * during a transfer.
355  * @node: List entry.
356  * @is_in_client_list: true if the client owns this descriptor.
357  * @cyclic: true if this is a cyclic job
358  *
359  * This descriptor is used for both logical and physical transfers.
360  */
361 struct d40_desc {
362 	/* LLI physical */
363 	struct d40_phy_lli_bidir	 lli_phy;
364 	/* LLI logical */
365 	struct d40_log_lli_bidir	 lli_log;
366 
367 	struct d40_lli_pool		 lli_pool;
368 	int				 lli_len;
369 	int				 lli_current;
370 	int				 lcla_alloc;
371 
372 	struct dma_async_tx_descriptor	 txd;
373 	struct list_head		 node;
374 
375 	bool				 is_in_client_list;
376 	bool				 cyclic;
377 };
378 
379 /**
380  * struct d40_lcla_pool - LCLA pool settings and data.
381  *
382  * @base: The virtual address of LCLA. 18 bit aligned.
383  * @base_unaligned: The orignal kmalloc pointer, if kmalloc is used.
384  * This pointer is only there for clean-up on error.
385  * @pages: The number of pages needed for all physical channels.
386  * Only used later for clean-up on error
387  * @lock: Lock to protect the content in this struct.
388  * @alloc_map: big map over which LCLA entry is own by which job.
389  */
390 struct d40_lcla_pool {
391 	void		*base;
392 	dma_addr_t	dma_addr;
393 	void		*base_unaligned;
394 	int		 pages;
395 	spinlock_t	 lock;
396 	struct d40_desc	**alloc_map;
397 };
398 
399 /**
400  * struct d40_phy_res - struct for handling eventlines mapped to physical
401  * channels.
402  *
403  * @lock: A lock protection this entity.
404  * @reserved: True if used by secure world or otherwise.
405  * @num: The physical channel number of this entity.
406  * @allocated_src: Bit mapped to show which src event line's are mapped to
407  * this physical channel. Can also be free or physically allocated.
408  * @allocated_dst: Same as for src but is dst.
409  * allocated_dst and allocated_src uses the D40_ALLOC* defines as well as
410  * event line number.
411  * @use_soft_lli: To mark if the linked lists of channel are managed by SW.
412  */
413 struct d40_phy_res {
414 	spinlock_t lock;
415 	bool	   reserved;
416 	int	   num;
417 	u32	   allocated_src;
418 	u32	   allocated_dst;
419 	bool	   use_soft_lli;
420 };
421 
422 struct d40_base;
423 
424 /**
425  * struct d40_chan - Struct that describes a channel.
426  *
427  * @lock: A spinlock to protect this struct.
428  * @log_num: The logical number, if any of this channel.
429  * @pending_tx: The number of pending transfers. Used between interrupt handler
430  * and tasklet.
431  * @busy: Set to true when transfer is ongoing on this channel.
432  * @phy_chan: Pointer to physical channel which this instance runs on. If this
433  * point is NULL, then the channel is not allocated.
434  * @chan: DMA engine handle.
435  * @tasklet: Tasklet that gets scheduled from interrupt context to complete a
436  * transfer and call client callback.
437  * @client: Cliented owned descriptor list.
438  * @pending_queue: Submitted jobs, to be issued by issue_pending()
439  * @active: Active descriptor.
440  * @done: Completed jobs
441  * @queue: Queued jobs.
442  * @prepare_queue: Prepared jobs.
443  * @dma_cfg: The client configuration of this dma channel.
444  * @configured: whether the dma_cfg configuration is valid
445  * @base: Pointer to the device instance struct.
446  * @src_def_cfg: Default cfg register setting for src.
447  * @dst_def_cfg: Default cfg register setting for dst.
448  * @log_def: Default logical channel settings.
449  * @lcpa: Pointer to dst and src lcpa settings.
450  * @runtime_addr: runtime configured address.
451  * @runtime_direction: runtime configured direction.
452  *
453  * This struct can either "be" a logical or a physical channel.
454  */
455 struct d40_chan {
456 	spinlock_t			 lock;
457 	int				 log_num;
458 	int				 pending_tx;
459 	bool				 busy;
460 	struct d40_phy_res		*phy_chan;
461 	struct dma_chan			 chan;
462 	struct tasklet_struct		 tasklet;
463 	struct list_head		 client;
464 	struct list_head		 pending_queue;
465 	struct list_head		 active;
466 	struct list_head		 done;
467 	struct list_head		 queue;
468 	struct list_head		 prepare_queue;
469 	struct stedma40_chan_cfg	 dma_cfg;
470 	bool				 configured;
471 	struct d40_base			*base;
472 	/* Default register configurations */
473 	u32				 src_def_cfg;
474 	u32				 dst_def_cfg;
475 	struct d40_def_lcsp		 log_def;
476 	struct d40_log_lli_full		*lcpa;
477 	/* Runtime reconfiguration */
478 	dma_addr_t			runtime_addr;
479 	enum dma_transfer_direction	runtime_direction;
480 };
481 
482 /**
483  * struct d40_gen_dmac - generic values to represent u8500/u8540 DMA
484  * controller
485  *
486  * @backup: the pointer to the registers address array for backup
487  * @backup_size: the size of the registers address array for backup
488  * @realtime_en: the realtime enable register
489  * @realtime_clear: the realtime clear register
490  * @high_prio_en: the high priority enable register
491  * @high_prio_clear: the high priority clear register
492  * @interrupt_en: the interrupt enable register
493  * @interrupt_clear: the interrupt clear register
494  * @il: the pointer to struct d40_interrupt_lookup
495  * @il_size: the size of d40_interrupt_lookup array
496  * @init_reg: the pointer to the struct d40_reg_val
497  * @init_reg_size: the size of d40_reg_val array
498  */
499 struct d40_gen_dmac {
500 	u32				*backup;
501 	u32				 backup_size;
502 	u32				 realtime_en;
503 	u32				 realtime_clear;
504 	u32				 high_prio_en;
505 	u32				 high_prio_clear;
506 	u32				 interrupt_en;
507 	u32				 interrupt_clear;
508 	struct d40_interrupt_lookup	*il;
509 	u32				 il_size;
510 	struct d40_reg_val		*init_reg;
511 	u32				 init_reg_size;
512 };
513 
514 /**
515  * struct d40_base - The big global struct, one for each probe'd instance.
516  *
517  * @interrupt_lock: Lock used to make sure one interrupt is handle a time.
518  * @execmd_lock: Lock for execute command usage since several channels share
519  * the same physical register.
520  * @dev: The device structure.
521  * @virtbase: The virtual base address of the DMA's register.
522  * @rev: silicon revision detected.
523  * @clk: Pointer to the DMA clock structure.
524  * @phy_start: Physical memory start of the DMA registers.
525  * @phy_size: Size of the DMA register map.
526  * @irq: The IRQ number.
527  * @num_memcpy_chans: The number of channels used for memcpy (mem-to-mem
528  * transfers).
529  * @num_phy_chans: The number of physical channels. Read from HW. This
530  * is the number of available channels for this driver, not counting "Secure
531  * mode" allocated physical channels.
532  * @num_log_chans: The number of logical channels. Calculated from
533  * num_phy_chans.
534  * @dma_both: dma_device channels that can do both memcpy and slave transfers.
535  * @dma_slave: dma_device channels that can do only do slave transfers.
536  * @dma_memcpy: dma_device channels that can do only do memcpy transfers.
537  * @phy_chans: Room for all possible physical channels in system.
538  * @log_chans: Room for all possible logical channels in system.
539  * @lookup_log_chans: Used to map interrupt number to logical channel. Points
540  * to log_chans entries.
541  * @lookup_phy_chans: Used to map interrupt number to physical channel. Points
542  * to phy_chans entries.
543  * @plat_data: Pointer to provided platform_data which is the driver
544  * configuration.
545  * @lcpa_regulator: Pointer to hold the regulator for the esram bank for lcla.
546  * @phy_res: Vector containing all physical channels.
547  * @lcla_pool: lcla pool settings and data.
548  * @lcpa_base: The virtual mapped address of LCPA.
549  * @phy_lcpa: The physical address of the LCPA.
550  * @lcpa_size: The size of the LCPA area.
551  * @desc_slab: cache for descriptors.
552  * @reg_val_backup: Here the values of some hardware registers are stored
553  * before the DMA is powered off. They are restored when the power is back on.
554  * @reg_val_backup_v4: Backup of registers that only exits on dma40 v3 and
555  * later
556  * @reg_val_backup_chan: Backup data for standard channel parameter registers.
557  * @gcc_pwr_off_mask: Mask to maintain the channels that can be turned off.
558  * @initialized: true if the dma has been initialized
559  * @gen_dmac: the struct for generic registers values to represent u8500/8540
560  * DMA controller
561  */
562 struct d40_base {
563 	spinlock_t			 interrupt_lock;
564 	spinlock_t			 execmd_lock;
565 	struct device			 *dev;
566 	void __iomem			 *virtbase;
567 	u8				  rev:4;
568 	struct clk			 *clk;
569 	phys_addr_t			  phy_start;
570 	resource_size_t			  phy_size;
571 	int				  irq;
572 	int				  num_memcpy_chans;
573 	int				  num_phy_chans;
574 	int				  num_log_chans;
575 	struct device_dma_parameters	  dma_parms;
576 	struct dma_device		  dma_both;
577 	struct dma_device		  dma_slave;
578 	struct dma_device		  dma_memcpy;
579 	struct d40_chan			 *phy_chans;
580 	struct d40_chan			 *log_chans;
581 	struct d40_chan			**lookup_log_chans;
582 	struct d40_chan			**lookup_phy_chans;
583 	struct stedma40_platform_data	 *plat_data;
584 	struct regulator		 *lcpa_regulator;
585 	/* Physical half channels */
586 	struct d40_phy_res		 *phy_res;
587 	struct d40_lcla_pool		  lcla_pool;
588 	void				 *lcpa_base;
589 	dma_addr_t			  phy_lcpa;
590 	resource_size_t			  lcpa_size;
591 	struct kmem_cache		 *desc_slab;
592 	u32				  reg_val_backup[BACKUP_REGS_SZ];
593 	u32				  reg_val_backup_v4[BACKUP_REGS_SZ_MAX];
594 	u32				 *reg_val_backup_chan;
595 	u16				  gcc_pwr_off_mask;
596 	bool				  initialized;
597 	struct d40_gen_dmac		  gen_dmac;
598 };
599 
600 static struct device *chan2dev(struct d40_chan *d40c)
601 {
602 	return &d40c->chan.dev->device;
603 }
604 
605 static bool chan_is_physical(struct d40_chan *chan)
606 {
607 	return chan->log_num == D40_PHY_CHAN;
608 }
609 
610 static bool chan_is_logical(struct d40_chan *chan)
611 {
612 	return !chan_is_physical(chan);
613 }
614 
615 static void __iomem *chan_base(struct d40_chan *chan)
616 {
617 	return chan->base->virtbase + D40_DREG_PCBASE +
618 	       chan->phy_chan->num * D40_DREG_PCDELTA;
619 }
620 
621 #define d40_err(dev, format, arg...)		\
622 	dev_err(dev, "[%s] " format, __func__, ## arg)
623 
624 #define chan_err(d40c, format, arg...)		\
625 	d40_err(chan2dev(d40c), format, ## arg)
626 
627 static int d40_pool_lli_alloc(struct d40_chan *d40c, struct d40_desc *d40d,
628 			      int lli_len)
629 {
630 	bool is_log = chan_is_logical(d40c);
631 	u32 align;
632 	void *base;
633 
634 	if (is_log)
635 		align = sizeof(struct d40_log_lli);
636 	else
637 		align = sizeof(struct d40_phy_lli);
638 
639 	if (lli_len == 1) {
640 		base = d40d->lli_pool.pre_alloc_lli;
641 		d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
642 		d40d->lli_pool.base = NULL;
643 	} else {
644 		d40d->lli_pool.size = lli_len * 2 * align;
645 
646 		base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
647 		d40d->lli_pool.base = base;
648 
649 		if (d40d->lli_pool.base == NULL)
650 			return -ENOMEM;
651 	}
652 
653 	if (is_log) {
654 		d40d->lli_log.src = PTR_ALIGN(base, align);
655 		d40d->lli_log.dst = d40d->lli_log.src + lli_len;
656 
657 		d40d->lli_pool.dma_addr = 0;
658 	} else {
659 		d40d->lli_phy.src = PTR_ALIGN(base, align);
660 		d40d->lli_phy.dst = d40d->lli_phy.src + lli_len;
661 
662 		d40d->lli_pool.dma_addr = dma_map_single(d40c->base->dev,
663 							 d40d->lli_phy.src,
664 							 d40d->lli_pool.size,
665 							 DMA_TO_DEVICE);
666 
667 		if (dma_mapping_error(d40c->base->dev,
668 				      d40d->lli_pool.dma_addr)) {
669 			kfree(d40d->lli_pool.base);
670 			d40d->lli_pool.base = NULL;
671 			d40d->lli_pool.dma_addr = 0;
672 			return -ENOMEM;
673 		}
674 	}
675 
676 	return 0;
677 }
678 
679 static void d40_pool_lli_free(struct d40_chan *d40c, struct d40_desc *d40d)
680 {
681 	if (d40d->lli_pool.dma_addr)
682 		dma_unmap_single(d40c->base->dev, d40d->lli_pool.dma_addr,
683 				 d40d->lli_pool.size, DMA_TO_DEVICE);
684 
685 	kfree(d40d->lli_pool.base);
686 	d40d->lli_pool.base = NULL;
687 	d40d->lli_pool.size = 0;
688 	d40d->lli_log.src = NULL;
689 	d40d->lli_log.dst = NULL;
690 	d40d->lli_phy.src = NULL;
691 	d40d->lli_phy.dst = NULL;
692 }
693 
694 static int d40_lcla_alloc_one(struct d40_chan *d40c,
695 			      struct d40_desc *d40d)
696 {
697 	unsigned long flags;
698 	int i;
699 	int ret = -EINVAL;
700 
701 	spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
702 
703 	/*
704 	 * Allocate both src and dst at the same time, therefore the half
705 	 * start on 1 since 0 can't be used since zero is used as end marker.
