xref: /openbmc/linux/drivers/dma/ste_dma40.c (revision 275876e2)
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/log2.h>
18 #include <linux/pm.h>
19 #include <linux/pm_runtime.h>
20 #include <linux/err.h>
21 #include <linux/of.h>
22 #include <linux/of_dma.h>
23 #include <linux/amba/bus.h>
24 #include <linux/regulator/consumer.h>
25 #include <linux/platform_data/dma-ste-dma40.h>
26 
27 #include "dmaengine.h"
28 #include "ste_dma40_ll.h"
29 
30 #define D40_NAME "dma40"
31 
32 #define D40_PHY_CHAN -1
33 
34 /* For masking out/in 2 bit channel positions */
35 #define D40_CHAN_POS(chan)  (2 * (chan / 2))
36 #define D40_CHAN_POS_MASK(chan) (0x3 << D40_CHAN_POS(chan))
37 
38 /* Maximum iterations taken before giving up suspending a channel */
39 #define D40_SUSPEND_MAX_IT 500
40 
41 /* Milliseconds */
42 #define DMA40_AUTOSUSPEND_DELAY	100
43 
44 /* Hardware requirement on LCLA alignment */
45 #define LCLA_ALIGNMENT 0x40000
46 
47 /* Max number of links per event group */
48 #define D40_LCLA_LINK_PER_EVENT_GRP 128
49 #define D40_LCLA_END D40_LCLA_LINK_PER_EVENT_GRP
50 
51 /* Max number of logical channels per physical channel */
52 #define D40_MAX_LOG_CHAN_PER_PHY 32
53 
54 /* Attempts before giving up to trying to get pages that are aligned */
55 #define MAX_LCLA_ALLOC_ATTEMPTS 256
56 
57 /* Bit markings for allocation map */
58 #define D40_ALLOC_FREE		BIT(31)
59 #define D40_ALLOC_PHY		BIT(30)
60 #define D40_ALLOC_LOG_FREE	0
61 
62 #define D40_MEMCPY_MAX_CHANS	8
63 
64 /* Reserved event lines for memcpy only. */
65 #define DB8500_DMA_MEMCPY_EV_0	51
66 #define DB8500_DMA_MEMCPY_EV_1	56
67 #define DB8500_DMA_MEMCPY_EV_2	57
68 #define DB8500_DMA_MEMCPY_EV_3	58
69 #define DB8500_DMA_MEMCPY_EV_4	59
70 #define DB8500_DMA_MEMCPY_EV_5	60
71 
72 static int dma40_memcpy_channels[] = {
73 	DB8500_DMA_MEMCPY_EV_0,
74 	DB8500_DMA_MEMCPY_EV_1,
75 	DB8500_DMA_MEMCPY_EV_2,
76 	DB8500_DMA_MEMCPY_EV_3,
77 	DB8500_DMA_MEMCPY_EV_4,
78 	DB8500_DMA_MEMCPY_EV_5,
79 };
80 
81 /* Default configuration for physcial memcpy */
82 static struct stedma40_chan_cfg dma40_memcpy_conf_phy = {
83 	.mode = STEDMA40_MODE_PHYSICAL,
84 	.dir = DMA_MEM_TO_MEM,
85 
86 	.src_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
87 	.src_info.psize = STEDMA40_PSIZE_PHY_1,
88 	.src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
89 
90 	.dst_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
91 	.dst_info.psize = STEDMA40_PSIZE_PHY_1,
92 	.dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
93 };
94 
95 /* Default configuration for logical memcpy */
96 static struct stedma40_chan_cfg dma40_memcpy_conf_log = {
97 	.mode = STEDMA40_MODE_LOGICAL,
98 	.dir = DMA_MEM_TO_MEM,
99 
100 	.src_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
101 	.src_info.psize = STEDMA40_PSIZE_LOG_1,
102 	.src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
103 
104 	.dst_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
105 	.dst_info.psize = STEDMA40_PSIZE_LOG_1,
106 	.dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
107 };
108 
109 /**
110  * enum 40_command - The different commands and/or statuses.
111  *
112  * @D40_DMA_STOP: DMA channel command STOP or status STOPPED,
113  * @D40_DMA_RUN: The DMA channel is RUNNING of the command RUN.
114  * @D40_DMA_SUSPEND_REQ: Request the DMA to SUSPEND as soon as possible.
115  * @D40_DMA_SUSPENDED: The DMA channel is SUSPENDED.
116  */
117 enum d40_command {
118 	D40_DMA_STOP		= 0,
119 	D40_DMA_RUN		= 1,
120 	D40_DMA_SUSPEND_REQ	= 2,
121 	D40_DMA_SUSPENDED	= 3
122 };
123 
124 /*
125  * enum d40_events - The different Event Enables for the event lines.
126  *
127  * @D40_DEACTIVATE_EVENTLINE: De-activate Event line, stopping the logical chan.
128  * @D40_ACTIVATE_EVENTLINE: Activate the Event line, to start a logical chan.
129  * @D40_SUSPEND_REQ_EVENTLINE: Requesting for suspending a event line.
130  * @D40_ROUND_EVENTLINE: Status check for event line.
131  */
132 
133 enum d40_events {
134 	D40_DEACTIVATE_EVENTLINE	= 0,
135 	D40_ACTIVATE_EVENTLINE		= 1,
136 	D40_SUSPEND_REQ_EVENTLINE	= 2,
137 	D40_ROUND_EVENTLINE		= 3
138 };
139 
140 /*
141  * These are the registers that has to be saved and later restored
142  * when the DMA hw is powered off.
143  * TODO: Add save/restore of D40_DREG_GCC on dma40 v3 or later, if that works.
144  */
145 static u32 d40_backup_regs[] = {
146 	D40_DREG_LCPA,
147 	D40_DREG_LCLA,
148 	D40_DREG_PRMSE,
149 	D40_DREG_PRMSO,
150 	D40_DREG_PRMOE,
151 	D40_DREG_PRMOO,
152 };
153 
154 #define BACKUP_REGS_SZ ARRAY_SIZE(d40_backup_regs)
155 
156 /*
157  * since 9540 and 8540 has the same HW revision
158  * use v4a for 9540 or ealier
159  * use v4b for 8540 or later
160  * HW revision:
161  * DB8500ed has revision 0
162  * DB8500v1 has revision 2
163  * DB8500v2 has revision 3
164  * AP9540v1 has revision 4
165  * DB8540v1 has revision 4
166  * TODO: Check if all these registers have to be saved/restored on dma40 v4a
167  */
168 static u32 d40_backup_regs_v4a[] = {
169 	D40_DREG_PSEG1,
170 	D40_DREG_PSEG2,
171 	D40_DREG_PSEG3,
172 	D40_DREG_PSEG4,
173 	D40_DREG_PCEG1,
174 	D40_DREG_PCEG2,
175 	D40_DREG_PCEG3,
176 	D40_DREG_PCEG4,
177 	D40_DREG_RSEG1,
178 	D40_DREG_RSEG2,
179 	D40_DREG_RSEG3,
180 	D40_DREG_RSEG4,
181 	D40_DREG_RCEG1,
182 	D40_DREG_RCEG2,
183 	D40_DREG_RCEG3,
184 	D40_DREG_RCEG4,
185 };
186 
187 #define BACKUP_REGS_SZ_V4A ARRAY_SIZE(d40_backup_regs_v4a)
188 
189 static u32 d40_backup_regs_v4b[] = {
190 	D40_DREG_CPSEG1,
191 	D40_DREG_CPSEG2,
192 	D40_DREG_CPSEG3,
193 	D40_DREG_CPSEG4,
194 	D40_DREG_CPSEG5,
195 	D40_DREG_CPCEG1,
196 	D40_DREG_CPCEG2,
197 	D40_DREG_CPCEG3,
198 	D40_DREG_CPCEG4,
199 	D40_DREG_CPCEG5,
200 	D40_DREG_CRSEG1,
201 	D40_DREG_CRSEG2,
202 	D40_DREG_CRSEG3,
203 	D40_DREG_CRSEG4,
204 	D40_DREG_CRSEG5,
205 	D40_DREG_CRCEG1,
206 	D40_DREG_CRCEG2,
207 	D40_DREG_CRCEG3,
208 	D40_DREG_CRCEG4,
209 	D40_DREG_CRCEG5,
210 };
211 
212 #define BACKUP_REGS_SZ_V4B ARRAY_SIZE(d40_backup_regs_v4b)
213 
214 static u32 d40_backup_regs_chan[] = {
215 	D40_CHAN_REG_SSCFG,
216 	D40_CHAN_REG_SSELT,
217 	D40_CHAN_REG_SSPTR,
218 	D40_CHAN_REG_SSLNK,
219 	D40_CHAN_REG_SDCFG,
220 	D40_CHAN_REG_SDELT,
221 	D40_CHAN_REG_SDPTR,
222 	D40_CHAN_REG_SDLNK,
223 };
224 
225 #define BACKUP_REGS_SZ_MAX ((BACKUP_REGS_SZ_V4A > BACKUP_REGS_SZ_V4B) ? \
226 			     BACKUP_REGS_SZ_V4A : BACKUP_REGS_SZ_V4B)
227 
228 /**
229  * struct d40_interrupt_lookup - lookup table for interrupt handler
230  *
231  * @src: Interrupt mask register.