706 	 */
707 	for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
708 		int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
709 
710 		if (!d40c->base->lcla_pool.alloc_map[idx]) {
711 			d40c->base->lcla_pool.alloc_map[idx] = d40d;
712 			d40d->lcla_alloc++;
713 			ret = i;
714 			break;
715 		}
716 	}
717 
718 	spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
719 
720 	return ret;
721 }
722 
723 static int d40_lcla_free_all(struct d40_chan *d40c,
724 			     struct d40_desc *d40d)
725 {
726 	unsigned long flags;
727 	int i;
728 	int ret = -EINVAL;
729 
730 	if (chan_is_physical(d40c))
731 		return 0;
732 
733 	spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
734 
735 	for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
736 		int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
737 
738 		if (d40c->base->lcla_pool.alloc_map[idx] == d40d) {
739 			d40c->base->lcla_pool.alloc_map[idx] = NULL;
740 			d40d->lcla_alloc--;
741 			if (d40d->lcla_alloc == 0) {
742 				ret = 0;
743 				break;
744 			}
745 		}
746 	}
747 
748 	spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
749 
750 	return ret;
751 
752 }
753 
754 static void d40_desc_remove(struct d40_desc *d40d)
755 {
756 	list_del(&d40d->node);
757 }
758 
759 static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
760 {
761 	struct d40_desc *desc = NULL;
762 
763 	if (!list_empty(&d40c->client)) {
764 		struct d40_desc *d;
765 		struct d40_desc *_d;
766 
767 		list_for_each_entry_safe(d, _d, &d40c->client, node) {
768 			if (async_tx_test_ack(&d->txd)) {
769 				d40_desc_remove(d);
770 				desc = d;
771 				memset(desc, 0, sizeof(*desc));
772 				break;
773 			}
774 		}
775 	}
776 
777 	if (!desc)
778 		desc = kmem_cache_zalloc(d40c->base->desc_slab, GFP_NOWAIT);
779 
780 	if (desc)
781 		INIT_LIST_HEAD(&desc->node);
782 
783 	return desc;
784 }
785 
786 static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
787 {
788 
789 	d40_pool_lli_free(d40c, d40d);
790 	d40_lcla_free_all(d40c, d40d);
791 	kmem_cache_free(d40c->base->desc_slab, d40d);
792 }
793 
794 static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
795 {
796 	list_add_tail(&desc->node, &d40c->active);
797 }
798 
799 static void d40_phy_lli_load(struct d40_chan *chan, struct d40_desc *desc)
800 {
801 	struct d40_phy_lli *lli_dst = desc->lli_phy.dst;
802 	struct d40_phy_lli *lli_src = desc->lli_phy.src;
803 	void __iomem *base = chan_base(chan);
804 
805 	writel(lli_src->reg_cfg, base + D40_CHAN_REG_SSCFG);
806 	writel(lli_src->reg_elt, base + D40_CHAN_REG_SSELT);
807 	writel(lli_src->reg_ptr, base + D40_CHAN_REG_SSPTR);
808 	writel(lli_src->reg_lnk, base + D40_CHAN_REG_SSLNK);
809 
810 	writel(lli_dst->reg_cfg, base + D40_CHAN_REG_SDCFG);
811 	writel(lli_dst->reg_elt, base + D40_CHAN_REG_SDELT);
812 	writel(lli_dst->reg_ptr, base + D40_CHAN_REG_SDPTR);
813 	writel(lli_dst->reg_lnk, base + D40_CHAN_REG_SDLNK);
814 }
815 
816 static void d40_desc_done(struct d40_chan *d40c, struct d40_desc *desc)
817 {
818 	list_add_tail(&desc->node, &d40c->done);
819 }
820 
821 static void d40_log_lli_to_lcxa(struct d40_chan *chan, struct d40_desc *desc)
822 {
823 	struct d40_lcla_pool *pool = &chan->base->lcla_pool;
824 	struct d40_log_lli_bidir *lli = &desc->lli_log;
825 	int lli_current = desc->lli_current;
826 	int lli_len = desc->lli_len;
827 	bool cyclic = desc->cyclic;
828 	int curr_lcla = -EINVAL;
829 	int first_lcla = 0;
830 	bool use_esram_lcla = chan->base->plat_data->use_esram_lcla;
831 	bool linkback;
832 
833 	/*
834 	 * We may have partially running cyclic transfers, in case we did't get
835 	 * enough LCLA entries.
836 	 */
837 	linkback = cyclic && lli_current == 0;
838 
839 	/*
840 	 * For linkback, we need one LCLA even with only one link, because we
841 	 * can't link back to the one in LCPA space
842 	 */
843 	if (linkback || (lli_len - lli_current > 1)) {
844 		/*
845 		 * If the channel is expected to use only soft_lli don't
846 		 * allocate a lcla. This is to avoid a HW issue that exists
847 		 * in some controller during a peripheral to memory transfer
848 		 * that uses linked lists.
849 		 */
850 		if (!(chan->phy_chan->use_soft_lli &&
851 			chan->dma_cfg.dir == DMA_DEV_TO_MEM))
852 			curr_lcla = d40_lcla_alloc_one(chan, desc);
853 
854 		first_lcla = curr_lcla;
855 	}
856 
857 	/*
858 	 * For linkback, we normally load the LCPA in the loop since we need to
859 	 * link it to the second LCLA and not the first.  However, if we
860 	 * couldn't even get a first LCLA, then we have to run in LCPA and
861 	 * reload manually.
862 	 */
863 	if (!linkback || curr_lcla == -EINVAL) {
864 		unsigned int flags = 0;
865 
866 		if (curr_lcla == -EINVAL)
867 			flags |= LLI_TERM_INT;
868 
869 		d40_log_lli_lcpa_write(chan->lcpa,
870 				       &lli->dst[lli_current],
871 				       &lli->src[lli_current],
872 				       curr_lcla,
873 				       flags);
874 		lli_current++;
875 	}
876 
877 	if (curr_lcla < 0)
878 		goto out;
879 
880 	for (; lli_current < lli_len; lli_current++) {
881 		unsigned int lcla_offset = chan->phy_chan->num * 1024 +
882 					   8 * curr_lcla * 2;
883 		struct d40_log_lli *lcla = pool->base + lcla_offset;
884 		unsigned int flags = 0;
885 		int next_lcla;
886 
887 		if (lli_current + 1 < lli_len)
888 			next_lcla = d40_lcla_alloc_one(chan, desc);
889 		else
890 			next_lcla = linkback ? first_lcla : -EINVAL;
891 
892 		if (cyclic || next_lcla == -EINVAL)
893 			flags |= LLI_TERM_INT;
894 
895 		if (linkback && curr_lcla == first_lcla) {
896 			/* First link goes in both LCPA and LCLA */
897 			d40_log_lli_lcpa_write(chan->lcpa,
898 					       &lli->dst[lli_current],
899 					       &lli->src[lli_current],
900 					       next_lcla, flags);
901 		}
902 
903 		/*
904 		 * One unused LCLA in the cyclic case if the very first
905 		 * next_lcla fails...
906 		 */
907 		d40_log_lli_lcla_write(lcla,
908 				       &lli->dst[lli_current],
909 				       &lli->src[lli_current],
910 				       next_lcla, flags);
911 
912 		/*
913 		 * Cache maintenance is not needed if lcla is
914 		 * mapped in esram
915 		 */
916 		if (!use_esram_lcla) {
917 			dma_sync_single_range_for_device(chan->base->dev,
918 						pool->dma_addr, lcla_offset,
919 						2 * sizeof(struct d40_log_lli),
920 						DMA_TO_DEVICE);
921 		}
922 		curr_lcla = next_lcla;
923 
924 		if (curr_lcla == -EINVAL || curr_lcla == first_lcla) {
925 			lli_current++;
926 			break;
927 		}
928 	}
929 
930 out:
931 	desc->lli_current = lli_current;
932 }
933 
934 static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
935 {
936 	if (chan_is_physical(d40c)) {
937 		d40_phy_lli_load(d40c, d40d);
938 		d40d->lli_current = d40d->lli_len;
939 	} else
940 		d40_log_lli_to_lcxa(d40c, d40d);
941 }
942 
943 static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
944 {
945 	struct d40_desc *d;
946 
947 	if (list_empty(&d40c->active))
948 		return NULL;
949 
950 	d = list_first_entry(&d40c->active,
951 			     struct d40_desc,
952 			     node);
953 	return d;
954 }
955 
956 /* remove desc from current queue and add it to the pending_queue */
957 static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
958 {
959 	d40_desc_remove(desc);
960 	desc->is_in_client_list = false;
961 	list_add_tail(&desc->node, &d40c->pending_queue);
962 }
963 
964 static struct d40_desc *d40_first_pending(struct d40_chan *d40c)
965 {
966 	struct d40_desc *d;
967 
968 	if (list_empty(&d40c->pending_queue))
969 		return NULL;
970 
971 	d = list_first_entry(&d40c->pending_queue,
972 			     struct d40_desc,
973 			     node);
974 	return d;
975 }
976 
977 static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
978 {
979 	struct d40_desc *d;
980 
981 	if (list_empty(&d40c->queue))
982 		return NULL;
983 
984 	d = list_first_entry(&d40c->queue,
985 			     struct d40_desc,
986 			     node);
987 	return d;
988 }
989 
990 static struct d40_desc *d40_first_done(struct d40_chan *d40c)
991 {
992 	if (list_empty(&d40c->done))
993 		return NULL;
994 
995 	return list_first_entry(&d40c->done, struct d40_desc, node);
996 }
997 
998 static int d40_psize_2_burst_size(bool is_log, int psize)
999 {
1000 	if (is_log) {
1001 		if (psize == STEDMA40_PSIZE_LOG_1)
1002 			return 1;
1003 	} else {
1004 		if (psize == STEDMA40_PSIZE_PHY_1)
1005 			return 1;
1006 	}
1007 
1008 	return 2 << psize;
1009 }
1010 
1011 /*
1012  * The dma only supports transmitting packages up to
1013  * STEDMA40_MAX_SEG_SIZE * data_width, where data_width is stored in Bytes.
1014  *
1015  * Calculate the total number of dma elements required to send the entire sg list.
1016  */
1017 static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2)
1018 {
1019 	int dmalen;
1020 	u32 max_w = max(data_width1, data_width2);
1021 	u32 min_w = min(data_width1, data_width2);
1022 	u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE * min_w, max_w);
1023 
1024 	if (seg_max > STEDMA40_MAX_SEG_SIZE)
1025 		seg_max -= max_w;
1026 
1027 	if (!IS_ALIGNED(size, max_w))
1028 		return -EINVAL;
1029 
1030 	if (size <= seg_max)
1031 		dmalen = 1;
1032 	else {
1033 		dmalen = size / seg_max;
1034 		if (dmalen * seg_max < size)
1035 			dmalen++;
1036 	}
1037 	return dmalen;
1038 }
1039 
1040 static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len,
1041 			   u32 data_width1, u32 data_width2)
1042 {
1043 	struct scatterlist *sg;
1044 	int i;
1045 	int len = 0;
1046 	int ret;
1047 
1048 	for_each_sg(sgl, sg, sg_len, i) {
1049 		ret = d40_size_2_dmalen(sg_dma_len(sg),
1050 					data_width1, data_width2);
1051 		if (ret < 0)
1052 			return ret;
1053 		len += ret;
1054 	}
1055 	return len;
1056 }
1057 
1058 
1059 #ifdef CONFIG_PM
1060 static void dma40_backup(void __iomem *baseaddr, u32 *backup,
1061 			 u32 *regaddr, int num, bool save)
1062 {
1063 	int i;
1064 
1065 	for (i = 0; i < num; i++) {
1066 		void __iomem *addr = baseaddr + regaddr[i];
1067 
1068 		if (save)
1069 			backup[i] = readl_relaxed(addr);
1070 		else
1071 			writel_relaxed(backup[i], addr);
1072 	}
1073 }
1074 
1075 static void d40_save_restore_registers(struct d40_base *base, bool save)
1076 {
1077 	int i;
1078 
1079 	/* Save/Restore channel specific registers */
1080 	for (i = 0; i < base->num_phy_chans; i++) {
1081 		void __iomem *addr;
1082 		int idx;
1083 
1084 		if (base->phy_res[i].reserved)
1085 			continue;
1086 
1087 		addr = base->virtbase + D40_DREG_PCBASE + i * D40_DREG_PCDELTA;
1088 		idx = i * ARRAY_SIZE(d40_backup_regs_chan);
1089 
1090 		dma40_backup(addr, &base->reg_val_backup_chan[idx],
1091 			     d40_backup_regs_chan,
1092 			     ARRAY_SIZE(d40_backup_regs_chan),
1093 			     save);
1094 	}
1095 
1096 	/* Save/Restore global registers */
1097 	dma40_backup(base->virtbase, base->reg_val_backup,
1098 		     d40_backup_regs, ARRAY_SIZE(d40_backup_regs),
1099 		     save);
1100 
1101 	/* Save/Restore registers only existing on dma40 v3 and later */
1102 	if (base->gen_dmac.backup)
1103 		dma40_backup(base->virtbase, base->reg_val_backup_v4,
1104 			     base->gen_dmac.backup,
1105 			base->gen_dmac.backup_size,
1106 			save);
1107 }
1108 #else
1109 static void d40_save_restore_registers(struct d40_base *base, bool save)
1110 {
1111 }
1112 #endif
1113 
1114 static int __d40_execute_command_phy(struct d40_chan *d40c,
1115 				     enum d40_command command)
1116 {
1117 	u32 status;
1118 	int i;
1119 	void __iomem *active_reg;
1120 	int ret = 0;
1121 	unsigned long flags;
1122 	u32 wmask;
1123 
1124 	if (command == D40_DMA_STOP) {
1125 		ret = __d40_execute_command_phy(d40c, D40_DMA_SUSPEND_REQ);
1126 		if (ret)
1127 			return ret;
1128 	}
1129 
1130 	spin_lock_irqsave(&d40c->base->execmd_lock, flags);
1131 
1132 	if (d40c->phy_chan->num % 2 == 0)
1133 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1134 	else
1135 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1136 
1137 	if (command == D40_DMA_SUSPEND_REQ) {
1138 		status = (readl(active_reg) &
1139 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1140 			D40_CHAN_POS(d40c->phy_chan->num);
1141 
1142 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1143 			goto done;
1144 	}
1145 
1146 	wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
1147 	writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
1148 	       active_reg);
1149 
1150 	if (command == D40_DMA_SUSPEND_REQ) {
1151 
1152 		for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
1153 			status = (readl(active_reg) &
1154 				  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1155 				D40_CHAN_POS(d40c->phy_chan->num);
1156 
1157 			cpu_relax();
1158 			/*
1159 			 * Reduce the number of bus accesses while
1160 			 * waiting for the DMA to suspend.