232  * @clr: Interrupt clear register.
233  * @is_error: true if this is an error interrupt.
234  * @offset: start delta in the lookup_log_chans in d40_base. If equals to
235  * D40_PHY_CHAN, the lookup_phy_chans shall be used instead.
236  */
237 struct d40_interrupt_lookup {
238 	u32 src;
239 	u32 clr;
240 	bool is_error;
241 	int offset;
242 };
243 
244 
245 static struct d40_interrupt_lookup il_v4a[] = {
246 	{D40_DREG_LCTIS0, D40_DREG_LCICR0, false,  0},
247 	{D40_DREG_LCTIS1, D40_DREG_LCICR1, false, 32},
248 	{D40_DREG_LCTIS2, D40_DREG_LCICR2, false, 64},
249 	{D40_DREG_LCTIS3, D40_DREG_LCICR3, false, 96},
250 	{D40_DREG_LCEIS0, D40_DREG_LCICR0, true,   0},
251 	{D40_DREG_LCEIS1, D40_DREG_LCICR1, true,  32},
252 	{D40_DREG_LCEIS2, D40_DREG_LCICR2, true,  64},
253 	{D40_DREG_LCEIS3, D40_DREG_LCICR3, true,  96},
254 	{D40_DREG_PCTIS,  D40_DREG_PCICR,  false, D40_PHY_CHAN},
255 	{D40_DREG_PCEIS,  D40_DREG_PCICR,  true,  D40_PHY_CHAN},
256 };
257 
258 static struct d40_interrupt_lookup il_v4b[] = {
259 	{D40_DREG_CLCTIS1, D40_DREG_CLCICR1, false,  0},
260 	{D40_DREG_CLCTIS2, D40_DREG_CLCICR2, false, 32},
261 	{D40_DREG_CLCTIS3, D40_DREG_CLCICR3, false, 64},
262 	{D40_DREG_CLCTIS4, D40_DREG_CLCICR4, false, 96},
263 	{D40_DREG_CLCTIS5, D40_DREG_CLCICR5, false, 128},
264 	{D40_DREG_CLCEIS1, D40_DREG_CLCICR1, true,   0},
265 	{D40_DREG_CLCEIS2, D40_DREG_CLCICR2, true,  32},
266 	{D40_DREG_CLCEIS3, D40_DREG_CLCICR3, true,  64},
267 	{D40_DREG_CLCEIS4, D40_DREG_CLCICR4, true,  96},
268 	{D40_DREG_CLCEIS5, D40_DREG_CLCICR5, true,  128},
269 	{D40_DREG_CPCTIS,  D40_DREG_CPCICR,  false, D40_PHY_CHAN},
270 	{D40_DREG_CPCEIS,  D40_DREG_CPCICR,  true,  D40_PHY_CHAN},
271 };
272 
273 /**
274  * struct d40_reg_val - simple lookup struct
275  *
276  * @reg: The register.
277  * @val: The value that belongs to the register in reg.
278  */
279 struct d40_reg_val {
280 	unsigned int reg;
281 	unsigned int val;
282 };
283 
284 static __initdata struct d40_reg_val dma_init_reg_v4a[] = {
285 	/* Clock every part of the DMA block from start */
286 	{ .reg = D40_DREG_GCC,    .val = D40_DREG_GCC_ENABLE_ALL},
287 
288 	/* Interrupts on all logical channels */
289 	{ .reg = D40_DREG_LCMIS0, .val = 0xFFFFFFFF},
290 	{ .reg = D40_DREG_LCMIS1, .val = 0xFFFFFFFF},
291 	{ .reg = D40_DREG_LCMIS2, .val = 0xFFFFFFFF},
292 	{ .reg = D40_DREG_LCMIS3, .val = 0xFFFFFFFF},
293 	{ .reg = D40_DREG_LCICR0, .val = 0xFFFFFFFF},
294 	{ .reg = D40_DREG_LCICR1, .val = 0xFFFFFFFF},
295 	{ .reg = D40_DREG_LCICR2, .val = 0xFFFFFFFF},
296 	{ .reg = D40_DREG_LCICR3, .val = 0xFFFFFFFF},
297 	{ .reg = D40_DREG_LCTIS0, .val = 0xFFFFFFFF},
298 	{ .reg = D40_DREG_LCTIS1, .val = 0xFFFFFFFF},
299 	{ .reg = D40_DREG_LCTIS2, .val = 0xFFFFFFFF},
300 	{ .reg = D40_DREG_LCTIS3, .val = 0xFFFFFFFF}
301 };
302 static __initdata struct d40_reg_val dma_init_reg_v4b[] = {
303 	/* Clock every part of the DMA block from start */
304 	{ .reg = D40_DREG_GCC,    .val = D40_DREG_GCC_ENABLE_ALL},
305 
306 	/* Interrupts on all logical channels */
307 	{ .reg = D40_DREG_CLCMIS1, .val = 0xFFFFFFFF},
308 	{ .reg = D40_DREG_CLCMIS2, .val = 0xFFFFFFFF},
309 	{ .reg = D40_DREG_CLCMIS3, .val = 0xFFFFFFFF},
310 	{ .reg = D40_DREG_CLCMIS4, .val = 0xFFFFFFFF},
311 	{ .reg = D40_DREG_CLCMIS5, .val = 0xFFFFFFFF},
312 	{ .reg = D40_DREG_CLCICR1, .val = 0xFFFFFFFF},
313 	{ .reg = D40_DREG_CLCICR2, .val = 0xFFFFFFFF},
314 	{ .reg = D40_DREG_CLCICR3, .val = 0xFFFFFFFF},
315 	{ .reg = D40_DREG_CLCICR4, .val = 0xFFFFFFFF},
316 	{ .reg = D40_DREG_CLCICR5, .val = 0xFFFFFFFF},
317 	{ .reg = D40_DREG_CLCTIS1, .val = 0xFFFFFFFF},
318 	{ .reg = D40_DREG_CLCTIS2, .val = 0xFFFFFFFF},
319 	{ .reg = D40_DREG_CLCTIS3, .val = 0xFFFFFFFF},
320 	{ .reg = D40_DREG_CLCTIS4, .val = 0xFFFFFFFF},
321 	{ .reg = D40_DREG_CLCTIS5, .val = 0xFFFFFFFF}
322 };
323 
324 /**
325  * struct d40_lli_pool - Structure for keeping LLIs in memory
326  *
327  * @base: Pointer to memory area when the pre_alloc_lli's are not large
328  * enough, IE bigger than the most common case, 1 dst and 1 src. NULL if
329  * pre_alloc_lli is used.
330  * @dma_addr: DMA address, if mapped
331  * @size: The size in bytes of the memory at base or the size of pre_alloc_lli.
332  * @pre_alloc_lli: Pre allocated area for the most common case of transfers,
333  * one buffer to one buffer.
334  */
335 struct d40_lli_pool {
336 	void	*base;
337 	int	 size;
338 	dma_addr_t	dma_addr;
339 	/* Space for dst and src, plus an extra for padding */
340 	u8	 pre_alloc_lli[3 * sizeof(struct d40_phy_lli)];
341 };
342 
343 /**
344  * struct d40_desc - A descriptor is one DMA job.
345  *
346  * @lli_phy: LLI settings for physical channel. Both src and dst=
347  * points into the lli_pool, to base if lli_len > 1 or to pre_alloc_lli if
348  * lli_len equals one.
349  * @lli_log: Same as above but for logical channels.
350  * @lli_pool: The pool with two entries pre-allocated.
351  * @lli_len: Number of llis of current descriptor.
352  * @lli_current: Number of transferred llis.
353  * @lcla_alloc: Number of LCLA entries allocated.
354  * @txd: DMA engine struct. Used for among other things for communication
355  * during a transfer.
356  * @node: List entry.
357  * @is_in_client_list: true if the client owns this descriptor.
358  * @cyclic: true if this is a cyclic job
359  *
360  * This descriptor is used for both logical and physical transfers.
361  */
362 struct d40_desc {
363 	/* LLI physical */
364 	struct d40_phy_lli_bidir	 lli_phy;
365 	/* LLI logical */
366 	struct d40_log_lli_bidir	 lli_log;
367 
368 	struct d40_lli_pool		 lli_pool;
369 	int				 lli_len;
370 	int				 lli_current;
371 	int				 lcla_alloc;
372 
373 	struct dma_async_tx_descriptor	 txd;
374 	struct list_head		 node;
375 
376 	bool				 is_in_client_list;
377 	bool				 cyclic;
378 };
379 
380 /**
381  * struct d40_lcla_pool - LCLA pool settings and data.
382  *
383  * @base: The virtual address of LCLA. 18 bit aligned.
384  * @base_unaligned: The orignal kmalloc pointer, if kmalloc is used.
385  * This pointer is only there for clean-up on error.
386  * @pages: The number of pages needed for all physical channels.
387  * Only used later for clean-up on error
388  * @lock: Lock to protect the content in this struct.