1161 			 */
1162 			udelay(3);
1163 
1164 			if (status == D40_DMA_STOP ||
1165 			    status == D40_DMA_SUSPENDED)
1166 				break;
1167 		}
1168 
1169 		if (i == D40_SUSPEND_MAX_IT) {
1170 			chan_err(d40c,
1171 				"unable to suspend the chl %d (log: %d) status %x\n",
1172 				d40c->phy_chan->num, d40c->log_num,
1173 				status);
1174 			dump_stack();
1175 			ret = -EBUSY;
1176 		}
1177 
1178 	}
1179 done:
1180 	spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
1181 	return ret;
1182 }
1183 
1184 static void d40_term_all(struct d40_chan *d40c)
1185 {
1186 	struct d40_desc *d40d;
1187 	struct d40_desc *_d;
1188 
1189 	/* Release completed descriptors */
1190 	while ((d40d = d40_first_done(d40c))) {
1191 		d40_desc_remove(d40d);
1192 		d40_desc_free(d40c, d40d);
1193 	}
1194 
1195 	/* Release active descriptors */
1196 	while ((d40d = d40_first_active_get(d40c))) {
1197 		d40_desc_remove(d40d);
1198 		d40_desc_free(d40c, d40d);
1199 	}
1200 
1201 	/* Release queued descriptors waiting for transfer */
1202 	while ((d40d = d40_first_queued(d40c))) {
1203 		d40_desc_remove(d40d);
1204 		d40_desc_free(d40c, d40d);
1205 	}
1206 
1207 	/* Release pending descriptors */
1208 	while ((d40d = d40_first_pending(d40c))) {
1209 		d40_desc_remove(d40d);
1210 		d40_desc_free(d40c, d40d);
1211 	}
1212 
1213 	/* Release client owned descriptors */
1214 	if (!list_empty(&d40c->client))
1215 		list_for_each_entry_safe(d40d, _d, &d40c->client, node) {
1216 			d40_desc_remove(d40d);
1217 			d40_desc_free(d40c, d40d);
1218 		}
1219 
1220 	/* Release descriptors in prepare queue */
1221 	if (!list_empty(&d40c->prepare_queue))
1222 		list_for_each_entry_safe(d40d, _d,
1223 					 &d40c->prepare_queue, node) {
1224 			d40_desc_remove(d40d);
1225 			d40_desc_free(d40c, d40d);
1226 		}
1227 
1228 	d40c->pending_tx = 0;
1229 }
1230 
1231 static void __d40_config_set_event(struct d40_chan *d40c,
1232 				   enum d40_events event_type, u32 event,
1233 				   int reg)
1234 {
1235 	void __iomem *addr = chan_base(d40c) + reg;
1236 	int tries;
1237 	u32 status;
1238 
1239 	switch (event_type) {
1240 
1241 	case D40_DEACTIVATE_EVENTLINE:
1242 
1243 		writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
1244 		       | ~D40_EVENTLINE_MASK(event), addr);
1245 		break;
1246 
1247 	case D40_SUSPEND_REQ_EVENTLINE:
1248 		status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1249 			  D40_EVENTLINE_POS(event);
1250 
1251 		if (status == D40_DEACTIVATE_EVENTLINE ||
1252 		    status == D40_SUSPEND_REQ_EVENTLINE)
1253 			break;
1254 
1255 		writel((D40_SUSPEND_REQ_EVENTLINE << D40_EVENTLINE_POS(event))
1256 		       | ~D40_EVENTLINE_MASK(event), addr);
1257 
1258 		for (tries = 0 ; tries < D40_SUSPEND_MAX_IT; tries++) {
1259 
1260 			status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1261 				  D40_EVENTLINE_POS(event);
1262 
1263 			cpu_relax();
1264 			/*
1265 			 * Reduce the number of bus accesses while
1266 			 * waiting for the DMA to suspend.
1267 			 */
1268 			udelay(3);
1269 
1270 			if (status == D40_DEACTIVATE_EVENTLINE)
1271 				break;
1272 		}
1273 
1274 		if (tries == D40_SUSPEND_MAX_IT) {
1275 			chan_err(d40c,
1276 				"unable to stop the event_line chl %d (log: %d)"
1277 				"status %x\n", d40c->phy_chan->num,
1278 				 d40c->log_num, status);
1279 		}
1280 		break;
1281 
1282 	case D40_ACTIVATE_EVENTLINE:
1283 	/*
1284 	 * The hardware sometimes doesn't register the enable when src and dst
1285 	 * event lines are active on the same logical channel.  Retry to ensure
1286 	 * it does.  Usually only one retry is sufficient.
1287 	 */
1288 		tries = 100;
1289 		while (--tries) {
1290 			writel((D40_ACTIVATE_EVENTLINE <<
1291 				D40_EVENTLINE_POS(event)) |
1292 				~D40_EVENTLINE_MASK(event), addr);
1293 
1294 			if (readl(addr) & D40_EVENTLINE_MASK(event))
1295 				break;
1296 		}
1297 
1298 		if (tries != 99)
1299 			dev_dbg(chan2dev(d40c),
1300 				"[%s] workaround enable S%cLNK (%d tries)\n",
1301 				__func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D',
1302 				100 - tries);
1303 
1304 		WARN_ON(!tries);
1305 		break;
1306 
1307 	case D40_ROUND_EVENTLINE:
1308 		BUG();
1309 		break;
1310 
1311 	}
1312 }
1313 
1314 static void d40_config_set_event(struct d40_chan *d40c,
1315 				 enum d40_events event_type)
1316 {
1317 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
1318 
1319 	/* Enable event line connected to device (or memcpy) */
1320 	if ((d40c->dma_cfg.dir == DMA_DEV_TO_MEM) ||
1321 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
1322 		__d40_config_set_event(d40c, event_type, event,
1323 				       D40_CHAN_REG_SSLNK);
1324 
1325 	if (d40c->dma_cfg.dir !=  DMA_DEV_TO_MEM)
1326 		__d40_config_set_event(d40c, event_type, event,
1327 				       D40_CHAN_REG_SDLNK);
1328 }
1329 
1330 static u32 d40_chan_has_events(struct d40_chan *d40c)
1331 {
1332 	void __iomem *chanbase = chan_base(d40c);
1333 	u32 val;
1334 
1335 	val = readl(chanbase + D40_CHAN_REG_SSLNK);
1336 	val |= readl(chanbase + D40_CHAN_REG_SDLNK);
1337 
1338 	return val;
1339 }
1340 
1341 static int
1342 __d40_execute_command_log(struct d40_chan *d40c, enum d40_command command)
1343 {
1344 	unsigned long flags;
1345 	int ret = 0;
1346 	u32 active_status;
1347 	void __iomem *active_reg;
1348 
1349 	if (d40c->phy_chan->num % 2 == 0)
1350 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1351 	else
1352 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1353 
1354 
1355 	spin_lock_irqsave(&d40c->phy_chan->lock, flags);
1356 
1357 	switch (command) {
1358 	case D40_DMA_STOP:
1359 	case D40_DMA_SUSPEND_REQ:
1360 
1361 		active_status = (readl(active_reg) &
1362 				 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1363 				 D40_CHAN_POS(d40c->phy_chan->num);
1364 
1365 		if (active_status == D40_DMA_RUN)
1366 			d40_config_set_event(d40c, D40_SUSPEND_REQ_EVENTLINE);
1367 		else
1368 			d40_config_set_event(d40c, D40_DEACTIVATE_EVENTLINE);
1369 
1370 		if (!d40_chan_has_events(d40c) && (command == D40_DMA_STOP))
1371 			ret = __d40_execute_command_phy(d40c, command);
1372 
1373 		break;
1374 
1375 	case D40_DMA_RUN:
1376 
1377 		d40_config_set_event(d40c, D40_ACTIVATE_EVENTLINE);
1378 		ret = __d40_execute_command_phy(d40c, command);
1379 		break;
1380 
1381 	case D40_DMA_SUSPENDED:
1382 		BUG();
1383 		break;
1384 	}
1385 
1386 	spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
1387 	return ret;
1388 }
1389 
1390 static int d40_channel_execute_command(struct d40_chan *d40c,
1391 				       enum d40_command command)
1392 {
1393 	if (chan_is_logical(d40c))
1394 		return __d40_execute_command_log(d40c, command);
1395 	else
1396 		return __d40_execute_command_phy(d40c, command);
1397 }
1398 
1399 static u32 d40_get_prmo(struct d40_chan *d40c)
1400 {
1401 	static const unsigned int phy_map[] = {
1402 		[STEDMA40_PCHAN_BASIC_MODE]
1403 			= D40_DREG_PRMO_PCHAN_BASIC,
1404 		[STEDMA40_PCHAN_MODULO_MODE]
1405 			= D40_DREG_PRMO_PCHAN_MODULO,
1406 		[STEDMA40_PCHAN_DOUBLE_DST_MODE]
1407 			= D40_DREG_PRMO_PCHAN_DOUBLE_DST,
1408 	};
1409 	static const unsigned int log_map[] = {
1410 		[STEDMA40_LCHAN_SRC_PHY_DST_LOG]
1411 			= D40_DREG_PRMO_LCHAN_SRC_PHY_DST_LOG,
1412 		[STEDMA40_LCHAN_SRC_LOG_DST_PHY]
1413 			= D40_DREG_PRMO_LCHAN_SRC_LOG_DST_PHY,
1414 		[STEDMA40_LCHAN_SRC_LOG_DST_LOG]
1415 			= D40_DREG_PRMO_LCHAN_SRC_LOG_DST_LOG,
1416 	};
1417 
1418 	if (chan_is_physical(d40c))
1419 		return phy_map[d40c->dma_cfg.mode_opt];
1420 	else
1421 		return log_map[d40c->dma_cfg.mode_opt];
1422 }
1423 
1424 static void d40_config_write(struct d40_chan *d40c)
1425 {
1426 	u32 addr_base;
1427 	u32 var;
1428 
1429 	/* Odd addresses are even addresses + 4 */
1430 	addr_base = (d40c->phy_chan->num % 2) * 4;
1431 	/* Setup channel mode to logical or physical */
1432 	var = ((u32)(chan_is_logical(d40c)) + 1) <<
1433 		D40_CHAN_POS(d40c->phy_chan->num);
1434 	writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
1435 
1436 	/* Setup operational mode option register */
1437 	var = d40_get_prmo(d40c) << D40_CHAN_POS(d40c->phy_chan->num);
1438 
1439 	writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
1440 
1441 	if (chan_is_logical(d40c)) {
1442 		int lidx = (d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS)
1443 			   & D40_SREG_ELEM_LOG_LIDX_MASK;
1444 		void __iomem *chanbase = chan_base(d40c);
1445 
1446 		/* Set default config for CFG reg */
1447 		writel(d40c->src_def_cfg, chanbase + D40_CHAN_REG_SSCFG);
1448 		writel(d40c->dst_def_cfg, chanbase + D40_CHAN_REG_SDCFG);
1449 
1450 		/* Set LIDX for lcla */
1451 		writel(lidx, chanbase + D40_CHAN_REG_SSELT);
1452 		writel(lidx, chanbase + D40_CHAN_REG_SDELT);
1453 
1454 		/* Clear LNK which will be used by d40_chan_has_events() */
1455 		writel(0, chanbase + D40_CHAN_REG_SSLNK);
1456 		writel(0, chanbase + D40_CHAN_REG_SDLNK);
1457 	}
1458 }
1459 
1460 static u32 d40_residue(struct d40_chan *d40c)
1461 {
1462 	u32 num_elt;
1463 
1464 	if (chan_is_logical(d40c))
1465 		num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
1466 			>> D40_MEM_LCSP2_ECNT_POS;
1467 	else {
1468 		u32 val = readl(chan_base(d40c) + D40_CHAN_REG_SDELT);
1469 		num_elt = (val & D40_SREG_ELEM_PHY_ECNT_MASK)
1470 			  >> D40_SREG_ELEM_PHY_ECNT_POS;
1471 	}
1472 
1473 	return num_elt * d40c->dma_cfg.dst_info.data_width;
1474 }
1475 
1476 static bool d40_tx_is_linked(struct d40_chan *d40c)
1477 {
1478 	bool is_link;
1479 
1480 	if (chan_is_logical(d40c))
1481 		is_link = readl(&d40c->lcpa->lcsp3) &  D40_MEM_LCSP3_DLOS_MASK;
1482 	else
1483 		is_link = readl(chan_base(d40c) + D40_CHAN_REG_SDLNK)
1484 			  & D40_SREG_LNK_PHYS_LNK_MASK;
1485 
1486 	return is_link;
1487 }
1488 
1489 static int d40_pause(struct d40_chan *d40c)
1490 {
1491 	int res = 0;
1492 	unsigned long flags;
1493 
1494 	if (!d40c->busy)
1495 		return 0;
1496 
1497 	pm_runtime_get_sync(d40c->base->dev);
1498 	spin_lock_irqsave(&d40c->lock, flags);
1499 
1500 	res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1501 
1502 	pm_runtime_mark_last_busy(d40c->base->dev);
1503 	pm_runtime_put_autosuspend(d40c->base->dev);
1504 	spin_unlock_irqrestore(&d40c->lock, flags);
1505 	return res;
1506 }
1507 
1508 static int d40_resume(struct d40_chan *d40c)
1509 {
1510 	int res = 0;
1511 	unsigned long flags;
1512 
1513 	if (!d40c->busy)
1514 		return 0;
1515 
1516 	spin_lock_irqsave(&d40c->lock, flags);
1517 	pm_runtime_get_sync(d40c->base->dev);
1518 
1519 	/* If bytes left to transfer or linked tx resume job */
1520 	if (d40_residue(d40c) || d40_tx_is_linked(d40c))
1521 		res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1522 
1523 	pm_runtime_mark_last_busy(d40c->base->dev);
1524 	pm_runtime_put_autosuspend(d40c->base->dev);
1525 	spin_unlock_irqrestore(&d40c->lock, flags);
1526 	return res;
1527 }
1528 
1529 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
1530 {
1531 	struct d40_chan *d40c = container_of(tx->chan,
1532 					     struct d40_chan,
1533 					     chan);
1534 	struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
1535 	unsigned long flags;
1536 	dma_cookie_t cookie;
1537 
1538 	spin_lock_irqsave(&d40c->lock, flags);
1539 	cookie = dma_cookie_assign(tx);
1540 	d40_desc_queue(d40c, d40d);
1541 	spin_unlock_irqrestore(&d40c->lock, flags);
1542 
1543 	return cookie;
1544 }
1545 
1546 static int d40_start(struct d40_chan *d40c)
1547 {
1548 	return d40_channel_execute_command(d40c, D40_DMA_RUN);
1549 }
1550 
1551 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
1552 {
1553 	struct d40_desc *d40d;
1554 	int err;
1555 
1556 	/* Start queued jobs, if any */
1557 	d40d = d40_first_queued(d40c);
1558 
1559 	if (d40d != NULL) {
1560 		if (!d40c->busy) {
1561 			d40c->busy = true;
1562 			pm_runtime_get_sync(d40c->base->dev);
1563 		}
1564 
1565 		/* Remove from queue */
1566 		d40_desc_remove(d40d);
1567 
1568 		/* Add to active queue */
1569 		d40_desc_submit(d40c, d40d);
1570 
1571 		/* Initiate DMA job */
1572 		d40_desc_load(d40c, d40d);
1573 
1574 		/* Start dma job */
1575 		err = d40_start(d40c);
1576 
1577 		if (err)
1578 			return NULL;
1579 	}
1580 
1581 	return d40d;
1582 }
1583 
1584 /* called from interrupt context */
1585 static void dma_tc_handle(struct d40_chan *d40c)
1586 {
1587 	struct d40_desc *d40d;
1588 
1589 	/* Get first active entry from list */
1590 	d40d = d40_first_active_get(d40c);
1591 
1592 	if (d40d == NULL)
1593 		return;
1594 
1595 	if (d40d->cyclic) {
1596 		/*
1597 		 * If this was a paritially loaded list, we need to reloaded
1598 		 * it, and only when the list is completed.  We need to check
1599 		 * for done because the interrupt will hit for every link, and
1600 		 * not just the last one.