389  * @alloc_map: big map over which LCLA entry is own by which job.
390  */
391 struct d40_lcla_pool {
392 	void		*base;
393 	dma_addr_t	dma_addr;
394 	void		*base_unaligned;
395 	int		 pages;
396 	spinlock_t	 lock;
397 	struct d40_desc	**alloc_map;
398 };
399 
400 /**
401  * struct d40_phy_res - struct for handling eventlines mapped to physical
402  * channels.
403  *
404  * @lock: A lock protection this entity.
405  * @reserved: True if used by secure world or otherwise.
406  * @num: The physical channel number of this entity.
407  * @allocated_src: Bit mapped to show which src event line's are mapped to
408  * this physical channel. Can also be free or physically allocated.
409  * @allocated_dst: Same as for src but is dst.
410  * allocated_dst and allocated_src uses the D40_ALLOC* defines as well as
411  * event line number.
412  * @use_soft_lli: To mark if the linked lists of channel are managed by SW.
413  */
414 struct d40_phy_res {
415 	spinlock_t lock;
416 	bool	   reserved;
417 	int	   num;
418 	u32	   allocated_src;
419 	u32	   allocated_dst;
420 	bool	   use_soft_lli;
421 };
422 
423 struct d40_base;
424 
425 /**
426  * struct d40_chan - Struct that describes a channel.
427  *
428  * @lock: A spinlock to protect this struct.
429  * @log_num: The logical number, if any of this channel.
430  * @pending_tx: The number of pending transfers. Used between interrupt handler
431  * and tasklet.
432  * @busy: Set to true when transfer is ongoing on this channel.
433  * @phy_chan: Pointer to physical channel which this instance runs on. If this
434  * point is NULL, then the channel is not allocated.
435  * @chan: DMA engine handle.
436  * @tasklet: Tasklet that gets scheduled from interrupt context to complete a
437  * transfer and call client callback.
438  * @client: Cliented owned descriptor list.
439  * @pending_queue: Submitted jobs, to be issued by issue_pending()
440  * @active: Active descriptor.
441  * @done: Completed jobs
442  * @queue: Queued jobs.
443  * @prepare_queue: Prepared jobs.
444  * @dma_cfg: The client configuration of this dma channel.
445  * @configured: whether the dma_cfg configuration is valid
446  * @base: Pointer to the device instance struct.
447  * @src_def_cfg: Default cfg register setting for src.
448  * @dst_def_cfg: Default cfg register setting for dst.
449  * @log_def: Default logical channel settings.
450  * @lcpa: Pointer to dst and src lcpa settings.
451  * @runtime_addr: runtime configured address.
452  * @runtime_direction: runtime configured direction.
453  *
454  * This struct can either "be" a logical or a physical channel.
455  */
456 struct d40_chan {
457 	spinlock_t			 lock;
458 	int				 log_num;
459 	int				 pending_tx;
460 	bool				 busy;
461 	struct d40_phy_res		*phy_chan;
462 	struct dma_chan			 chan;
463 	struct tasklet_struct		 tasklet;
464 	struct list_head		 client;
465 	struct list_head		 pending_queue;
466 	struct list_head		 active;
467 	struct list_head		 done;
468 	struct list_head		 queue;
469 	struct list_head		 prepare_queue;
470 	struct stedma40_chan_cfg	 dma_cfg;
471 	bool				 configured;
472 	struct d40_base			*base;
473 	/* Default register configurations */
474 	u32				 src_def_cfg;
475 	u32				 dst_def_cfg;
476 	struct d40_def_lcsp		 log_def;
477 	struct d40_log_lli_full		*lcpa;
478 	/* Runtime reconfiguration */
479 	dma_addr_t			runtime_addr;
480 	enum dma_transfer_direction	runtime_direction;
481 };
482 
483 /**
484  * struct d40_gen_dmac - generic values to represent u8500/u8540 DMA
485  * controller
486  *
487  * @backup: the pointer to the registers address array for backup
488  * @backup_size: the size of the registers address array for backup
489  * @realtime_en: the realtime enable register
490  * @realtime_clear: the realtime clear register
491  * @high_prio_en: the high priority enable register
492  * @high_prio_clear: the high priority clear register
493  * @interrupt_en: the interrupt enable register
494  * @interrupt_clear: the interrupt clear register
495  * @il: the pointer to struct d40_interrupt_lookup
496  * @il_size: the size of d40_interrupt_lookup array
497  * @init_reg: the pointer to the struct d40_reg_val
498  * @init_reg_size: the size of d40_reg_val array
499  */
500 struct d40_gen_dmac {
501 	u32				*backup;
502 	u32				 backup_size;
503 	u32				 realtime_en;
504 	u32				 realtime_clear;
505 	u32				 high_prio_en;
506 	u32				 high_prio_clear;
507 	u32				 interrupt_en;
508 	u32				 interrupt_clear;
509 	struct d40_interrupt_lookup	*il;
510 	u32				 il_size;
511 	struct d40_reg_val		*init_reg;
512 	u32				 init_reg_size;
513 };
514 
515 /**
516  * struct d40_base - The big global struct, one for each probe'd instance.
517  *
518  * @interrupt_lock: Lock used to make sure one interrupt is handle a time.
519  * @execmd_lock: Lock for execute command usage since several channels share
520  * the same physical register.
521  * @dev: The device structure.
522  * @virtbase: The virtual base address of the DMA's register.
523  * @rev: silicon revision detected.
524  * @clk: Pointer to the DMA clock structure.
525  * @phy_start: Physical memory start of the DMA registers.
526  * @phy_size: Size of the DMA register map.
527  * @irq: The IRQ number.
528  * @num_memcpy_chans: The number of channels used for memcpy (mem-to-mem
529  * transfers).
530  * @num_phy_chans: The number of physical channels. Read from HW. This
531  * is the number of available channels for this driver, not counting "Secure
532  * mode" allocated physical channels.
533  * @num_log_chans: The number of logical channels. Calculated from
534  * num_phy_chans.
535  * @dma_both: dma_device channels that can do both memcpy and slave transfers.
536  * @dma_slave: dma_device channels that can do only do slave transfers.
537  * @dma_memcpy: dma_device channels that can do only do memcpy transfers.
538  * @phy_chans: Room for all possible physical channels in system.
539  * @log_chans: Room for all possible logical channels in system.
540  * @lookup_log_chans: Used to map interrupt number to logical channel. Points
541  * to log_chans entries.
542  * @lookup_phy_chans: Used to map interrupt number to physical channel. Points
543  * to phy_chans entries.
544  * @plat_data: Pointer to provided platform_data which is the driver
545  * configuration.
546  * @lcpa_regulator: Pointer to hold the regulator for the esram bank for lcla.
547  * @phy_res: Vector containing all physical channels.
548  * @lcla_pool: lcla pool settings and data.
549  * @lcpa_base: The virtual mapped address of LCPA.
550  * @phy_lcpa: The physical address of the LCPA.
551  * @lcpa_size: The size of the LCPA area.
552  * @desc_slab: cache for descriptors.
553  * @reg_val_backup: Here the values of some hardware registers are stored
554  * before the DMA is powered off. They are restored when the power is back on.
555  * @reg_val_backup_v4: Backup of registers that only exits on dma40 v3 and
556  * later
557  * @reg_val_backup_chan: Backup data for standard channel parameter registers.
558  * @gcc_pwr_off_mask: Mask to maintain the channels that can be turned off.