1601 		 */
1602 		if (d40d->lli_current < d40d->lli_len
1603 		    && !d40_tx_is_linked(d40c)
1604 		    && !d40_residue(d40c)) {
1605 			d40_lcla_free_all(d40c, d40d);
1606 			d40_desc_load(d40c, d40d);
1607 			(void) d40_start(d40c);
1608 
1609 			if (d40d->lli_current == d40d->lli_len)
1610 				d40d->lli_current = 0;
1611 		}
1612 	} else {
1613 		d40_lcla_free_all(d40c, d40d);
1614 
1615 		if (d40d->lli_current < d40d->lli_len) {
1616 			d40_desc_load(d40c, d40d);
1617 			/* Start dma job */
1618 			(void) d40_start(d40c);
1619 			return;
1620 		}
1621 
1622 		if (d40_queue_start(d40c) == NULL) {
1623 			d40c->busy = false;
1624 
1625 			pm_runtime_mark_last_busy(d40c->base->dev);
1626 			pm_runtime_put_autosuspend(d40c->base->dev);
1627 		}
1628 
1629 		d40_desc_remove(d40d);
1630 		d40_desc_done(d40c, d40d);
1631 	}
1632 
1633 	d40c->pending_tx++;
1634 	tasklet_schedule(&d40c->tasklet);
1635 
1636 }
1637 
1638 static void dma_tasklet(unsigned long data)
1639 {
1640 	struct d40_chan *d40c = (struct d40_chan *) data;
1641 	struct d40_desc *d40d;
1642 	unsigned long flags;
1643 	dma_async_tx_callback callback;
1644 	void *callback_param;
1645 
1646 	spin_lock_irqsave(&d40c->lock, flags);
1647 
1648 	/* Get first entry from the done list */
1649 	d40d = d40_first_done(d40c);
1650 	if (d40d == NULL) {
1651 		/* Check if we have reached here for cyclic job */
1652 		d40d = d40_first_active_get(d40c);
1653 		if (d40d == NULL || !d40d->cyclic)
1654 			goto err;
1655 	}
1656 
1657 	if (!d40d->cyclic)
1658 		dma_cookie_complete(&d40d->txd);
1659 
1660 	/*
1661 	 * If terminating a channel pending_tx is set to zero.
1662 	 * This prevents any finished active jobs to return to the client.
1663 	 */
1664 	if (d40c->pending_tx == 0) {
1665 		spin_unlock_irqrestore(&d40c->lock, flags);
1666 		return;
1667 	}
1668 
1669 	/* Callback to client */
1670 	callback = d40d->txd.callback;
1671 	callback_param = d40d->txd.callback_param;
1672 
1673 	if (!d40d->cyclic) {
1674 		if (async_tx_test_ack(&d40d->txd)) {
1675 			d40_desc_remove(d40d);
1676 			d40_desc_free(d40c, d40d);
1677 		} else if (!d40d->is_in_client_list) {
1678 			d40_desc_remove(d40d);
1679 			d40_lcla_free_all(d40c, d40d);
1680 			list_add_tail(&d40d->node, &d40c->client);
1681 			d40d->is_in_client_list = true;
1682 		}
1683 	}
1684 
1685 	d40c->pending_tx--;
1686 
1687 	if (d40c->pending_tx)
1688 		tasklet_schedule(&d40c->tasklet);
1689 
1690 	spin_unlock_irqrestore(&d40c->lock, flags);
1691 
1692 	if (callback && (d40d->txd.flags & DMA_PREP_INTERRUPT))
1693 		callback(callback_param);
1694 
1695 	return;
1696 
1697 err:
1698 	/* Rescue manouver if receiving double interrupts */
1699 	if (d40c->pending_tx > 0)
1700 		d40c->pending_tx--;
1701 	spin_unlock_irqrestore(&d40c->lock, flags);
1702 }
1703 
1704 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1705 {
1706 	int i;
1707 	u32 idx;
1708 	u32 row;
1709 	long chan = -1;
1710 	struct d40_chan *d40c;
1711 	unsigned long flags;
1712 	struct d40_base *base = data;
1713 	u32 regs[base->gen_dmac.il_size];
1714 	struct d40_interrupt_lookup *il = base->gen_dmac.il;
1715 	u32 il_size = base->gen_dmac.il_size;
1716 
1717 	spin_lock_irqsave(&base->interrupt_lock, flags);
1718 
1719 	/* Read interrupt status of both logical and physical channels */
1720 	for (i = 0; i < il_size; i++)
1721 		regs[i] = readl(base->virtbase + il[i].src);
1722 
1723 	for (;;) {
1724 
1725 		chan = find_next_bit((unsigned long *)regs,
1726 				     BITS_PER_LONG * il_size, chan + 1);
1727 
1728 		/* No more set bits found? */
1729 		if (chan == BITS_PER_LONG * il_size)
1730 			break;
1731 
1732 		row = chan / BITS_PER_LONG;
1733 		idx = chan & (BITS_PER_LONG - 1);
1734 
1735 		if (il[row].offset == D40_PHY_CHAN)
1736 			d40c = base->lookup_phy_chans[idx];
1737 		else
1738 			d40c = base->lookup_log_chans[il[row].offset + idx];
1739 
1740 		if (!d40c) {
1741 			/*
1742 			 * No error because this can happen if something else
1743 			 * in the system is using the channel.
1744 			 */
1745 			continue;
1746 		}
1747 
1748 		/* ACK interrupt */
1749 		writel(BIT(idx), base->virtbase + il[row].clr);
1750 
1751 		spin_lock(&d40c->lock);
1752 
1753 		if (!il[row].is_error)
1754 			dma_tc_handle(d40c);
1755 		else
1756 			d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1757 				chan, il[row].offset, idx);
1758 
1759 		spin_unlock(&d40c->lock);
1760 	}
1761 
1762 	spin_unlock_irqrestore(&base->interrupt_lock, flags);
1763 
1764 	return IRQ_HANDLED;
1765 }
1766 
1767 static int d40_validate_conf(struct d40_chan *d40c,
1768 			     struct stedma40_chan_cfg *conf)
1769 {
1770 	int res = 0;
1771 	bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1772 
1773 	if (!conf->dir) {
1774 		chan_err(d40c, "Invalid direction.\n");
1775 		res = -EINVAL;
1776 	}
1777 
1778 	if ((is_log && conf->dev_type > d40c->base->num_log_chans)  ||
1779 	    (!is_log && conf->dev_type > d40c->base->num_phy_chans) ||
1780 	    (conf->dev_type < 0)) {
1781 		chan_err(d40c, "Invalid device type (%d)\n", conf->dev_type);
1782 		res = -EINVAL;
1783 	}
1784 
1785 	if (conf->dir == DMA_DEV_TO_DEV) {
1786 		/*
1787 		 * DMAC HW supports it. Will be added to this driver,
1788 		 * in case any dma client requires it.
1789 		 */
1790 		chan_err(d40c, "periph to periph not supported\n");
1791 		res = -EINVAL;
1792 	}
1793 
1794 	if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1795 	    conf->src_info.data_width !=
1796 	    d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1797 	    conf->dst_info.data_width) {
1798 		/*
1799 		 * The DMAC hardware only supports
1800 		 * src (burst x width) == dst (burst x width)
1801 		 */
1802 
1803 		chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1804 		res = -EINVAL;
1805 	}
1806 
1807 	return res;
1808 }
1809 
1810 static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1811 			       bool is_src, int log_event_line, bool is_log,
1812 			       bool *first_user)
1813 {
1814 	unsigned long flags;
1815 	spin_lock_irqsave(&phy->lock, flags);
1816 
1817 	*first_user = ((phy->allocated_src | phy->allocated_dst)
1818 			== D40_ALLOC_FREE);
1819 
1820 	if (!is_log) {
1821 		/* Physical interrupts are masked per physical full channel */
1822 		if (phy->allocated_src == D40_ALLOC_FREE &&
1823 		    phy->allocated_dst == D40_ALLOC_FREE) {
1824 			phy->allocated_dst = D40_ALLOC_PHY;
1825 			phy->allocated_src = D40_ALLOC_PHY;
1826 			goto found;
1827 		} else
1828 			goto not_found;
1829 	}
1830 
1831 	/* Logical channel */
1832 	if (is_src) {
1833 		if (phy->allocated_src == D40_ALLOC_PHY)
1834 			goto not_found;
1835 
1836 		if (phy->allocated_src == D40_ALLOC_FREE)
1837 			phy->allocated_src = D40_ALLOC_LOG_FREE;
1838 
1839 		if (!(phy->allocated_src & BIT(log_event_line))) {
1840 			phy->allocated_src |= BIT(log_event_line);
1841 			goto found;
1842 		} else
1843 			goto not_found;
1844 	} else {
1845 		if (phy->allocated_dst == D40_ALLOC_PHY)
1846 			goto not_found;
1847 
1848 		if (phy->allocated_dst == D40_ALLOC_FREE)
1849 			phy->allocated_dst = D40_ALLOC_LOG_FREE;
1850 
1851 		if (!(phy->allocated_dst & BIT(log_event_line))) {
1852 			phy->allocated_dst |= BIT(log_event_line);
1853 			goto found;
1854 		} else
1855 			goto not_found;
1856 	}
1857 
1858 not_found:
1859 	spin_unlock_irqrestore(&phy->lock, flags);
1860 	return false;
1861 found:
1862 	spin_unlock_irqrestore(&phy->lock, flags);
1863 	return true;
1864 }
1865 
1866 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1867 			       int log_event_line)
1868 {
1869 	unsigned long flags;
1870 	bool is_free = false;
1871 
1872 	spin_lock_irqsave(&phy->lock, flags);
1873 	if (!log_event_line) {
1874 		phy->allocated_dst = D40_ALLOC_FREE;
1875 		phy->allocated_src = D40_ALLOC_FREE;
1876 		is_free = true;
1877 		goto out;
1878 	}
1879 
1880 	/* Logical channel */
1881 	if (is_src) {
1882 		phy->allocated_src &= ~BIT(log_event_line);
1883 		if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1884 			phy->allocated_src = D40_ALLOC_FREE;
1885 	} else {
1886 		phy->allocated_dst &= ~BIT(log_event_line);
1887 		if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1888 			phy->allocated_dst = D40_ALLOC_FREE;
1889 	}
1890 
1891 	is_free = ((phy->allocated_src | phy->allocated_dst) ==
1892 		   D40_ALLOC_FREE);
1893 
1894 out:
1895 	spin_unlock_irqrestore(&phy->lock, flags);
1896 
1897 	return is_free;
1898 }
1899 
1900 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1901 {
1902 	int dev_type = d40c->dma_cfg.dev_type;
1903 	int event_group;
1904 	int event_line;
1905 	struct d40_phy_res *phys;
1906 	int i;
1907 	int j;
1908 	int log_num;
1909 	int num_phy_chans;
1910 	bool is_src;
1911 	bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1912 
1913 	phys = d40c->base->phy_res;
1914 	num_phy_chans = d40c->base->num_phy_chans;
1915 
1916 	if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
1917 		log_num = 2 * dev_type;
1918 		is_src = true;
1919 	} else if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
1920 		   d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1921 		/* dst event lines are used for logical memcpy */
1922 		log_num = 2 * dev_type + 1;
1923 		is_src = false;
1924 	} else
1925 		return -EINVAL;
1926 
1927 	event_group = D40_TYPE_TO_GROUP(dev_type);
1928 	event_line = D40_TYPE_TO_EVENT(dev_type);
1929 
1930 	if (!is_log) {
1931 		if (d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1932 			/* Find physical half channel */
1933 			if (d40c->dma_cfg.use_fixed_channel) {
1934 				i = d40c->dma_cfg.phy_channel;
1935 				if (d40_alloc_mask_set(&phys[i], is_src,
1936 						       0, is_log,
1937 						       first_phy_user))
1938 					goto found_phy;
1939 			} else {
1940 				for (i = 0; i < num_phy_chans; i++) {
1941 					if (d40_alloc_mask_set(&phys[i], is_src,
1942 						       0, is_log,
1943 						       first_phy_user))
1944 						goto found_phy;
1945 				}
1946 			}
1947 		} else
1948 			for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1949 				int phy_num = j  + event_group * 2;
1950 				for (i = phy_num; i < phy_num + 2; i++) {
1951 					if (d40_alloc_mask_set(&phys[i],
1952 							       is_src,
1953 							       0,
1954 							       is_log,
1955 							       first_phy_user))
1956 						goto found_phy;
1957 				}
1958 			}
1959 		return -EINVAL;
1960 found_phy:
1961 		d40c->phy_chan = &phys[i];
1962 		d40c->log_num = D40_PHY_CHAN;
1963 		goto out;
1964 	}
1965 	if (dev_type == -1)
1966 		return -EINVAL;
1967 
1968 	/* Find logical channel */
1969 	for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1970 		int phy_num = j + event_group * 2;
1971 
1972 		if (d40c->dma_cfg.use_fixed_channel) {
1973 			i = d40c->dma_cfg.phy_channel;
1974 
1975 			if ((i != phy_num) && (i != phy_num + 1)) {
1976 				dev_err(chan2dev(d40c),
1977 					"invalid fixed phy channel %d\n", i);
1978 				return -EINVAL;
1979 			}
1980 
1981 			if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1982 					       is_log, first_phy_user))
1983 				goto found_log;
1984 
1985 			dev_err(chan2dev(d40c),
1986 				"could not allocate fixed phy channel %d\n", i);
1987 			return -EINVAL;
1988 		}
1989 
1990 		/*
1991 		 * Spread logical channels across all available physical rather
1992 		 * than pack every logical channel at the first available phy
1993 		 * channels.
1994 		 */
1995 		if (is_src) {
1996 			for (i = phy_num; i < phy_num + 2; i++) {
1997 				if (d40_alloc_mask_set(&phys[i], is_src,
1998 						       event_line, is_log,
1999 						       first_phy_user))
2000 					goto found_log;
2001 			}
2002 		} else {
2003 			for (i = phy_num + 1; i >= phy_num; i--) {
2004 				if (d40_alloc_mask_set(&phys[i], is_src,
2005 						       event_line, is_log,
2006 						       first_phy_user))
2007 					goto found_log;
2008 			}
2009 		}
2010 	}
2011 	return -EINVAL;
2012 
2013 found_log:
2014 	d40c->phy_chan = &phys[i];
2015 	d40c->log_num = log_num;
2016 out:
2017 
2018 	if (is_log)
2019 		d40c->base->lookup_log_chans[d40c->log_num] = d40c;
2020 	else
2021 		d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
2022 
2023 	return 0;
2024 
2025 }
2026 
2027 static int d40_config_memcpy(struct d40_chan *d40c)
2028 {
2029 	dma_cap_mask_t cap = d40c->chan.device->cap_mask;
2030 
2031 	if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
2032 		d40c->dma_cfg = dma40_memcpy_conf_log;
2033 		d40c->dma_cfg.dev_type = dma40_memcpy_channels[d40c->chan.chan_id];
2034 
2035 		d40_log_cfg(&d40c->dma_cfg,
2036 			    &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2037 
2038 	} else if (dma_has_cap(DMA_MEMCPY, cap) &&
2039 		   dma_has_cap(DMA_SLAVE, cap)) {
2040 		d40c->dma_cfg = dma40_memcpy_conf_phy;
2041 
2042 		/* Generate interrrupt at end of transfer or relink. */
2043 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_TIM_POS);
2044 
2045 		/* Generate interrupt on error. */
2046 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
2047 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
2048 
2049 	} else {
2050 		chan_err(d40c, "No memcpy\n");
2051 		return -EINVAL;
2052 	}
2053 
2054 	return 0;
2055 }
2056 
2057 static int d40_free_dma(struct d40_chan *d40c)
2058 {
2059 
2060 	int res = 0;
2061 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2062 	struct d40_phy_res *phy = d40c->phy_chan;
2063 	bool is_src;
2064 
2065 	/* Terminate all queued and active transfers */
2066 	d40_term_all(d40c);
2067 
2068 	if (phy == NULL) {
2069 		chan_err(d40c, "phy == null\n");
2070 		return -EINVAL;
2071 	}
2072 
2073 	if (phy->allocated_src == D40_ALLOC_FREE &&
2074 	    phy->allocated_dst == D40_ALLOC_FREE) {
2075 		chan_err(d40c, "channel already free\n");
2076 		return -EINVAL;
2077 	}
2078 
2079 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2080 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM)
2081 		is_src = false;
2082 	else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2083 		is_src = true;
2084 	else {
2085 		chan_err(d40c, "Unknown direction\n");
2086 		return -EINVAL;
2087 	}
2088 
2089 	pm_runtime_get_sync(d40c->base->dev);
2090 	res = d40_channel_execute_command(d40c, D40_DMA_STOP);
2091 	if (res) {
2092 		chan_err(d40c, "stop failed\n");
2093 		goto out;
2094 	}
2095 
2096 	d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
2097 
2098 	if (chan_is_logical(d40c))
2099 		d40c->base->lookup_log_chans[d40c->log_num] = NULL;
2100 	else
2101 		d40c->base->lookup_phy_chans[phy->num] = NULL;
2102 
2103 	if (d40c->busy) {
2104 		pm_runtime_mark_last_busy(d40c->base->dev);
2105 		pm_runtime_put_autosuspend(d40c->base->dev);
2106 	}
2107 
2108 	d40c->busy = false;
2109 	d40c->phy_chan = NULL;
2110 	d40c->configured = false;
2111 out:
2112 
2113 	pm_runtime_mark_last_busy(d40c->base->dev);
2114 	pm_runtime_put_autosuspend(d40c->base->dev);
2115 	return res;
2116 }
2117 
2118 static bool d40_is_paused(struct d40_chan *d40c)
2119 {
2120 	void __iomem *chanbase = chan_base(d40c);
2121 	bool is_paused = false;
2122 	unsigned long flags;
2123 	void __iomem *active_reg;
2124 	u32 status;
2125 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2126 
2127 	spin_lock_irqsave(&d40c->lock, flags);
2128 
2129 	if (chan_is_physical(d40c)) {
2130 		if (d40c->phy_chan->num % 2 == 0)
2131 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
2132 		else
2133 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
2134 
2135 		status = (readl(active_reg) &
2136 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
2137 			D40_CHAN_POS(d40c->phy_chan->num);
2138 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
2139 			is_paused = true;
2140 
2141 		goto _exit;
2142 	}
2143 
2144 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2145 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
2146 		status = readl(chanbase + D40_CHAN_REG_SDLNK);
2147 	} else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
2148 		status = readl(chanbase + D40_CHAN_REG_SSLNK);
2149 	} else {
2150 		chan_err(d40c, "Unknown direction\n");
2151 		goto _exit;
2152 	}
2153 
2154 	status = (status & D40_EVENTLINE_MASK(event)) >>
2155 		D40_EVENTLINE_POS(event);
2156 
2157 	if (status != D40_DMA_RUN)
2158 		is_paused = true;
2159 _exit:
2160 	spin_unlock_irqrestore(&d40c->lock, flags);
2161 	return is_paused;
2162 
2163 }
2164 
2165 static u32 stedma40_residue(struct dma_chan *chan)
2166 {
2167 	struct d40_chan *d40c =
2168 		container_of(chan, struct d40_chan, chan);
2169 	u32 bytes_left;
2170 	unsigned long flags;
2171 
2172 	spin_lock_irqsave(&d40c->lock, flags);
2173 	bytes_left = d40_residue(d40c);
2174 	spin_unlock_irqrestore(&d40c->lock, flags);
2175 
2176 	return bytes_left;
2177 }
2178 
2179 static int
2180 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
2181 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2182 		unsigned int sg_len, dma_addr_t src_dev_addr,
2183 		dma_addr_t dst_dev_addr)
2184 {
2185 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2186 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2187 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2188 	int ret;
2189 
2190 	ret = d40_log_sg_to_lli(sg_src, sg_len,
2191 				src_dev_addr,
2192 				desc->lli_log.src,
2193 				chan->log_def.lcsp1,
2194 				src_info->data_width,
2195 				dst_info->data_width);
2196 
2197 	ret = d40_log_sg_to_lli(sg_dst, sg_len,
2198 				dst_dev_addr,
2199 				desc->lli_log.dst,
2200 				chan->log_def.lcsp3,
2201 				dst_info->data_width,
2202 				src_info->data_width);
2203 
2204 	return ret < 0 ? ret : 0;
2205 }
2206 
2207 static int
2208 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2209 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2210 		unsigned int sg_len, dma_addr_t src_dev_addr,
2211 		dma_addr_t dst_dev_addr)
2212 {
2213 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2214 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2215 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2216 	unsigned long flags = 0;
2217 	int ret;
2218 
2219 	if (desc->cyclic)
2220 		flags |= LLI_CYCLIC | LLI_TERM_INT;
2221 
2222 	ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2223 				desc->lli_phy.src,
2224 				virt_to_phys(desc->lli_phy.src),
2225 				chan->src_def_cfg,
2226 				src_info, dst_info, flags);
2227 
2228 	ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2229 				desc->lli_phy.dst,
2230 				virt_to_phys(desc->lli_phy.dst),
2231 				chan->dst_def_cfg,
2232 				dst_info, src_info, flags);
2233 
2234 	dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2235 				   desc->lli_pool.size, DMA_TO_DEVICE);
2236 
2237 	return ret < 0 ? ret : 0;
2238 }
2239 
2240 static struct d40_desc *
2241 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2242 	      unsigned int sg_len, unsigned long dma_flags)
2243 {
2244 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2245 	struct d40_desc *desc;
2246 	int ret;
2247 
2248 	desc = d40_desc_get(chan);
2249 	if (!desc)
2250 		return NULL;
2251 
2252 	desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2253 					cfg->dst_info.data_width);
2254 	if (desc->lli_len < 0) {
2255 		chan_err(chan, "Unaligned size\n");
2256 		goto err;
2257 	}
2258 
2259 	ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2260 	if (ret < 0) {
2261 		chan_err(chan, "Could not allocate lli\n");
2262 		goto err;
2263 	}
2264 
2265 	desc->lli_current = 0;
2266 	desc->txd.flags = dma_flags;
2267 	desc->txd.tx_submit = d40_tx_submit;
2268 
2269 	dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2270 
2271 	return desc;
2272 
2273 err:
2274 	d40_desc_free(chan, desc);
2275 	return NULL;
2276 }
2277 
2278 static struct dma_async_tx_descriptor *
2279 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2280 	    struct scatterlist *sg_dst, unsigned int sg_len,
2281 	    enum dma_transfer_direction direction, unsigned long dma_flags)
2282 {
2283 	struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2284 	dma_addr_t src_dev_addr = 0;
2285 	dma_addr_t dst_dev_addr = 0;
2286 	struct d40_desc *desc;
2287 	unsigned long flags;
2288 	int ret;
2289 
2290 	if (!chan->phy_chan) {
2291 		chan_err(chan, "Cannot prepare unallocated channel\n");
2292 		return NULL;
2293 	}
2294 
2295 	spin_lock_irqsave(&chan->lock, flags);
2296 
2297 	desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2298 	if (desc == NULL)
2299 		goto err;
2300 
2301 	if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2302 		desc->cyclic = true;
2303 
2304 	if (direction == DMA_DEV_TO_MEM)
2305 		src_dev_addr = chan->runtime_addr;
2306 	else if (direction == DMA_MEM_TO_DEV)
2307 		dst_dev_addr = chan->runtime_addr;
2308 
2309 	if (chan_is_logical(chan))
2310 		ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2311 				      sg_len, src_dev_addr, dst_dev_addr);
2312 	else
2313 		ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2314 				      sg_len, src_dev_addr, dst_dev_addr);
2315 
2316 	if (ret) {
2317 		chan_err(chan, "Failed to prepare %s sg job: %d\n",
2318 			 chan_is_logical(chan) ? "log" : "phy", ret);
2319 		goto err;
2320 	}
2321 
2322 	/*
2323 	 * add descriptor to the prepare queue in order to be able
2324 	 * to free them later in terminate_all
2325 	 */
2326 	list_add_tail(&desc->node, &chan->prepare_queue);
2327 
2328 	spin_unlock_irqrestore(&chan->lock, flags);
2329 
2330 	return &desc->txd;
2331 
2332 err:
2333 	if (desc)
2334 		d40_desc_free(chan, desc);
2335 	spin_unlock_irqrestore(&chan->lock, flags);
2336 	return NULL;
2337 }
2338 
2339 bool stedma40_filter(struct dma_chan *chan, void *data)
2340 {
2341 	struct stedma40_chan_cfg *info = data;
2342 	struct d40_chan *d40c =
2343 		container_of(chan, struct d40_chan, chan);
2344 	int err;
2345 
2346 	if (data) {
2347 		err = d40_validate_conf(d40c, info);
2348 		if (!err)
2349 			d40c->dma_cfg = *info;
2350 	} else
2351 		err = d40_config_memcpy(d40c);
2352 
2353 	if (!err)
2354 		d40c->configured = true;
2355 
2356 	return err == 0;
2357 }
2358 EXPORT_SYMBOL(stedma40_filter);
2359 
2360 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2361 {
2362 	bool realtime = d40c->dma_cfg.realtime;
2363 	bool highprio = d40c->dma_cfg.high_priority;
2364 	u32 rtreg;
2365 	u32 event = D40_TYPE_TO_EVENT(dev_type);
2366 	u32 group = D40_TYPE_TO_GROUP(dev_type);
2367 	u32 bit = BIT(event);
2368 	u32 prioreg;
2369 	struct d40_gen_dmac *dmac = &d40c->base->gen_dmac;
2370 
2371 	rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear;
2372 	/*
2373 	 * Due to a hardware bug, in some cases a logical channel triggered by
2374 	 * a high priority destination event line can generate extra packet
2375 	 * transactions.
2376 	 *
2377 	 * The workaround is to not set the high priority level for the
2378 	 * destination event lines that trigger logical channels.
2379 	 */
2380 	if (!src && chan_is_logical(d40c))
2381 		highprio = false;
2382 
2383 	prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear;
2384 
2385 	/* Destination event lines are stored in the upper halfword */
2386 	if (!src)
2387 		bit <<= 16;
2388 
2389 	writel(bit, d40c->base->virtbase + prioreg + group * 4);
2390 	writel(bit, d40c->base->virtbase + rtreg + group * 4);
2391 }
2392 
2393 static void d40_set_prio_realtime(struct d40_chan *d40c)
2394 {
2395 	if (d40c->base->rev < 3)
2396 		return;
2397 
2398 	if ((d40c->dma_cfg.dir ==  DMA_DEV_TO_MEM) ||
2399 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2400 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, true);
2401 
2402 	if ((d40c->dma_cfg.dir ==  DMA_MEM_TO_DEV) ||
2403 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2404 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, false);
2405 }
2406 
2407 #define D40_DT_FLAGS_MODE(flags)       ((flags >> 0) & 0x1)
2408 #define D40_DT_FLAGS_DIR(flags)        ((flags >> 1) & 0x1)
2409 #define D40_DT_FLAGS_BIG_ENDIAN(flags) ((flags >> 2) & 0x1)
2410 #define D40_DT_FLAGS_FIXED_CHAN(flags) ((flags >> 3) & 0x1)
2411 
2412 static struct dma_chan *d40_xlate(struct of_phandle_args *dma_spec,
2413 				  struct of_dma *ofdma)
2414 {
2415 	struct stedma40_chan_cfg cfg;
2416 	dma_cap_mask_t cap;
2417 	u32 flags;
2418 
2419 	memset(&cfg, 0, sizeof(struct stedma40_chan_cfg));
2420 
2421 	dma_cap_zero(cap);
2422 	dma_cap_set(DMA_SLAVE, cap);
2423 
2424 	cfg.dev_type = dma_spec->args[0];
2425 	flags = dma_spec->args[2];
2426 
2427 	switch (D40_DT_FLAGS_MODE(flags)) {
2428 	case 0: cfg.mode = STEDMA40_MODE_LOGICAL; break;
2429 	case 1: cfg.mode = STEDMA40_MODE_PHYSICAL; break;
2430 	}
2431 
2432 	switch (D40_DT_FLAGS_DIR(flags)) {
2433 	case 0:
2434 		cfg.dir = DMA_MEM_TO_DEV;
2435 		cfg.dst_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2436 		break;
2437 	case 1:
2438 		cfg.dir = DMA_DEV_TO_MEM;
2439 		cfg.src_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2440 		break;
2441 	}
2442 
2443 	if (D40_DT_FLAGS_FIXED_CHAN(flags)) {
2444 		cfg.phy_channel = dma_spec->args[1];
2445 		cfg.use_fixed_channel = true;
2446 	}
2447 
2448 	return dma_request_channel(cap, stedma40_filter, &cfg);
2449 }
2450 
2451 /* DMA ENGINE functions */
2452 static int d40_alloc_chan_resources(struct dma_chan *chan)
2453 {
2454 	int err;
2455 	unsigned long flags;
2456 	struct d40_chan *d40c =
2457 		container_of(chan, struct d40_chan, chan);
2458 	bool is_free_phy;
2459 	spin_lock_irqsave(&d40c->lock, flags);
2460 
2461 	dma_cookie_init(chan);
2462 
2463 	/* If no dma configuration is set use default configuration (memcpy) */
2464 	if (!d40c->configured) {
2465 		err = d40_config_memcpy(d40c);
2466 		if (err) {
2467 			chan_err(d40c, "Failed to configure memcpy channel\n");
2468 			goto fail;
2469 		}
2470 	}
2471 
2472 	err = d40_allocate_channel(d40c, &is_free_phy);
2473 	if (err) {
2474 		chan_err(d40c, "Failed to allocate channel\n");
2475 		d40c->configured = false;
2476 		goto fail;
2477 	}
2478 
2479 	pm_runtime_get_sync(d40c->base->dev);
2480 
2481 	d40_set_prio_realtime(d40c);
2482 
2483 	if (chan_is_logical(d40c)) {
2484 		if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2485 			d40c->lcpa = d40c->base->lcpa_base +
2486 				d40c->dma_cfg.dev_type * D40_LCPA_CHAN_SIZE;
2487 		else
2488 			d40c->lcpa = d40c->base->lcpa_base +
2489 				d40c->dma_cfg.dev_type *
2490 				D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2491 
2492 		/* Unmask the Global Interrupt Mask. */
2493 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2494 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2495 	}
2496 
2497 	dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2498 		 chan_is_logical(d40c) ? "logical" : "physical",
2499 		 d40c->phy_chan->num,
2500 		 d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2501 
2502 
2503 	/*
2504 	 * Only write channel configuration to the DMA if the physical
2505 	 * resource is free. In case of multiple logical channels
2506 	 * on the same physical resource, only the first write is necessary.