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 	struct d40_gen_dmac		  gen_dmac;
597 };
598 
599 static struct device *chan2dev(struct d40_chan *d40c)
600 {
601 	return &d40c->chan.dev->device;
602 }
603 
604 static bool chan_is_physical(struct d40_chan *chan)
605 {
606 	return chan->log_num == D40_PHY_CHAN;
607 }
608 
609 static bool chan_is_logical(struct d40_chan *chan)
610 {
611 	return !chan_is_physical(chan);
612 }
613 
614 static void __iomem *chan_base(struct d40_chan *chan)
615 {
616 	return chan->base->virtbase + D40_DREG_PCBASE +
617 	       chan->phy_chan->num * D40_DREG_PCDELTA;
618 }
619 
620 #define d40_err(dev, format, arg...)		\
621 	dev_err(dev, "[%s] " format, __func__, ## arg)
622 
623 #define chan_err(d40c, format, arg...)		\
624 	d40_err(chan2dev(d40c), format, ## arg)
625 
626 static int d40_pool_lli_alloc(struct d40_chan *d40c, struct d40_desc *d40d,
627 			      int lli_len)
628 {
629 	bool is_log = chan_is_logical(d40c);
630 	u32 align;
631 	void *base;
632 
633 	if (is_log)
634 		align = sizeof(struct d40_log_lli);
635 	else
636 		align = sizeof(struct d40_phy_lli);
637 
638 	if (lli_len == 1) {
639 		base = d40d->lli_pool.pre_alloc_lli;
640 		d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
641 		d40d->lli_pool.base = NULL;
642 	} else {
643 		d40d->lli_pool.size = lli_len * 2 * align;
644 
645 		base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
646 		d40d->lli_pool.base = base;
647 
648 		if (d40d->lli_pool.base == NULL)
649 			return -ENOMEM;
650 	}
651 
652 	if (is_log) {
653 		d40d->lli_log.src = PTR_ALIGN(base, align);
654 		d40d->lli_log.dst = d40d->lli_log.src + lli_len;
655 
656 		d40d->lli_pool.dma_addr = 0;
657 	} else {
658 		d40d->lli_phy.src = PTR_ALIGN(base, align);
659 		d40d->lli_phy.dst = d40d->lli_phy.src + lli_len;
660 
661 		d40d->lli_pool.dma_addr = dma_map_single(d40c->base->dev,
662 							 d40d->lli_phy.src,
663 							 d40d->lli_pool.size,
664 							 DMA_TO_DEVICE);
665 
666 		if (dma_mapping_error(d40c->base->dev,
667 				      d40d->lli_pool.dma_addr)) {
668 			kfree(d40d->lli_pool.base);
669 			d40d->lli_pool.base = NULL;
670 			d40d->lli_pool.dma_addr = 0;
671 			return -ENOMEM;
672 		}
673 	}
674 
675 	return 0;
676 }
677 
678 static void d40_pool_lli_free(struct d40_chan *d40c, struct d40_desc *d40d)
679 {
680 	if (d40d->lli_pool.dma_addr)
681 		dma_unmap_single(d40c->base->dev, d40d->lli_pool.dma_addr,
682 				 d40d->lli_pool.size, DMA_TO_DEVICE);
683 
684 	kfree(d40d->lli_pool.base);
685 	d40d->lli_pool.base = NULL;
686 	d40d->lli_pool.size = 0;
687 	d40d->lli_log.src = NULL;
688 	d40d->lli_log.dst = NULL;
689 	d40d->lli_phy.src = NULL;
690 	d40d->lli_phy.dst = NULL;
691 }
692 
693 static int d40_lcla_alloc_one(struct d40_chan *d40c,
694 			      struct d40_desc *d40d)
695 {
696 	unsigned long flags;
697 	int i;
698 	int ret = -EINVAL;
699 
700 	spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
701 
702 	/*
703 	 * Allocate both src and dst at the same time, therefore the half
704 	 * start on 1 since 0 can't be used since zero is used as end marker.
705 	 */
706 	for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
707 		int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
708 
709 		if (!d40c->base->lcla_pool.alloc_map[idx]) {
710 			d40c->base->lcla_pool.alloc_map[idx] = d40d;
711 			d40d->lcla_alloc++;
712 			ret = i;
713 			break;
714 		}
715 	}
716 
717 	spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
718 
719 	return ret;
720 }
721 
722 static int d40_lcla_free_all(struct d40_chan *d40c,
723 			     struct d40_desc *d40d)
724 {
725 	unsigned long flags;
726 	int i;
727 	int ret = -EINVAL;
728 
729 	if (chan_is_physical(d40c))
730 		return 0;
731 
732 	spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
733 
734 	for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
735 		int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
736 
737 		if (d40c->base->lcla_pool.alloc_map[idx] == d40d) {
738 			d40c->base->lcla_pool.alloc_map[idx] = NULL;
739 			d40d->lcla_alloc--;
740 			if (d40d->lcla_alloc == 0) {
741 				ret = 0;
742 				break;
743 			}
744 		}
745 	}
746 
747 	spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
748 
749 	return ret;
750 
751 }
752 
753 static void d40_desc_remove(struct d40_desc *d40d)
754 {
755 	list_del(&d40d->node);
756 }
757 
758 static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
759 {
760 	struct d40_desc *desc = NULL;
761 
762 	if (!list_empty(&d40c->client)) {
763 		struct d40_desc *d;
764 		struct d40_desc *_d;
765 
766 		list_for_each_entry_safe(d, _d, &d40c->client, node) {
767 			if (async_tx_test_ack(&d->txd)) {
768 				d40_desc_remove(d);
769 				desc = d;
770 				memset(desc, 0, sizeof(*desc));
771 				break;
772 			}
773 		}
774 	}
775 
776 	if (!desc)
777 		desc = kmem_cache_zalloc(d40c->base->desc_slab, GFP_NOWAIT);
778 
779 	if (desc)
780 		INIT_LIST_HEAD(&desc->node);
781 
782 	return desc;
783 }
784 
785 static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
786 {
787 
788 	d40_pool_lli_free(d40c, d40d);
789 	d40_lcla_free_all(d40c, d40d);
790 	kmem_cache_free(d40c->base->desc_slab, d40d);
791 }
792 
793 static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
794 {
795 	list_add_tail(&desc->node, &d40c->active);
796 }
797 
798 static void d40_phy_lli_load(struct d40_chan *chan, struct d40_desc *desc)
799 {
800 	struct d40_phy_lli *lli_dst = desc->lli_phy.dst;
801 	struct d40_phy_lli *lli_src = desc->lli_phy.src;
802 	void __iomem *base = chan_base(chan);
803 
804 	writel(lli_src->reg_cfg, base + D40_CHAN_REG_SSCFG);
805 	writel(lli_src->reg_elt, base + D40_CHAN_REG_SSELT);
806 	writel(lli_src->reg_ptr, base + D40_CHAN_REG_SSPTR);
807 	writel(lli_src->reg_lnk, base + D40_CHAN_REG_SSLNK);
808 
809 	writel(lli_dst->reg_cfg, base + D40_CHAN_REG_SDCFG);
810 	writel(lli_dst->reg_elt, base + D40_CHAN_REG_SDELT);
811 	writel(lli_dst->reg_ptr, base + D40_CHAN_REG_SDPTR);
812 	writel(lli_dst->reg_lnk, base + D40_CHAN_REG_SDLNK);
813 }
814 
815 static void d40_desc_done(struct d40_chan *d40c, struct d40_desc *desc)
816 {
817 	list_add_tail(&desc->node, &d40c->done);
818 }
819 
820 static void d40_log_lli_to_lcxa(struct d40_chan *chan, struct d40_desc *desc)
821 {
822 	struct d40_lcla_pool *pool = &chan->base->lcla_pool;
823 	struct d40_log_lli_bidir *lli = &desc->lli_log;
824 	int lli_current = desc->lli_current;
825 	int lli_len = desc->lli_len;
826 	bool cyclic = desc->cyclic;
827 	int curr_lcla = -EINVAL;
828 	int first_lcla = 0;
829 	bool use_esram_lcla = chan->base->plat_data->use_esram_lcla;
830 	bool linkback;
831 
832 	/*
833 	 * We may have partially running cyclic transfers, in case we did't get
834 	 * enough LCLA entries.
835 	 */
836 	linkback = cyclic && lli_current == 0;
837 
838 	/*
839 	 * For linkback, we need one LCLA even with only one link, because we
840 	 * can't link back to the one in LCPA space
841 	 */
842 	if (linkback || (lli_len - lli_current > 1)) {
843 		/*
844 		 * If the channel is expected to use only soft_lli don't
845 		 * allocate a lcla. This is to avoid a HW issue that exists
846 		 * in some controller during a peripheral to memory transfer
847 		 * that uses linked lists.
848 		 */
849 		if (!(chan->phy_chan->use_soft_lli &&
850 			chan->dma_cfg.dir == DMA_DEV_TO_MEM))
851 			curr_lcla = d40_lcla_alloc_one(chan, desc);
852 
853 		first_lcla = curr_lcla;
854 	}
855 
856 	/*
857 	 * For linkback, we normally load the LCPA in the loop since we need to
858 	 * link it to the second LCLA and not the first.  However, if we
859 	 * couldn't even get a first LCLA, then we have to run in LCPA and
860 	 * reload manually.
861 	 */
862 	if (!linkback || curr_lcla == -EINVAL) {
863 		unsigned int flags = 0;
864 
865 		if (curr_lcla == -EINVAL)
866 			flags |= LLI_TERM_INT;
867 
868 		d40_log_lli_lcpa_write(chan->lcpa,
869 				       &lli->dst[lli_current],
870 				       &lli->src[lli_current],
871 				       curr_lcla,
872 				       flags);
873 		lli_current++;
874 	}
875 
876 	if (curr_lcla < 0)
877 		goto out;
878 
879 	for (; lli_current < lli_len; lli_current++) {
880 		unsigned int lcla_offset = chan->phy_chan->num * 1024 +
881 					   8 * curr_lcla * 2;
882 		struct d40_log_lli *lcla = pool->base + lcla_offset;
883 		unsigned int flags = 0;
884 		int next_lcla;
885 
886 		if (lli_current + 1 < lli_len)
887 			next_lcla = d40_lcla_alloc_one(chan, desc);
888 		else
889 			next_lcla = linkback ? first_lcla : -EINVAL;
890 
891 		if (cyclic || next_lcla == -EINVAL)
892 			flags |= LLI_TERM_INT;
893 
894 		if (linkback && curr_lcla == first_lcla) {
895 			/* First link goes in both LCPA and LCLA */
896 			d40_log_lli_lcpa_write(chan->lcpa,
897 					       &lli->dst[lli_current],
898 					       &lli->src[lli_current],
899 					       next_lcla, flags);
900 		}
901 
902 		/*
903 		 * One unused LCLA in the cyclic case if the very first
904 		 * next_lcla fails...