2507 	 */
2508 	if (is_free_phy)
2509 		d40_config_write(d40c);
2510 fail:
2511 	pm_runtime_mark_last_busy(d40c->base->dev);
2512 	pm_runtime_put_autosuspend(d40c->base->dev);
2513 	spin_unlock_irqrestore(&d40c->lock, flags);
2514 	return err;
2515 }
2516 
2517 static void d40_free_chan_resources(struct dma_chan *chan)
2518 {
2519 	struct d40_chan *d40c =
2520 		container_of(chan, struct d40_chan, chan);
2521 	int err;
2522 	unsigned long flags;
2523 
2524 	if (d40c->phy_chan == NULL) {
2525 		chan_err(d40c, "Cannot free unallocated channel\n");
2526 		return;
2527 	}
2528 
2529 	spin_lock_irqsave(&d40c->lock, flags);
2530 
2531 	err = d40_free_dma(d40c);
2532 
2533 	if (err)
2534 		chan_err(d40c, "Failed to free channel\n");
2535 	spin_unlock_irqrestore(&d40c->lock, flags);
2536 }
2537 
2538 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2539 						       dma_addr_t dst,
2540 						       dma_addr_t src,
2541 						       size_t size,
2542 						       unsigned long dma_flags)
2543 {
2544 	struct scatterlist dst_sg;
2545 	struct scatterlist src_sg;
2546 
2547 	sg_init_table(&dst_sg, 1);
2548 	sg_init_table(&src_sg, 1);
2549 
2550 	sg_dma_address(&dst_sg) = dst;
2551 	sg_dma_address(&src_sg) = src;
2552 
2553 	sg_dma_len(&dst_sg) = size;
2554 	sg_dma_len(&src_sg) = size;
2555 
2556 	return d40_prep_sg(chan, &src_sg, &dst_sg, 1, DMA_NONE, dma_flags);
2557 }
2558 
2559 static struct dma_async_tx_descriptor *
2560 d40_prep_memcpy_sg(struct dma_chan *chan,
2561 		   struct scatterlist *dst_sg, unsigned int dst_nents,
2562 		   struct scatterlist *src_sg, unsigned int src_nents,
2563 		   unsigned long dma_flags)
2564 {
2565 	if (dst_nents != src_nents)
2566 		return NULL;
2567 
2568 	return d40_prep_sg(chan, src_sg, dst_sg, src_nents, DMA_NONE, dma_flags);
2569 }
2570 
2571 static struct dma_async_tx_descriptor *
2572 d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2573 		  unsigned int sg_len, enum dma_transfer_direction direction,
2574 		  unsigned long dma_flags, void *context)
2575 {
2576 	if (!is_slave_direction(direction))
2577 		return NULL;
2578 
2579 	return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2580 }
2581 
2582 static struct dma_async_tx_descriptor *
2583 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2584 		     size_t buf_len, size_t period_len,
2585 		     enum dma_transfer_direction direction, unsigned long flags,
2586 		     void *context)
2587 {
2588 	unsigned int periods = buf_len / period_len;
2589 	struct dma_async_tx_descriptor *txd;
2590 	struct scatterlist *sg;
2591 	int i;
2592 
2593 	sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2594 	if (!sg)
2595 		return NULL;
2596 
2597 	for (i = 0; i < periods; i++) {
2598 		sg_dma_address(&sg[i]) = dma_addr;
2599 		sg_dma_len(&sg[i]) = period_len;
2600 		dma_addr += period_len;
2601 	}
2602 
2603 	sg[periods].offset = 0;
2604 	sg_dma_len(&sg[periods]) = 0;
2605 	sg[periods].page_link =
2606 		((unsigned long)sg | 0x01) & ~0x02;
2607 
2608 	txd = d40_prep_sg(chan, sg, sg, periods, direction,
2609 			  DMA_PREP_INTERRUPT);
2610 
2611 	kfree(sg);
2612 
2613 	return txd;
2614 }
2615 
2616 static enum dma_status d40_tx_status(struct dma_chan *chan,
2617 				     dma_cookie_t cookie,
2618 				     struct dma_tx_state *txstate)
2619 {
2620 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2621 	enum dma_status ret;
2622 
2623 	if (d40c->phy_chan == NULL) {
2624 		chan_err(d40c, "Cannot read status of unallocated channel\n");
2625 		return -EINVAL;
2626 	}
2627 
2628 	ret = dma_cookie_status(chan, cookie, txstate);
2629 	if (ret != DMA_SUCCESS)
2630 		dma_set_residue(txstate, stedma40_residue(chan));
2631 
2632 	if (d40_is_paused(d40c))
2633 		ret = DMA_PAUSED;
2634 
2635 	return ret;
2636 }
2637 
2638 static void d40_issue_pending(struct dma_chan *chan)
2639 {
2640 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2641 	unsigned long flags;
2642 
2643 	if (d40c->phy_chan == NULL) {
2644 		chan_err(d40c, "Channel is not allocated!\n");
2645 		return;
2646 	}
2647 
2648 	spin_lock_irqsave(&d40c->lock, flags);
2649 
2650 	list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2651 
2652 	/* Busy means that queued jobs are already being processed */
2653 	if (!d40c->busy)
2654 		(void) d40_queue_start(d40c);
2655 
2656 	spin_unlock_irqrestore(&d40c->lock, flags);
2657 }
2658 
2659 static void d40_terminate_all(struct dma_chan *chan)
2660 {
2661 	unsigned long flags;
2662 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2663 	int ret;
2664 
2665 	spin_lock_irqsave(&d40c->lock, flags);
2666 
2667 	pm_runtime_get_sync(d40c->base->dev);
2668 	ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2669 	if (ret)
2670 		chan_err(d40c, "Failed to stop channel\n");
2671 
2672 	d40_term_all(d40c);
2673 	pm_runtime_mark_last_busy(d40c->base->dev);
2674 	pm_runtime_put_autosuspend(d40c->base->dev);
2675 	if (d40c->busy) {
2676 		pm_runtime_mark_last_busy(d40c->base->dev);
2677 		pm_runtime_put_autosuspend(d40c->base->dev);
2678 	}
2679 	d40c->busy = false;
2680 
2681 	spin_unlock_irqrestore(&d40c->lock, flags);
2682 }
2683 
2684 static int
2685 dma40_config_to_halfchannel(struct d40_chan *d40c,
2686 			    struct stedma40_half_channel_info *info,
2687 			    u32 maxburst)
2688 {
2689 	int psize;
2690 
2691 	if (chan_is_logical(d40c)) {
2692 		if (maxburst >= 16)
2693 			psize = STEDMA40_PSIZE_LOG_16;
2694 		else if (maxburst >= 8)
2695 			psize = STEDMA40_PSIZE_LOG_8;
2696 		else if (maxburst >= 4)
2697 			psize = STEDMA40_PSIZE_LOG_4;
2698 		else
2699 			psize = STEDMA40_PSIZE_LOG_1;
2700 	} else {
2701 		if (maxburst >= 16)
2702 			psize = STEDMA40_PSIZE_PHY_16;
2703 		else if (maxburst >= 8)
2704 			psize = STEDMA40_PSIZE_PHY_8;
2705 		else if (maxburst >= 4)
2706 			psize = STEDMA40_PSIZE_PHY_4;
2707 		else
2708 			psize = STEDMA40_PSIZE_PHY_1;
2709 	}
2710 
2711 	info->psize = psize;
2712 	info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2713 
2714 	return 0;
2715 }
2716 
2717 /* Runtime reconfiguration extension */
2718 static int d40_set_runtime_config(struct dma_chan *chan,
2719 				  struct dma_slave_config *config)
2720 {
2721 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2722 	struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2723 	enum dma_slave_buswidth src_addr_width, dst_addr_width;
2724 	dma_addr_t config_addr;
2725 	u32 src_maxburst, dst_maxburst;
2726 	int ret;
2727 
2728 	src_addr_width = config->src_addr_width;
2729 	src_maxburst = config->src_maxburst;
2730 	dst_addr_width = config->dst_addr_width;
2731 	dst_maxburst = config->dst_maxburst;
2732 
2733 	if (config->direction == DMA_DEV_TO_MEM) {
2734 		config_addr = config->src_addr;
2735 
2736 		if (cfg->dir != DMA_DEV_TO_MEM)
2737 			dev_dbg(d40c->base->dev,
2738 				"channel was not configured for peripheral "
2739 				"to memory transfer (%d) overriding\n",
2740 				cfg->dir);
2741 		cfg->dir = DMA_DEV_TO_MEM;
2742 
2743 		/* Configure the memory side */
2744 		if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2745 			dst_addr_width = src_addr_width;
2746 		if (dst_maxburst == 0)
2747 			dst_maxburst = src_maxburst;
2748 
2749 	} else if (config->direction == DMA_MEM_TO_DEV) {
2750 		config_addr = config->dst_addr;
2751 
2752 		if (cfg->dir != DMA_MEM_TO_DEV)
2753 			dev_dbg(d40c->base->dev,
2754 				"channel was not configured for memory "
2755 				"to peripheral transfer (%d) overriding\n",
2756 				cfg->dir);
2757 		cfg->dir = DMA_MEM_TO_DEV;
2758 
2759 		/* Configure the memory side */
2760 		if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2761 			src_addr_width = dst_addr_width;
2762 		if (src_maxburst == 0)
2763 			src_maxburst = dst_maxburst;
2764 	} else {
2765 		dev_err(d40c->base->dev,
2766 			"unrecognized channel direction %d\n",
2767 			config->direction);
2768 		return -EINVAL;
2769 	}
2770 
2771 	if (config_addr <= 0) {
2772 		dev_err(d40c->base->dev, "no address supplied\n");
2773 		return -EINVAL;
2774 	}
2775 
2776 	if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2777 		dev_err(d40c->base->dev,
2778 			"src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2779 			src_maxburst,
2780 			src_addr_width,
2781 			dst_maxburst,
2782 			dst_addr_width);
2783 		return -EINVAL;
2784 	}
2785 
2786 	if (src_maxburst > 16) {
2787 		src_maxburst = 16;
2788 		dst_maxburst = src_maxburst * src_addr_width / dst_addr_width;
2789 	} else if (dst_maxburst > 16) {
2790 		dst_maxburst = 16;
2791 		src_maxburst = dst_maxburst * dst_addr_width / src_addr_width;
2792 	}
2793 
2794 	/* Only valid widths are; 1, 2, 4 and 8. */
2795 	if (src_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2796 	    src_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2797 	    dst_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2798 	    dst_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2799 	    ((src_addr_width > 1) && (src_addr_width & 1)) ||
2800 	    ((dst_addr_width > 1) && (dst_addr_width & 1)))
2801 		return -EINVAL;
2802 
2803 	cfg->src_info.data_width = src_addr_width;
2804 	cfg->dst_info.data_width = dst_addr_width;
2805 
2806 	ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2807 					  src_maxburst);
2808 	if (ret)
2809 		return ret;
2810 
2811 	ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2812 					  dst_maxburst);
2813 	if (ret)
2814 		return ret;
2815 
2816 	/* Fill in register values */
2817 	if (chan_is_logical(d40c))
2818 		d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2819 	else
2820 		d40_phy_cfg(cfg, &d40c->src_def_cfg, &d40c->dst_def_cfg);
2821 
2822 	/* These settings will take precedence later */
2823 	d40c->runtime_addr = config_addr;
2824 	d40c->runtime_direction = config->direction;
2825 	dev_dbg(d40c->base->dev,
2826 		"configured channel %s for %s, data width %d/%d, "
2827 		"maxburst %d/%d elements, LE, no flow control\n",
2828 		dma_chan_name(chan),
2829 		(config->direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2830 		src_addr_width, dst_addr_width,
2831 		src_maxburst, dst_maxburst);
2832 
2833 	return 0;
2834 }
2835 
2836 static int d40_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
2837 		       unsigned long arg)
2838 {
2839 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2840 
2841 	if (d40c->phy_chan == NULL) {
2842 		chan_err(d40c, "Channel is not allocated!\n");
2843 		return -EINVAL;
2844 	}
2845 
2846 	switch (cmd) {
2847 	case DMA_TERMINATE_ALL:
2848 		d40_terminate_all(chan);
2849 		return 0;
2850 	case DMA_PAUSE:
2851 		return d40_pause(d40c);
2852 	case DMA_RESUME:
2853 		return d40_resume(d40c);
2854 	case DMA_SLAVE_CONFIG:
2855 		return d40_set_runtime_config(chan,
2856 			(struct dma_slave_config *) arg);
2857 	default:
2858 		break;
2859 	}
2860 
2861 	/* Other commands are unimplemented */
2862 	return -ENXIO;
2863 }
2864 
2865 /* Initialization functions */
2866 
2867 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2868 				 struct d40_chan *chans, int offset,
2869 				 int num_chans)
2870 {
2871 	int i = 0;
2872 	struct d40_chan *d40c;
2873 
2874 	INIT_LIST_HEAD(&dma->channels);
2875 
2876 	for (i = offset; i < offset + num_chans; i++) {
2877 		d40c = &chans[i];
2878 		d40c->base = base;
2879 		d40c->chan.device = dma;
2880 
2881 		spin_lock_init(&d40c->lock);
2882 
2883 		d40c->log_num = D40_PHY_CHAN;
2884 
2885 		INIT_LIST_HEAD(&d40c->done);
2886 		INIT_LIST_HEAD(&d40c->active);
2887 		INIT_LIST_HEAD(&d40c->queue);
2888 		INIT_LIST_HEAD(&d40c->pending_queue);
2889 		INIT_LIST_HEAD(&d40c->client);
2890 		INIT_LIST_HEAD(&d40c->prepare_queue);
2891 
2892 		tasklet_init(&d40c->tasklet, dma_tasklet,
2893 			     (unsigned long) d40c);
2894 
2895 		list_add_tail(&d40c->chan.device_node,
2896 			      &dma->channels);
2897 	}
2898 }
2899 
2900 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2901 {
2902 	if (dma_has_cap(DMA_SLAVE, dev->cap_mask))
2903 		dev->device_prep_slave_sg = d40_prep_slave_sg;
2904 
2905 	if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2906 		dev->device_prep_dma_memcpy = d40_prep_memcpy;
2907 
2908 		/*
2909 		 * This controller can only access address at even
2910 		 * 32bit boundaries, i.e. 2^2
2911 		 */
2912 		dev->copy_align = 2;
2913 	}
2914 
2915 	if (dma_has_cap(DMA_SG, dev->cap_mask))
2916 		dev->device_prep_dma_sg = d40_prep_memcpy_sg;
2917 
2918 	if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2919 		dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2920 
2921 	dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2922 	dev->device_free_chan_resources = d40_free_chan_resources;
2923 	dev->device_issue_pending = d40_issue_pending;
2924 	dev->device_tx_status = d40_tx_status;
2925 	dev->device_control = d40_control;
2926 	dev->dev = base->dev;
2927 }
2928 
2929 static int __init d40_dmaengine_init(struct d40_base *base,
2930 				     int num_reserved_chans)
2931 {
2932 	int err ;
2933 
2934 	d40_chan_init(base, &base->dma_slave, base->log_chans,
2935 		      0, base->num_log_chans);
2936 
2937 	dma_cap_zero(base->dma_slave.cap_mask);
2938 	dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2939 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2940 
2941 	d40_ops_init(base, &base->dma_slave);
2942 
2943 	err = dma_async_device_register(&base->dma_slave);
2944 
2945 	if (err) {
2946 		d40_err(base->dev, "Failed to register slave channels\n");
2947 		goto failure1;
2948 	}
2949 
2950 	d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2951 		      base->num_log_chans, base->num_memcpy_chans);
2952 
2953 	dma_cap_zero(base->dma_memcpy.cap_mask);
2954 	dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2955 	dma_cap_set(DMA_SG, base->dma_memcpy.cap_mask);
2956 
2957 	d40_ops_init(base, &base->dma_memcpy);
2958 
2959 	err = dma_async_device_register(&base->dma_memcpy);
2960 
2961 	if (err) {
2962 		d40_err(base->dev,
2963 			"Failed to regsiter memcpy only channels\n");
2964 		goto failure2;
2965 	}
2966 
2967 	d40_chan_init(base, &base->dma_both, base->phy_chans,
2968 		      0, num_reserved_chans);
2969 
2970 	dma_cap_zero(base->dma_both.cap_mask);
2971 	dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2972 	dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2973 	dma_cap_set(DMA_SG, base->dma_both.cap_mask);
2974 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2975 