905 		 */
906 		d40_log_lli_lcla_write(lcla,
907 				       &lli->dst[lli_current],
908 				       &lli->src[lli_current],
909 				       next_lcla, flags);
910 
911 		/*
912 		 * Cache maintenance is not needed if lcla is
913 		 * mapped in esram
914 		 */
915 		if (!use_esram_lcla) {
916 			dma_sync_single_range_for_device(chan->base->dev,
917 						pool->dma_addr, lcla_offset,
918 						2 * sizeof(struct d40_log_lli),
919 						DMA_TO_DEVICE);
920 		}
921 		curr_lcla = next_lcla;
922 
923 		if (curr_lcla == -EINVAL || curr_lcla == first_lcla) {
924 			lli_current++;
925 			break;
926 		}
927 	}
928 
929 out:
930 	desc->lli_current = lli_current;
931 }
932 
933 static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
934 {
935 	if (chan_is_physical(d40c)) {
936 		d40_phy_lli_load(d40c, d40d);
937 		d40d->lli_current = d40d->lli_len;
938 	} else
939 		d40_log_lli_to_lcxa(d40c, d40d);
940 }
941 
942 static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
943 {
944 	struct d40_desc *d;
945 
946 	if (list_empty(&d40c->active))
947 		return NULL;
948 
949 	d = list_first_entry(&d40c->active,
950 			     struct d40_desc,
951 			     node);
952 	return d;
953 }
954 
955 /* remove desc from current queue and add it to the pending_queue */
956 static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
957 {
958 	d40_desc_remove(desc);
959 	desc->is_in_client_list = false;
960 	list_add_tail(&desc->node, &d40c->pending_queue);
961 }
962 
963 static struct d40_desc *d40_first_pending(struct d40_chan *d40c)
964 {
965 	struct d40_desc *d;
966 
967 	if (list_empty(&d40c->pending_queue))
968 		return NULL;
969 
970 	d = list_first_entry(&d40c->pending_queue,
971 			     struct d40_desc,
972 			     node);
973 	return d;
974 }
975 
976 static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
977 {
978 	struct d40_desc *d;
979 
980 	if (list_empty(&d40c->queue))
981 		return NULL;
982 
983 	d = list_first_entry(&d40c->queue,
984 			     struct d40_desc,
985 			     node);
986 	return d;
987 }
988 
989 static struct d40_desc *d40_first_done(struct d40_chan *d40c)
990 {
991 	if (list_empty(&d40c->done))
992 		return NULL;
993 
994 	return list_first_entry(&d40c->done, struct d40_desc, node);
995 }
996 
997 static int d40_psize_2_burst_size(bool is_log, int psize)
998 {
999 	if (is_log) {
1000 		if (psize == STEDMA40_PSIZE_LOG_1)
1001 			return 1;
1002 	} else {
1003 		if (psize == STEDMA40_PSIZE_PHY_1)
1004 			return 1;
1005 	}
1006 
1007 	return 2 << psize;
1008 }
1009 
1010 /*
1011  * The dma only supports transmitting packages up to
1012  * STEDMA40_MAX_SEG_SIZE * data_width, where data_width is stored in Bytes.
1013  *
1014  * Calculate the total number of dma elements required to send the entire sg list.
1015  */
1016 static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2)
1017 {
1018 	int dmalen;
1019 	u32 max_w = max(data_width1, data_width2);
1020 	u32 min_w = min(data_width1, data_width2);
1021 	u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE * min_w, max_w);
1022 
1023 	if (seg_max > STEDMA40_MAX_SEG_SIZE)
1024 		seg_max -= max_w;
1025 
1026 	if (!IS_ALIGNED(size, max_w))
1027 		return -EINVAL;
1028 
1029 	if (size <= seg_max)
1030 		dmalen = 1;
1031 	else {
1032 		dmalen = size / seg_max;
1033 		if (dmalen * seg_max < size)
1034 			dmalen++;
1035 	}
1036 	return dmalen;
1037 }
1038 
1039 static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len,
1040 			   u32 data_width1, u32 data_width2)
1041 {
1042 	struct scatterlist *sg;
1043 	int i;
1044 	int len = 0;
1045 	int ret;
1046 
1047 	for_each_sg(sgl, sg, sg_len, i) {
1048 		ret = d40_size_2_dmalen(sg_dma_len(sg),
1049 					data_width1, data_width2);
1050 		if (ret < 0)
1051 			return ret;
1052 		len += ret;
1053 	}
1054 	return len;
1055 }
1056 
1057 static int __d40_execute_command_phy(struct d40_chan *d40c,
1058 				     enum d40_command command)
1059 {
1060 	u32 status;
1061 	int i;
1062 	void __iomem *active_reg;
1063 	int ret = 0;
1064 	unsigned long flags;
1065 	u32 wmask;
1066 
1067 	if (command == D40_DMA_STOP) {
1068 		ret = __d40_execute_command_phy(d40c, D40_DMA_SUSPEND_REQ);
1069 		if (ret)
1070 			return ret;
1071 	}
1072 
1073 	spin_lock_irqsave(&d40c->base->execmd_lock, flags);
1074 
1075 	if (d40c->phy_chan->num % 2 == 0)
1076 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1077 	else
1078 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1079 
1080 	if (command == D40_DMA_SUSPEND_REQ) {
1081 		status = (readl(active_reg) &
1082 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1083 			D40_CHAN_POS(d40c->phy_chan->num);
1084 
1085 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1086 			goto done;
1087 	}
1088 
1089 	wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
1090 	writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
1091 	       active_reg);
1092 
1093 	if (command == D40_DMA_SUSPEND_REQ) {
1094 
1095 		for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
1096 			status = (readl(active_reg) &
1097 				  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1098 				D40_CHAN_POS(d40c->phy_chan->num);
1099 
1100 			cpu_relax();
1101 			/*
1102 			 * Reduce the number of bus accesses while
1103 			 * waiting for the DMA to suspend.
1104 			 */
1105 			udelay(3);
1106 
1107 			if (status == D40_DMA_STOP ||
1108 			    status == D40_DMA_SUSPENDED)
1109 				break;
1110 		}
1111 
1112 		if (i == D40_SUSPEND_MAX_IT) {
1113 			chan_err(d40c,
1114 				"unable to suspend the chl %d (log: %d) status %x\n",
1115 				d40c->phy_chan->num, d40c->log_num,
1116 				status);
1117 			dump_stack();
1118 			ret = -EBUSY;
1119 		}
1120 
1121 	}
1122 done:
1123 	spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
1124 	return ret;
1125 }
1126 
1127 static void d40_term_all(struct d40_chan *d40c)
1128 {
1129 	struct d40_desc *d40d;
1130 	struct d40_desc *_d;
1131 
1132 	/* Release completed descriptors */
1133 	while ((d40d = d40_first_done(d40c))) {
1134 		d40_desc_remove(d40d);
1135 		d40_desc_free(d40c, d40d);
1136 	}
1137 
1138 	/* Release active descriptors */
1139 	while ((d40d = d40_first_active_get(d40c))) {
1140 		d40_desc_remove(d40d);
1141 		d40_desc_free(d40c, d40d);
1142 	}
1143 
1144 	/* Release queued descriptors waiting for transfer */
1145 	while ((d40d = d40_first_queued(d40c))) {
1146 		d40_desc_remove(d40d);
1147 		d40_desc_free(d40c, d40d);
1148 	}
1149 
1150 	/* Release pending descriptors */
1151 	while ((d40d = d40_first_pending(d40c))) {
1152 		d40_desc_remove(d40d);
1153 		d40_desc_free(d40c, d40d);
1154 	}
1155 
1156 	/* Release client owned descriptors */
1157 	if (!list_empty(&d40c->client))
1158 		list_for_each_entry_safe(d40d, _d, &d40c->client, node) {
1159 			d40_desc_remove(d40d);
1160 			d40_desc_free(d40c, d40d);
1161 		}
1162 
1163 	/* Release descriptors in prepare queue */
1164 	if (!list_empty(&d40c->prepare_queue))
1165 		list_for_each_entry_safe(d40d, _d,
1166 					 &d40c->prepare_queue, node) {
1167 			d40_desc_remove(d40d);
1168 			d40_desc_free(d40c, d40d);
1169 		}
1170 
1171 	d40c->pending_tx = 0;
1172 }
1173 
1174 static void __d40_config_set_event(struct d40_chan *d40c,
1175 				   enum d40_events event_type, u32 event,
1176 				   int reg)
1177 {
1178 	void __iomem *addr = chan_base(d40c) + reg;
1179 	int tries;
1180 	u32 status;
1181 
1182 	switch (event_type) {
1183 
1184 	case D40_DEACTIVATE_EVENTLINE:
1185 
1186 		writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
1187 		       | ~D40_EVENTLINE_MASK(event), addr);
1188 		break;
1189 
1190 	case D40_SUSPEND_REQ_EVENTLINE:
1191 		status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1192 			  D40_EVENTLINE_POS(event);
1193 
1194 		if (status == D40_DEACTIVATE_EVENTLINE ||
1195 		    status == D40_SUSPEND_REQ_EVENTLINE)
1196 			break;
1197 
1198 		writel((D40_SUSPEND_REQ_EVENTLINE << D40_EVENTLINE_POS(event))
1199 		       | ~D40_EVENTLINE_MASK(event), addr);
1200 
1201 		for (tries = 0 ; tries < D40_SUSPEND_MAX_IT; tries++) {
1202 
1203 			status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1204 				  D40_EVENTLINE_POS(event);
1205 
1206 			cpu_relax();
1207 			/*
1208 			 * Reduce the number of bus accesses while
1209 			 * waiting for the DMA to suspend.
1210 			 */
1211 			udelay(3);
1212 
1213 			if (status == D40_DEACTIVATE_EVENTLINE)
1214 				break;
1215 		}
1216 
1217 		if (tries == D40_SUSPEND_MAX_IT) {
1218 			chan_err(d40c,
1219 				"unable to stop the event_line chl %d (log: %d)"
1220 				"status %x\n", d40c->phy_chan->num,
1221 				 d40c->log_num, status);
1222 		}
1223 		break;
1224 
1225 	case D40_ACTIVATE_EVENTLINE:
1226 	/*
1227 	 * The hardware sometimes doesn't register the enable when src and dst
1228 	 * event lines are active on the same logical channel.  Retry to ensure
1229 	 * it does.  Usually only one retry is sufficient.
1230 	 */
1231 		tries = 100;
1232 		while (--tries) {
1233 			writel((D40_ACTIVATE_EVENTLINE <<
1234 				D40_EVENTLINE_POS(event)) |
1235 				~D40_EVENTLINE_MASK(event), addr);
1236 
1237 			if (readl(addr) & D40_EVENTLINE_MASK(event))
1238 				break;
1239 		}
1240 
1241 		if (tries != 99)
1242 			dev_dbg(chan2dev(d40c),
1243 				"[%s] workaround enable S%cLNK (%d tries)\n",
1244 				__func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D',
1245 				100 - tries);
1246 
1247 		WARN_ON(!tries);
1248 		break;
1249 
1250 	case D40_ROUND_EVENTLINE:
1251 		BUG();
1252 		break;
1253 
1254 	}
1255 }
1256 
1257 static void d40_config_set_event(struct d40_chan *d40c,
1258 				 enum d40_events event_type)
1259 {
1260 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
1261 
1262 	/* Enable event line connected to device (or memcpy) */
1263 	if ((d40c->dma_cfg.dir == DMA_DEV_TO_MEM) ||
1264 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
1265 		__d40_config_set_event(d40c, event_type, event,
1266 				       D40_CHAN_REG_SSLNK);
1267 
1268 	if (d40c->dma_cfg.dir !=  DMA_DEV_TO_MEM)
1269 		__d40_config_set_event(d40c, event_type, event,
1270 				       D40_CHAN_REG_SDLNK);
1271 }
1272 
1273 static u32 d40_chan_has_events(struct d40_chan *d40c)
1274 {
1275 	void __iomem *chanbase = chan_base(d40c);
1276 	u32 val;
1277 
1278 	val = readl(chanbase + D40_CHAN_REG_SSLNK);
1279 	val |= readl(chanbase + D40_CHAN_REG_SDLNK);
1280 
1281 	return val;
1282 }
1283 
1284 static int
1285 __d40_execute_command_log(struct d40_chan *d40c, enum d40_command command)
1286 {
1287 	unsigned long flags;
1288 	int ret = 0;
1289 	u32 active_status;
1290 	void __iomem *active_reg;
1291 
1292 	if (d40c->phy_chan->num % 2 == 0)
1293 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1294 	else
1295 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1296 
1297 
1298 	spin_lock_irqsave(&d40c->phy_chan->lock, flags);
1299 
1300 	switch (command) {
1301 	case D40_DMA_STOP:
1302 	case D40_DMA_SUSPEND_REQ:
1303 
1304 		active_status = (readl(active_reg) &
1305 				 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1306 				 D40_CHAN_POS(d40c->phy_chan->num);
1307 
1308 		if (active_status == D40_DMA_RUN)
1309 			d40_config_set_event(d40c, D40_SUSPEND_REQ_EVENTLINE);
1310 		else
1311 			d40_config_set_event(d40c, D40_DEACTIVATE_EVENTLINE);
1312 
1313 		if (!d40_chan_has_events(d40c) && (command == D40_DMA_STOP))
1314 			ret = __d40_execute_command_phy(d40c, command);
1315 
1316 		break;
1317 
1318 	case D40_DMA_RUN:
1319 
1320 		d40_config_set_event(d40c, D40_ACTIVATE_EVENTLINE);
1321 		ret = __d40_execute_command_phy(d40c, command);
1322 		break;
1323 
1324 	case D40_DMA_SUSPENDED:
1325 		BUG();
1326 		break;
1327 	}
1328 
1329 	spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
1330 	return ret;
1331 }
1332 
1333 static int d40_channel_execute_command(struct d40_chan *d40c,
1334 				       enum d40_command command)
1335 {
1336 	if (chan_is_logical(d40c))
1337 		return __d40_execute_command_log(d40c, command);
1338 	else
1339 		return __d40_execute_command_phy(d40c, command);
1340 }
1341 
1342 static u32 d40_get_prmo(struct d40_chan *d40c)
1343 {
1344 	static const unsigned int phy_map[] = {
1345 		[STEDMA40_PCHAN_BASIC_MODE]
1346 			= D40_DREG_PRMO_PCHAN_BASIC,
1347 		[STEDMA40_PCHAN_MODULO_MODE]
1348 			= D40_DREG_PRMO_PCHAN_MODULO,
1349 		[STEDMA40_PCHAN_DOUBLE_DST_MODE]
1350 			= D40_DREG_PRMO_PCHAN_DOUBLE_DST,
1351 	};
1352 	static const unsigned int log_map[] = {
1353 		[STEDMA40_LCHAN_SRC_PHY_DST_LOG]
1354 			= D40_DREG_PRMO_LCHAN_SRC_PHY_DST_LOG,
1355 		[STEDMA40_LCHAN_SRC_LOG_DST_PHY]
1356 			= D40_DREG_PRMO_LCHAN_SRC_LOG_DST_PHY,
1357 		[STEDMA40_LCHAN_SRC_LOG_DST_LOG]
1358 			= D40_DREG_PRMO_LCHAN_SRC_LOG_DST_LOG,
1359 	};
1360 
1361 	if (chan_is_physical(d40c))
1362 		return phy_map[d40c->dma_cfg.mode_opt];
1363 	else
1364 		return log_map[d40c->dma_cfg.mode_opt];
1365 }
1366 
1367 static void d40_config_write(struct d40_chan *d40c)
1368 {
1369 	u32 addr_base;
1370 	u32 var;
1371 
1372 	/* Odd addresses are even addresses + 4 */
1373 	addr_base = (d40c->phy_chan->num % 2) * 4;
1374 	/* Setup channel mode to logical or physical */
1375 	var = ((u32)(chan_is_logical(d40c)) + 1) <<
1376 		D40_CHAN_POS(d40c->phy_chan->num);
1377 	writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
1378 
1379 	/* Setup operational mode option register */
1380 	var = d40_get_prmo(d40c) << D40_CHAN_POS(d40c->phy_chan->num);
1381 
1382 	writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
1383 
1384 	if (chan_is_logical(d40c)) {
1385 		int lidx = (d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS)
1386 			   & D40_SREG_ELEM_LOG_LIDX_MASK;
1387 		void __iomem *chanbase = chan_base(d40c);
1388 
1389 		/* Set default config for CFG reg */
1390 		writel(d40c->src_def_cfg, chanbase + D40_CHAN_REG_SSCFG);
1391 		writel(d40c->dst_def_cfg, chanbase + D40_CHAN_REG_SDCFG);
1392 
1393 		/* Set LIDX for lcla */
1394 		writel(lidx, chanbase + D40_CHAN_REG_SSELT);
1395 		writel(lidx, chanbase + D40_CHAN_REG_SDELT);
1396 
1397 		/* Clear LNK which will be used by d40_chan_has_events() */
1398 		writel(0, chanbase + D40_CHAN_REG_SSLNK);
1399 		writel(0, chanbase + D40_CHAN_REG_SDLNK);
1400 	}
1401 }
1402 
1403 static u32 d40_residue(struct d40_chan *d40c)
1404 {
1405 	u32 num_elt;
1406 
1407 	if (chan_is_logical(d40c))
1408 		num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
1409 			>> D40_MEM_LCSP2_ECNT_POS;
1410 	else {
1411 		u32 val = readl(chan_base(d40c) + D40_CHAN_REG_SDELT);
1412 		num_elt = (val & D40_SREG_ELEM_PHY_ECNT_MASK)
1413 			  >> D40_SREG_ELEM_PHY_ECNT_POS;
1414 	}
1415 
1416 	return num_elt * d40c->dma_cfg.dst_info.data_width;
1417 }
1418 
1419 static bool d40_tx_is_linked(struct d40_chan *d40c)
1420 {
1421 	bool is_link;
1422 
1423 	if (chan_is_logical(d40c))
1424 		is_link = readl(&d40c->lcpa->lcsp3) &  D40_MEM_LCSP3_DLOS_MASK;
1425 	else
1426 		is_link = readl(chan_base(d40c) + D40_CHAN_REG_SDLNK)
1427 			  & D40_SREG_LNK_PHYS_LNK_MASK;
1428 
1429 	return is_link;
1430 }
1431 
1432 static int d40_pause(struct d40_chan *d40c)
1433 {
1434 	int res = 0;
1435 	unsigned long flags;
1436 
1437 	if (!d40c->busy)
1438 		return 0;
1439 
1440 	spin_lock_irqsave(&d40c->lock, flags);