2976 	d40_ops_init(base, &base->dma_both);
2977 	err = dma_async_device_register(&base->dma_both);
2978 
2979 	if (err) {
2980 		d40_err(base->dev,
2981 			"Failed to register logical and physical capable channels\n");
2982 		goto failure3;
2983 	}
2984 	return 0;
2985 failure3:
2986 	dma_async_device_unregister(&base->dma_memcpy);
2987 failure2:
2988 	dma_async_device_unregister(&base->dma_slave);
2989 failure1:
2990 	return err;
2991 }
2992 
2993 /* Suspend resume functionality */
2994 #ifdef CONFIG_PM
2995 static int dma40_pm_suspend(struct device *dev)
2996 {
2997 	struct platform_device *pdev = to_platform_device(dev);
2998 	struct d40_base *base = platform_get_drvdata(pdev);
2999 	int ret = 0;
3000 
3001 	if (base->lcpa_regulator)
3002 		ret = regulator_disable(base->lcpa_regulator);
3003 	return ret;
3004 }
3005 
3006 static int dma40_runtime_suspend(struct device *dev)
3007 {
3008 	struct platform_device *pdev = to_platform_device(dev);
3009 	struct d40_base *base = platform_get_drvdata(pdev);
3010 
3011 	d40_save_restore_registers(base, true);
3012 
3013 	/* Don't disable/enable clocks for v1 due to HW bugs */
3014 	if (base->rev != 1)
3015 		writel_relaxed(base->gcc_pwr_off_mask,
3016 			       base->virtbase + D40_DREG_GCC);
3017 
3018 	return 0;
3019 }
3020 
3021 static int dma40_runtime_resume(struct device *dev)
3022 {
3023 	struct platform_device *pdev = to_platform_device(dev);
3024 	struct d40_base *base = platform_get_drvdata(pdev);
3025 
3026 	if (base->initialized)
3027 		d40_save_restore_registers(base, false);
3028 
3029 	writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
3030 		       base->virtbase + D40_DREG_GCC);
3031 	return 0;
3032 }
3033 
3034 static int dma40_resume(struct device *dev)
3035 {
3036 	struct platform_device *pdev = to_platform_device(dev);
3037 	struct d40_base *base = platform_get_drvdata(pdev);
3038 	int ret = 0;
3039 
3040 	if (base->lcpa_regulator)
3041 		ret = regulator_enable(base->lcpa_regulator);
3042 
3043 	return ret;
3044 }
3045 
3046 static const struct dev_pm_ops dma40_pm_ops = {
3047 	.suspend		= dma40_pm_suspend,
3048 	.runtime_suspend	= dma40_runtime_suspend,
3049 	.runtime_resume		= dma40_runtime_resume,
3050 	.resume			= dma40_resume,
3051 };
3052 #define DMA40_PM_OPS	(&dma40_pm_ops)
3053 #else
3054 #define DMA40_PM_OPS	NULL
3055 #endif
3056 
3057 /* Initialization functions. */
3058 
3059 static int __init d40_phy_res_init(struct d40_base *base)
3060 {
3061 	int i;
3062 	int num_phy_chans_avail = 0;
3063 	u32 val[2];
3064 	int odd_even_bit = -2;
3065 	int gcc = D40_DREG_GCC_ENA;
3066 
3067 	val[0] = readl(base->virtbase + D40_DREG_PRSME);
3068 	val[1] = readl(base->virtbase + D40_DREG_PRSMO);
3069 
3070 	for (i = 0; i < base->num_phy_chans; i++) {
3071 		base->phy_res[i].num = i;
3072 		odd_even_bit += 2 * ((i % 2) == 0);
3073 		if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
3074 			/* Mark security only channels as occupied */
3075 			base->phy_res[i].allocated_src = D40_ALLOC_PHY;
3076 			base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
3077 			base->phy_res[i].reserved = true;
3078 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3079 						       D40_DREG_GCC_SRC);
3080 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3081 						       D40_DREG_GCC_DST);
3082 
3083 
3084 		} else {
3085 			base->phy_res[i].allocated_src = D40_ALLOC_FREE;
3086 			base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
3087 			base->phy_res[i].reserved = false;
3088 			num_phy_chans_avail++;
3089 		}
3090 		spin_lock_init(&base->phy_res[i].lock);
3091 	}
3092 
3093 	/* Mark disabled channels as occupied */
3094 	for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
3095 		int chan = base->plat_data->disabled_channels[i];
3096 
3097 		base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
3098 		base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
3099 		base->phy_res[chan].reserved = true;
3100 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3101 					       D40_DREG_GCC_SRC);
3102 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3103 					       D40_DREG_GCC_DST);
3104 		num_phy_chans_avail--;
3105 	}
3106 
3107 	/* Mark soft_lli channels */
3108 	for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) {
3109 		int chan = base->plat_data->soft_lli_chans[i];
3110 
3111 		base->phy_res[chan].use_soft_lli = true;
3112 	}
3113 
3114 	dev_info(base->dev, "%d of %d physical DMA channels available\n",
3115 		 num_phy_chans_avail, base->num_phy_chans);
3116 
3117 	/* Verify settings extended vs standard */
3118 	val[0] = readl(base->virtbase + D40_DREG_PRTYP);
3119 
3120 	for (i = 0; i < base->num_phy_chans; i++) {
3121 
3122 		if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
3123 		    (val[0] & 0x3) != 1)
3124 			dev_info(base->dev,
3125 				 "[%s] INFO: channel %d is misconfigured (%d)\n",
3126 				 __func__, i, val[0] & 0x3);
3127 
3128 		val[0] = val[0] >> 2;
3129 	}
3130 
3131 	/*
3132 	 * To keep things simple, Enable all clocks initially.
3133 	 * The clocks will get managed later post channel allocation.
3134 	 * The clocks for the event lines on which reserved channels exists
3135 	 * are not managed here.
3136 	 */
3137 	writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3138 	base->gcc_pwr_off_mask = gcc;
3139 
3140 	return num_phy_chans_avail;
3141 }
3142 
3143 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
3144 {
3145 	struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3146 	struct clk *clk = NULL;
3147 	void __iomem *virtbase = NULL;
3148 	struct resource *res = NULL;
3149 	struct d40_base *base = NULL;
3150 	int num_log_chans = 0;
3151 	int num_phy_chans;
3152 	int num_memcpy_chans;
3153 	int clk_ret = -EINVAL;
3154 	int i;
3155 	u32 pid;
3156 	u32 cid;
3157 	u8 rev;
3158 
3159 	clk = clk_get(&pdev->dev, NULL);
3160 	if (IS_ERR(clk)) {
3161 		d40_err(&pdev->dev, "No matching clock found\n");
3162 		goto failure;
3163 	}
3164 
3165 	clk_ret = clk_prepare_enable(clk);
3166 	if (clk_ret) {
3167 		d40_err(&pdev->dev, "Failed to prepare/enable clock\n");
3168 		goto failure;
3169 	}
3170 
3171 	/* Get IO for DMAC base address */
3172 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
3173 	if (!res)
3174 		goto failure;
3175 
3176 	if (request_mem_region(res->start, resource_size(res),
3177 			       D40_NAME " I/O base") == NULL)
3178 		goto failure;
3179 
3180 	virtbase = ioremap(res->start, resource_size(res));
3181 	if (!virtbase)
3182 		goto failure;
3183 
3184 	/* This is just a regular AMBA PrimeCell ID actually */
3185 	for (pid = 0, i = 0; i < 4; i++)
3186 		pid |= (readl(virtbase + resource_size(res) - 0x20 + 4 * i)
3187 			& 255) << (i * 8);
3188 	for (cid = 0, i = 0; i < 4; i++)
3189 		cid |= (readl(virtbase + resource_size(res) - 0x10 + 4 * i)
3190 			& 255) << (i * 8);
3191 
3192 	if (cid != AMBA_CID) {
3193 		d40_err(&pdev->dev, "Unknown hardware! No PrimeCell ID\n");
3194 		goto failure;
3195 	}
3196 	if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
3197 		d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
3198 			AMBA_MANF_BITS(pid),
3199 			AMBA_VENDOR_ST);
3200 		goto failure;
3201 	}
3202 	/*
3203 	 * HW revision:
3204 	 * DB8500ed has revision 0
3205 	 * ? has revision 1
3206 	 * DB8500v1 has revision 2
3207 	 * DB8500v2 has revision 3
3208 	 * AP9540v1 has revision 4
3209 	 * DB8540v1 has revision 4
3210 	 */
3211 	rev = AMBA_REV_BITS(pid);
3212 	if (rev < 2) {
3213 		d40_err(&pdev->dev, "hardware revision: %d is not supported", rev);
3214 		goto failure;
3215 	}
3216 
3217 	/* The number of physical channels on this HW */
3218 	if (plat_data->num_of_phy_chans)
3219 		num_phy_chans = plat_data->num_of_phy_chans;
3220 	else
3221 		num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
3222 
3223 	/* The number of channels used for memcpy */
3224 	if (plat_data->num_of_memcpy_chans)
3225 		num_memcpy_chans = plat_data->num_of_memcpy_chans;
3226 	else
3227 		num_memcpy_chans = ARRAY_SIZE(dma40_memcpy_channels);
3228 
3229 	num_log_chans = num_phy_chans * D40_MAX_LOG_CHAN_PER_PHY;
3230 
3231 	dev_info(&pdev->dev,
3232 		 "hardware rev: %d @ %pa with %d physical and %d logical channels\n",
3233 		 rev, &res->start, num_phy_chans, num_log_chans);
3234 
3235 	base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
3236 		       (num_phy_chans + num_log_chans + num_memcpy_chans) *
3237 		       sizeof(struct d40_chan), GFP_KERNEL);
3238 
3239 	if (base == NULL) {
3240 		d40_err(&pdev->dev, "Out of memory\n");
3241 		goto failure;
3242 	}
3243 
3244 	base->rev = rev;
3245 	base->clk = clk;
3246 	base->num_memcpy_chans = num_memcpy_chans;
3247 	base->num_phy_chans = num_phy_chans;
3248 	base->num_log_chans = num_log_chans;
3249 	base->phy_start = res->start;
3250 	base->phy_size = resource_size(res);
3251 	base->virtbase = virtbase;
3252 	base->plat_data = plat_data;
3253 	base->dev = &pdev->dev;
3254 	base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3255 	base->log_chans = &base->phy_chans[num_phy_chans];
3256 
3257 	if (base->plat_data->num_of_phy_chans == 14) {
3258 		base->gen_dmac.backup = d40_backup_regs_v4b;
3259 		base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
3260 		base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
3261 		base->gen_dmac.interrupt_clear = D40_DREG_CPCICR;
3262 		base->gen_dmac.realtime_en = D40_DREG_CRSEG1;
3263 		base->gen_dmac.realtime_clear = D40_DREG_CRCEG1;
3264 		base->gen_dmac.high_prio_en = D40_DREG_CPSEG1;
3265 		base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1;
3266 		base->gen_dmac.il = il_v4b;
3267 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4b);
3268 		base->gen_dmac.init_reg = dma_init_reg_v4b;
3269 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
3270 	} else {
3271 		if (base->rev >= 3) {
3272 			base->gen_dmac.backup = d40_backup_regs_v4a;
3273 			base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
3274 		}
3275 		base->gen_dmac.interrupt_en = D40_DREG_PCMIS;
3276 		base->gen_dmac.interrupt_clear = D40_DREG_PCICR;
3277 		base->gen_dmac.realtime_en = D40_DREG_RSEG1;
3278 		base->gen_dmac.realtime_clear = D40_DREG_RCEG1;
3279 		base->gen_dmac.high_prio_en = D40_DREG_PSEG1;
3280 		base->gen_dmac.high_prio_clear = D40_DREG_PCEG1;
3281 		base->gen_dmac.il = il_v4a;
3282 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4a);
3283 		base->gen_dmac.init_reg = dma_init_reg_v4a;
3284 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a);
3285 	}
3286 
3287 	base->phy_res = kzalloc(num_phy_chans * sizeof(struct d40_phy_res),
3288 				GFP_KERNEL);
3289 	if (!base->phy_res)
3290 		goto failure;
3291 
3292 	base->lookup_phy_chans = kzalloc(num_phy_chans *
3293 					 sizeof(struct d40_chan *),
3294 					 GFP_KERNEL);
3295 	if (!base->lookup_phy_chans)
3296 		goto failure;
3297 
3298 	base->lookup_log_chans = kzalloc(num_log_chans *
3299 					 sizeof(struct d40_chan *),
3300 					 GFP_KERNEL);
3301 	if (!base->lookup_log_chans)
3302 		goto failure;
3303 
3304 	base->reg_val_backup_chan = kmalloc(base->num_phy_chans *
3305 					    sizeof(d40_backup_regs_chan),
3306 					    GFP_KERNEL);
3307 	if (!base->reg_val_backup_chan)
3308 		goto failure;
3309 
3310 	base->lcla_pool.alloc_map =
3311 		kzalloc(num_phy_chans * sizeof(struct d40_desc *)
3312 			* D40_LCLA_LINK_PER_EVENT_GRP, GFP_KERNEL);
3313 	if (!base->lcla_pool.alloc_map)
3314 		goto failure;
3315 
3316 	base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3317 					    0, SLAB_HWCACHE_ALIGN,
3318 					    NULL);
3319 	if (base->desc_slab == NULL)
3320 		goto failure;
3321 
3322 	return base;
3323 
3324 failure:
3325 	if (!clk_ret)
3326 		clk_disable_unprepare(clk);
3327 	if (!IS_ERR(clk))
3328 		clk_put(clk);
3329 	if (virtbase)
3330 		iounmap(virtbase);
3331 	if (res)
3332 		release_mem_region(res->start,
3333 				   resource_size(res));
3334 	if (virtbase)
3335 		iounmap(virtbase);
3336 
3337 	if (base) {
3338 		kfree(base->lcla_pool.alloc_map);
3339 		kfree(base->reg_val_backup_chan);
3340 		kfree(base->lookup_log_chans);
3341 		kfree(base->lookup_phy_chans);
3342 		kfree(base->phy_res);
3343 		kfree(base);
3344 	}
3345 
3346 	return NULL;
3347 }
3348 
3349 static void __init d40_hw_init(struct d40_base *base)
3350 {
3351 
3352 	int i;
3353 	u32 prmseo[2] = {0, 0};
3354 	u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3355 	u32 pcmis = 0;
3356 	u32 pcicr = 0;
3357 	struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg;
3358 	u32 reg_size = base->gen_dmac.init_reg_size;
3359 
3360 	for (i = 0; i < reg_size; i++)
3361 		writel(dma_init_reg[i].val,
3362 		       base->virtbase + dma_init_reg[i].reg);
3363 
3364 	/* Configure all our dma channels to default settings */
3365 	for (i = 0; i < base->num_phy_chans; i++) {
3366 
3367 		activeo[i % 2] = activeo[i % 2] << 2;
3368 
3369 		if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3370 		    == D40_ALLOC_PHY) {
3371 			activeo[i % 2] |= 3;
3372 			continue;
3373 		}
3374 
3375 		/* Enable interrupt # */
3376 		pcmis = (pcmis << 1) | 1;
3377 
3378 		/* Clear interrupt # */
3379 		pcicr = (pcicr << 1) | 1;
3380 
3381 		/* Set channel to physical mode */
3382 		prmseo[i % 2] = prmseo[i % 2] << 2;
3383 		prmseo[i % 2] |= 1;
3384 
3385 	}
3386 
3387 	writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3388 	writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3389 	writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3390 	writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3391 
3392 	/* Write which interrupt to enable */
3393 	writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en);
3394 
3395 	/* Write which interrupt to clear */
3396 	writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear);
3397 
3398 	/* These are __initdata and cannot be accessed after init */
3399 	base->gen_dmac.init_reg = NULL;
3400 	base->gen_dmac.init_reg_size = 0;
3401 }
3402 
3403 static int __init d40_lcla_allocate(struct d40_base *base)
3404 {
3405 	struct d40_lcla_pool *pool = &base->lcla_pool;
3406 	unsigned long *page_list;
3407 	int i, j;
3408 	int ret = 0;
3409 
3410 	/*
3411 	 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3412 	 * To full fill this hardware requirement without wasting 256 kb
3413 	 * we allocate pages until we get an aligned one.