1441 	pm_runtime_get_sync(d40c->base->dev);
1442 
1443 	res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1444 
1445 	pm_runtime_mark_last_busy(d40c->base->dev);
1446 	pm_runtime_put_autosuspend(d40c->base->dev);
1447 	spin_unlock_irqrestore(&d40c->lock, flags);
1448 	return res;
1449 }
1450 
1451 static int d40_resume(struct d40_chan *d40c)
1452 {
1453 	int res = 0;
1454 	unsigned long flags;
1455 
1456 	if (!d40c->busy)
1457 		return 0;
1458 
1459 	spin_lock_irqsave(&d40c->lock, flags);
1460 	pm_runtime_get_sync(d40c->base->dev);
1461 
1462 	/* If bytes left to transfer or linked tx resume job */
1463 	if (d40_residue(d40c) || d40_tx_is_linked(d40c))
1464 		res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1465 
1466 	pm_runtime_mark_last_busy(d40c->base->dev);
1467 	pm_runtime_put_autosuspend(d40c->base->dev);
1468 	spin_unlock_irqrestore(&d40c->lock, flags);
1469 	return res;
1470 }
1471 
1472 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
1473 {
1474 	struct d40_chan *d40c = container_of(tx->chan,
1475 					     struct d40_chan,
1476 					     chan);
1477 	struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
1478 	unsigned long flags;
1479 	dma_cookie_t cookie;
1480 
1481 	spin_lock_irqsave(&d40c->lock, flags);
1482 	cookie = dma_cookie_assign(tx);
1483 	d40_desc_queue(d40c, d40d);
1484 	spin_unlock_irqrestore(&d40c->lock, flags);
1485 
1486 	return cookie;
1487 }
1488 
1489 static int d40_start(struct d40_chan *d40c)
1490 {
1491 	return d40_channel_execute_command(d40c, D40_DMA_RUN);
1492 }
1493 
1494 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
1495 {
1496 	struct d40_desc *d40d;
1497 	int err;
1498 
1499 	/* Start queued jobs, if any */
1500 	d40d = d40_first_queued(d40c);
1501 
1502 	if (d40d != NULL) {
1503 		if (!d40c->busy) {
1504 			d40c->busy = true;
1505 			pm_runtime_get_sync(d40c->base->dev);
1506 		}
1507 
1508 		/* Remove from queue */
1509 		d40_desc_remove(d40d);
1510 
1511 		/* Add to active queue */
1512 		d40_desc_submit(d40c, d40d);
1513 
1514 		/* Initiate DMA job */
1515 		d40_desc_load(d40c, d40d);
1516 
1517 		/* Start dma job */
1518 		err = d40_start(d40c);
1519 
1520 		if (err)
1521 			return NULL;
1522 	}
1523 
1524 	return d40d;
1525 }
1526 
1527 /* called from interrupt context */
1528 static void dma_tc_handle(struct d40_chan *d40c)
1529 {
1530 	struct d40_desc *d40d;
1531 
1532 	/* Get first active entry from list */
1533 	d40d = d40_first_active_get(d40c);
1534 
1535 	if (d40d == NULL)
1536 		return;
1537 
1538 	if (d40d->cyclic) {
1539 		/*
1540 		 * If this was a paritially loaded list, we need to reloaded
1541 		 * it, and only when the list is completed.  We need to check
1542 		 * for done because the interrupt will hit for every link, and
1543 		 * not just the last one.
1544 		 */
1545 		if (d40d->lli_current < d40d->lli_len
1546 		    && !d40_tx_is_linked(d40c)
1547 		    && !d40_residue(d40c)) {
1548 			d40_lcla_free_all(d40c, d40d);
1549 			d40_desc_load(d40c, d40d);
1550 			(void) d40_start(d40c);
1551 
1552 			if (d40d->lli_current == d40d->lli_len)
1553 				d40d->lli_current = 0;
1554 		}
1555 	} else {
1556 		d40_lcla_free_all(d40c, d40d);
1557 
1558 		if (d40d->lli_current < d40d->lli_len) {
1559 			d40_desc_load(d40c, d40d);
1560 			/* Start dma job */
1561 			(void) d40_start(d40c);
1562 			return;
1563 		}
1564 
1565 		if (d40_queue_start(d40c) == NULL) {
1566 			d40c->busy = false;
1567 
1568 			pm_runtime_mark_last_busy(d40c->base->dev);
1569 			pm_runtime_put_autosuspend(d40c->base->dev);
1570 		}
1571 
1572 		d40_desc_remove(d40d);
1573 		d40_desc_done(d40c, d40d);
1574 	}
1575 
1576 	d40c->pending_tx++;
1577 	tasklet_schedule(&d40c->tasklet);
1578 
1579 }
1580 
1581 static void dma_tasklet(unsigned long data)
1582 {
1583 	struct d40_chan *d40c = (struct d40_chan *) data;
1584 	struct d40_desc *d40d;
1585 	unsigned long flags;
1586 	bool callback_active;
1587 	dma_async_tx_callback callback;
1588 	void *callback_param;
1589 
1590 	spin_lock_irqsave(&d40c->lock, flags);
1591 
1592 	/* Get first entry from the done list */
1593 	d40d = d40_first_done(d40c);
1594 	if (d40d == NULL) {
1595 		/* Check if we have reached here for cyclic job */
1596 		d40d = d40_first_active_get(d40c);
1597 		if (d40d == NULL || !d40d->cyclic)
1598 			goto err;
1599 	}
1600 
1601 	if (!d40d->cyclic)
1602 		dma_cookie_complete(&d40d->txd);
1603 
1604 	/*
1605 	 * If terminating a channel pending_tx is set to zero.
1606 	 * This prevents any finished active jobs to return to the client.
1607 	 */
1608 	if (d40c->pending_tx == 0) {
1609 		spin_unlock_irqrestore(&d40c->lock, flags);
1610 		return;
1611 	}
1612 
1613 	/* Callback to client */
1614 	callback_active = !!(d40d->txd.flags & DMA_PREP_INTERRUPT);
1615 	callback = d40d->txd.callback;
1616 	callback_param = d40d->txd.callback_param;
1617 
1618 	if (!d40d->cyclic) {
1619 		if (async_tx_test_ack(&d40d->txd)) {
1620 			d40_desc_remove(d40d);
1621 			d40_desc_free(d40c, d40d);
1622 		} else if (!d40d->is_in_client_list) {
1623 			d40_desc_remove(d40d);
1624 			d40_lcla_free_all(d40c, d40d);
1625 			list_add_tail(&d40d->node, &d40c->client);
1626 			d40d->is_in_client_list = true;
1627 		}
1628 	}
1629 
1630 	d40c->pending_tx--;
1631 
1632 	if (d40c->pending_tx)
1633 		tasklet_schedule(&d40c->tasklet);
1634 
1635 	spin_unlock_irqrestore(&d40c->lock, flags);
1636 
1637 	if (callback_active && callback)
1638 		callback(callback_param);
1639 
1640 	return;
1641 
1642 err:
1643 	/* Rescue manouver if receiving double interrupts */
1644 	if (d40c->pending_tx > 0)
1645 		d40c->pending_tx--;
1646 	spin_unlock_irqrestore(&d40c->lock, flags);
1647 }
1648 
1649 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1650 {
1651 	int i;
1652 	u32 idx;
1653 	u32 row;
1654 	long chan = -1;
1655 	struct d40_chan *d40c;
1656 	unsigned long flags;
1657 	struct d40_base *base = data;
1658 	u32 regs[base->gen_dmac.il_size];
1659 	struct d40_interrupt_lookup *il = base->gen_dmac.il;
1660 	u32 il_size = base->gen_dmac.il_size;
1661 
1662 	spin_lock_irqsave(&base->interrupt_lock, flags);
1663 
1664 	/* Read interrupt status of both logical and physical channels */
1665 	for (i = 0; i < il_size; i++)
1666 		regs[i] = readl(base->virtbase + il[i].src);
1667 
1668 	for (;;) {
1669 
1670 		chan = find_next_bit((unsigned long *)regs,
1671 				     BITS_PER_LONG * il_size, chan + 1);
1672 
1673 		/* No more set bits found? */
1674 		if (chan == BITS_PER_LONG * il_size)
1675 			break;
1676 
1677 		row = chan / BITS_PER_LONG;
1678 		idx = chan & (BITS_PER_LONG - 1);
1679 
1680 		if (il[row].offset == D40_PHY_CHAN)
1681 			d40c = base->lookup_phy_chans[idx];
1682 		else
1683 			d40c = base->lookup_log_chans[il[row].offset + idx];
1684 
1685 		if (!d40c) {
1686 			/*
1687 			 * No error because this can happen if something else
1688 			 * in the system is using the channel.