3414 	 */
3415 	page_list = kmalloc(sizeof(unsigned long) * MAX_LCLA_ALLOC_ATTEMPTS,
3416 			    GFP_KERNEL);
3417 
3418 	if (!page_list) {
3419 		ret = -ENOMEM;
3420 		goto failure;
3421 	}
3422 
3423 	/* Calculating how many pages that are required */
3424 	base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3425 
3426 	for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3427 		page_list[i] = __get_free_pages(GFP_KERNEL,
3428 						base->lcla_pool.pages);
3429 		if (!page_list[i]) {
3430 
3431 			d40_err(base->dev, "Failed to allocate %d pages.\n",
3432 				base->lcla_pool.pages);
3433 
3434 			for (j = 0; j < i; j++)
3435 				free_pages(page_list[j], base->lcla_pool.pages);
3436 			goto failure;
3437 		}
3438 
3439 		if ((virt_to_phys((void *)page_list[i]) &
3440 		     (LCLA_ALIGNMENT - 1)) == 0)
3441 			break;
3442 	}
3443 
3444 	for (j = 0; j < i; j++)
3445 		free_pages(page_list[j], base->lcla_pool.pages);
3446 
3447 	if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3448 		base->lcla_pool.base = (void *)page_list[i];
3449 	} else {
3450 		/*
3451 		 * After many attempts and no succees with finding the correct
3452 		 * alignment, try with allocating a big buffer.
3453 		 */
3454 		dev_warn(base->dev,
3455 			 "[%s] Failed to get %d pages @ 18 bit align.\n",
3456 			 __func__, base->lcla_pool.pages);
3457 		base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3458 							 base->num_phy_chans +
3459 							 LCLA_ALIGNMENT,
3460 							 GFP_KERNEL);
3461 		if (!base->lcla_pool.base_unaligned) {
3462 			ret = -ENOMEM;
3463 			goto failure;
3464 		}
3465 
3466 		base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3467 						 LCLA_ALIGNMENT);
3468 	}
3469 
3470 	pool->dma_addr = dma_map_single(base->dev, pool->base,
3471 					SZ_1K * base->num_phy_chans,
3472 					DMA_TO_DEVICE);
3473 	if (dma_mapping_error(base->dev, pool->dma_addr)) {
3474 		pool->dma_addr = 0;
3475 		ret = -ENOMEM;
3476 		goto failure;
3477 	}
3478 
3479 	writel(virt_to_phys(base->lcla_pool.base),
3480 	       base->virtbase + D40_DREG_LCLA);
3481 failure:
3482 	kfree(page_list);
3483 	return ret;
3484 }
3485 
3486 static int __init d40_of_probe(struct platform_device *pdev,
3487 			       struct device_node *np)
3488 {
3489 	struct stedma40_platform_data *pdata;
3490 	int num_phy = 0, num_memcpy = 0, num_disabled = 0;
3491 	const __be32 *list;
3492 
3493 	pdata = devm_kzalloc(&pdev->dev,
3494 			     sizeof(struct stedma40_platform_data),
3495 			     GFP_KERNEL);
3496 	if (!pdata)
3497 		return -ENOMEM;
3498 
3499 	/* If absent this value will be obtained from h/w. */
3500 	of_property_read_u32(np, "dma-channels", &num_phy);
3501 	if (num_phy > 0)
3502 		pdata->num_of_phy_chans = num_phy;
3503 
3504 	list = of_get_property(np, "memcpy-channels", &num_memcpy);
3505 	num_memcpy /= sizeof(*list);
3506 
3507 	if (num_memcpy > D40_MEMCPY_MAX_CHANS || num_memcpy <= 0) {
3508 		d40_err(&pdev->dev,
3509 			"Invalid number of memcpy channels specified (%d)\n",
3510 			num_memcpy);
3511 		return -EINVAL;
3512 	}
3513 	pdata->num_of_memcpy_chans = num_memcpy;
3514 
3515 	of_property_read_u32_array(np, "memcpy-channels",
3516 				   dma40_memcpy_channels,
3517 				   num_memcpy);
3518 
3519 	list = of_get_property(np, "disabled-channels", &num_disabled);
3520 	num_disabled /= sizeof(*list);
3521 
3522 	if (num_disabled >= STEDMA40_MAX_PHYS || num_disabled < 0) {
3523 		d40_err(&pdev->dev,
3524 			"Invalid number of disabled channels specified (%d)\n",
3525 			num_disabled);
3526 		return -EINVAL;
3527 	}
3528 
3529 	of_property_read_u32_array(np, "disabled-channels",
3530 				   pdata->disabled_channels,
3531 				   num_disabled);
3532 	pdata->disabled_channels[num_disabled] = -1;
3533 
3534 	pdev->dev.platform_data = pdata;
3535 
3536 	return 0;
3537 }
3538 
3539 static int __init d40_probe(struct platform_device *pdev)
3540 {
3541 	struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3542 	struct device_node *np = pdev->dev.of_node;
3543 	int ret = -ENOENT;
3544 	struct d40_base *base = NULL;
3545 	struct resource *res = NULL;
3546 	int num_reserved_chans;
3547 	u32 val;
3548 
3549 	if (!plat_data) {
3550 		if (np) {
3551 			if(d40_of_probe(pdev, np)) {
3552 				ret = -ENOMEM;
3553 				goto failure;
3554 			}
3555 		} else {
3556 			d40_err(&pdev->dev, "No pdata or Device Tree provided\n");
3557 			goto failure;
3558 		}
3559 	}
3560 
3561 	base = d40_hw_detect_init(pdev);
3562 	if (!base)
3563 		goto failure;
3564 
3565 	num_reserved_chans = d40_phy_res_init(base);
3566 
3567 	platform_set_drvdata(pdev, base);
3568 
3569 	spin_lock_init(&base->interrupt_lock);
3570 	spin_lock_init(&base->execmd_lock);
3571 
3572 	/* Get IO for logical channel parameter address */
3573 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
3574 	if (!res) {
3575 		ret = -ENOENT;
3576 		d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
3577 		goto failure;
3578 	}
3579 	base->lcpa_size = resource_size(res);
3580 	base->phy_lcpa = res->start;
3581 
3582 	if (request_mem_region(res->start, resource_size(res),
3583 			       D40_NAME " I/O lcpa") == NULL) {
3584 		ret = -EBUSY;
3585 		d40_err(&pdev->dev, "Failed to request LCPA region %pR\n", res);
3586 		goto failure;
3587 	}
3588 
3589 	/* We make use of ESRAM memory for this. */
3590 	val = readl(base->virtbase + D40_DREG_LCPA);
3591 	if (res->start != val && val != 0) {
3592 		dev_warn(&pdev->dev,
3593 			 "[%s] Mismatch LCPA dma 0x%x, def %pa\n",
3594 			 __func__, val, &res->start);
3595 	} else
3596 		writel(res->start, base->virtbase + D40_DREG_LCPA);
3597 
3598 	base->lcpa_base = ioremap(res->start, resource_size(res));
3599 	if (!base->lcpa_base) {
3600 		ret = -ENOMEM;
3601 		d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
3602 		goto failure;
3603 	}
3604 	/* If lcla has to be located in ESRAM we don't need to allocate */
3605 	if (base->plat_data->use_esram_lcla) {
3606 		res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3607 							"lcla_esram");
3608 		if (!res) {
3609 			ret = -ENOENT;
3610 			d40_err(&pdev->dev,
3611 				"No \"lcla_esram\" memory resource\n");
3612 			goto failure;
3613 		}
3614 		base->lcla_pool.base = ioremap(res->start,
3615 						resource_size(res));
3616 		if (!base->lcla_pool.base) {
3617 			ret = -ENOMEM;
3618 			d40_err(&pdev->dev, "Failed to ioremap LCLA region\n");
3619 			goto failure;
3620 		}
3621 		writel(res->start, base->virtbase + D40_DREG_LCLA);
3622 
3623 	} else {
3624 		ret = d40_lcla_allocate(base);
3625 		if (ret) {
3626 			d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
3627 			goto failure;
3628 		}
3629 	}
3630 
3631 	spin_lock_init(&base->lcla_pool.lock);
3632 
3633 	base->irq = platform_get_irq(pdev, 0);
3634 
3635 	ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3636 	if (ret) {
3637 		d40_err(&pdev->dev, "No IRQ defined\n");
3638 		goto failure;
3639 	}
3640 
3641 	pm_runtime_irq_safe(base->dev);
3642 	pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3643 	pm_runtime_use_autosuspend(base->dev);
3644 	pm_runtime_enable(base->dev);
3645 	pm_runtime_resume(base->dev);
3646 
3647 	if (base->plat_data->use_esram_lcla) {
3648 
3649 		base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3650 		if (IS_ERR(base->lcpa_regulator)) {
3651 			d40_err(&pdev->dev, "Failed to get lcpa_regulator\n");
3652 			ret = PTR_ERR(base->lcpa_regulator);
3653 			base->lcpa_regulator = NULL;
3654 			goto failure;
3655 		}
3656 
3657 		ret = regulator_enable(base->lcpa_regulator);
3658 		if (ret) {
3659 			d40_err(&pdev->dev,
3660 				"Failed to enable lcpa_regulator\n");
3661 			regulator_put(base->lcpa_regulator);
3662 			base->lcpa_regulator = NULL;
3663 			goto failure;
3664 		}
3665 	}
3666 
3667 	base->initialized = true;
3668 	ret = d40_dmaengine_init(base, num_reserved_chans);
3669 	if (ret)
3670 		goto failure;
3671 
3672 	base->dev->dma_parms = &base->dma_parms;
3673 	ret = dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
3674 	if (ret) {
3675 		d40_err(&pdev->dev, "Failed to set dma max seg size\n");
3676 		goto failure;
3677 	}
3678 
3679 	d40_hw_init(base);
3680 
3681 	if (np) {
3682 		ret = of_dma_controller_register(np, d40_xlate, NULL);
3683 		if (ret)
3684 			dev_err(&pdev->dev,
3685 				"could not register of_dma_controller\n");
3686 	}
3687 
3688 	dev_info(base->dev, "initialized\n");
3689 	return 0;
3690 
3691 failure:
3692 	if (base) {
3693 		if (base->desc_slab)
3694 			kmem_cache_destroy(base->desc_slab);
3695 		if (base->virtbase)
3696 			iounmap(base->virtbase);
3697 
3698 		if (base->lcla_pool.base && base->plat_data->use_esram_lcla) {
3699 			iounmap(base->lcla_pool.base);
3700 			base->lcla_pool.base = NULL;
3701 		}
3702 
3703 		if (base->lcla_pool.dma_addr)
3704 			dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3705 					 SZ_1K * base->num_phy_chans,
3706 					 DMA_TO_DEVICE);
3707 
3708 		if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3709 			free_pages((unsigned long)base->lcla_pool.base,
3710 				   base->lcla_pool.pages);
3711 
3712 		kfree(base->lcla_pool.base_unaligned);
3713 
3714 		if (base->phy_lcpa)
3715 			release_mem_region(base->phy_lcpa,
3716 					   base->lcpa_size);
3717 		if (base->phy_start)
3718 			release_mem_region(base->phy_start,
3719 					   base->phy_size);
3720 		if (base->clk) {
3721 			clk_disable_unprepare(base->clk);
3722 			clk_put(base->clk);
3723 		}
3724 
3725 		if (base->lcpa_regulator) {
3726 			regulator_disable(base->lcpa_regulator);
3727 			regulator_put(base->lcpa_regulator);
3728 		}
3729 
3730 		kfree(base->lcla_pool.alloc_map);
3731 		kfree(base->lookup_log_chans);
3732 		kfree(base->lookup_phy_chans);
3733 		kfree(base->phy_res);
3734 		kfree(base);
3735 	}
3736 
3737 	d40_err(&pdev->dev, "probe failed\n");
3738 	return ret;
3739 }
3740 
3741 static const struct of_device_id d40_match[] = {
3742         { .compatible = "stericsson,dma40", },
3743         {}
3744 };
3745 
3746 static struct platform_driver d40_driver = {
3747 	.driver = {
3748 		.owner = THIS_MODULE,
3749 		.name  = D40_NAME,
3750 		.pm = DMA40_PM_OPS,
3751 		.of_match_table = d40_match,
3752 	},
3753 };
3754 
3755 static int __init stedma40_init(void)
3756 {
3757 	return platform_driver_probe(&d40_driver, d40_probe);
3758 }
3759 subsys_initcall(stedma40_init);
3760