1689 			 */
1690 			continue;
1691 		}
1692 
1693 		/* ACK interrupt */
1694 		writel(BIT(idx), base->virtbase + il[row].clr);
1695 
1696 		spin_lock(&d40c->lock);
1697 
1698 		if (!il[row].is_error)
1699 			dma_tc_handle(d40c);
1700 		else
1701 			d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1702 				chan, il[row].offset, idx);
1703 
1704 		spin_unlock(&d40c->lock);
1705 	}
1706 
1707 	spin_unlock_irqrestore(&base->interrupt_lock, flags);
1708 
1709 	return IRQ_HANDLED;
1710 }
1711 
1712 static int d40_validate_conf(struct d40_chan *d40c,
1713 			     struct stedma40_chan_cfg *conf)
1714 {
1715 	int res = 0;
1716 	bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1717 
1718 	if (!conf->dir) {
1719 		chan_err(d40c, "Invalid direction.\n");
1720 		res = -EINVAL;
1721 	}
1722 
1723 	if ((is_log && conf->dev_type > d40c->base->num_log_chans)  ||
1724 	    (!is_log && conf->dev_type > d40c->base->num_phy_chans) ||
1725 	    (conf->dev_type < 0)) {
1726 		chan_err(d40c, "Invalid device type (%d)\n", conf->dev_type);
1727 		res = -EINVAL;
1728 	}
1729 
1730 	if (conf->dir == DMA_DEV_TO_DEV) {
1731 		/*
1732 		 * DMAC HW supports it. Will be added to this driver,
1733 		 * in case any dma client requires it.
1734 		 */
1735 		chan_err(d40c, "periph to periph not supported\n");
1736 		res = -EINVAL;
1737 	}
1738 
1739 	if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1740 	    conf->src_info.data_width !=
1741 	    d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1742 	    conf->dst_info.data_width) {
1743 		/*
1744 		 * The DMAC hardware only supports
1745 		 * src (burst x width) == dst (burst x width)
1746 		 */
1747 
1748 		chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1749 		res = -EINVAL;
1750 	}
1751 
1752 	return res;
1753 }
1754 
1755 static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1756 			       bool is_src, int log_event_line, bool is_log,
1757 			       bool *first_user)
1758 {
1759 	unsigned long flags;
1760 	spin_lock_irqsave(&phy->lock, flags);
1761 
1762 	*first_user = ((phy->allocated_src | phy->allocated_dst)
1763 			== D40_ALLOC_FREE);
1764 
1765 	if (!is_log) {
1766 		/* Physical interrupts are masked per physical full channel */
1767 		if (phy->allocated_src == D40_ALLOC_FREE &&
1768 		    phy->allocated_dst == D40_ALLOC_FREE) {
1769 			phy->allocated_dst = D40_ALLOC_PHY;
1770 			phy->allocated_src = D40_ALLOC_PHY;
1771 			goto found;
1772 		} else
1773 			goto not_found;
1774 	}
1775 
1776 	/* Logical channel */
1777 	if (is_src) {
1778 		if (phy->allocated_src == D40_ALLOC_PHY)
1779 			goto not_found;
1780 
1781 		if (phy->allocated_src == D40_ALLOC_FREE)
1782 			phy->allocated_src = D40_ALLOC_LOG_FREE;
1783 
1784 		if (!(phy->allocated_src & BIT(log_event_line))) {
1785 			phy->allocated_src |= BIT(log_event_line);
1786 			goto found;
1787 		} else
1788 			goto not_found;
1789 	} else {
1790 		if (phy->allocated_dst == D40_ALLOC_PHY)
1791 			goto not_found;
1792 
1793 		if (phy->allocated_dst == D40_ALLOC_FREE)
1794 			phy->allocated_dst = D40_ALLOC_LOG_FREE;
1795 
1796 		if (!(phy->allocated_dst & BIT(log_event_line))) {
1797 			phy->allocated_dst |= BIT(log_event_line);
1798 			goto found;
1799 		} else
1800 			goto not_found;
1801 	}
1802 
1803 not_found:
1804 	spin_unlock_irqrestore(&phy->lock, flags);
1805 	return false;
1806 found:
1807 	spin_unlock_irqrestore(&phy->lock, flags);
1808 	return true;
1809 }
1810 
1811 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1812 			       int log_event_line)
1813 {
1814 	unsigned long flags;
1815 	bool is_free = false;
1816 
1817 	spin_lock_irqsave(&phy->lock, flags);
1818 	if (!log_event_line) {
1819 		phy->allocated_dst = D40_ALLOC_FREE;
1820 		phy->allocated_src = D40_ALLOC_FREE;
1821 		is_free = true;
1822 		goto out;
1823 	}
1824 
1825 	/* Logical channel */
1826 	if (is_src) {
1827 		phy->allocated_src &= ~BIT(log_event_line);
1828 		if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1829 			phy->allocated_src = D40_ALLOC_FREE;
1830 	} else {
1831 		phy->allocated_dst &= ~BIT(log_event_line);
1832 		if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1833 			phy->allocated_dst = D40_ALLOC_FREE;
1834 	}
1835 
1836 	is_free = ((phy->allocated_src | phy->allocated_dst) ==
1837 		   D40_ALLOC_FREE);
1838 
1839 out:
1840 	spin_unlock_irqrestore(&phy->lock, flags);
1841 
1842 	return is_free;
1843 }
1844 
1845 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1846 {
1847 	int dev_type = d40c->dma_cfg.dev_type;
1848 	int event_group;
1849 	int event_line;
1850 	struct d40_phy_res *phys;
1851 	int i;
1852 	int j;
1853 	int log_num;
1854 	int num_phy_chans;
1855 	bool is_src;
1856 	bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1857 
1858 	phys = d40c->base->phy_res;
1859 	num_phy_chans = d40c->base->num_phy_chans;
1860 
1861 	if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
1862 		log_num = 2 * dev_type;
1863 		is_src = true;
1864 	} else if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
1865 		   d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1866 		/* dst event lines are used for logical memcpy */
1867 		log_num = 2 * dev_type + 1;
1868 		is_src = false;
1869 	} else
1870 		return -EINVAL;
1871 
1872 	event_group = D40_TYPE_TO_GROUP(dev_type);
1873 	event_line = D40_TYPE_TO_EVENT(dev_type);
1874 
1875 	if (!is_log) {
1876 		if (d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1877 			/* Find physical half channel */
1878 			if (d40c->dma_cfg.use_fixed_channel) {
1879 				i = d40c->dma_cfg.phy_channel;
1880 				if (d40_alloc_mask_set(&phys[i], is_src,
1881 						       0, is_log,
1882 						       first_phy_user))
1883 					goto found_phy;
1884 			} else {
1885 				for (i = 0; i < num_phy_chans; i++) {
1886 					if (d40_alloc_mask_set(&phys[i], is_src,
1887 						       0, is_log,
1888 						       first_phy_user))
1889 						goto found_phy;
1890 				}
1891 			}
1892 		} else
1893 			for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1894 				int phy_num = j  + event_group * 2;
1895 				for (i = phy_num; i < phy_num + 2; i++) {
1896 					if (d40_alloc_mask_set(&phys[i],
1897 							       is_src,
1898 							       0,
1899 							       is_log,
1900 							       first_phy_user))
1901 						goto found_phy;
1902 				}
1903 			}
1904 		return -EINVAL;
1905 found_phy:
1906 		d40c->phy_chan = &phys[i];
1907 		d40c->log_num = D40_PHY_CHAN;
1908 		goto out;
1909 	}
1910 	if (dev_type == -1)
1911 		return -EINVAL;
1912 
1913 	/* Find logical channel */
1914 	for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1915 		int phy_num = j + event_group * 2;
1916 
1917 		if (d40c->dma_cfg.use_fixed_channel) {
1918 			i = d40c->dma_cfg.phy_channel;
1919 
1920 			if ((i != phy_num) && (i != phy_num + 1)) {
1921 				dev_err(chan2dev(d40c),
1922 					"invalid fixed phy channel %d\n", i);
1923 				return -EINVAL;
1924 			}
1925 
1926 			if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1927 					       is_log, first_phy_user))
1928 				goto found_log;
1929 
1930 			dev_err(chan2dev(d40c),
1931 				"could not allocate fixed phy channel %d\n", i);
1932 			return -EINVAL;
1933 		}
1934 
1935 		/*
1936 		 * Spread logical channels across all available physical rather
1937 		 * than pack every logical channel at the first available phy
1938 		 * channels.
1939 		 */
1940 		if (is_src) {
1941 			for (i = phy_num; i < phy_num + 2; i++) {
1942 				if (d40_alloc_mask_set(&phys[i], is_src,
1943 						       event_line, is_log,
1944 						       first_phy_user))
1945 					goto found_log;
1946 			}
1947 		} else {
1948 			for (i = phy_num + 1; i >= phy_num; i--) {
1949 				if (d40_alloc_mask_set(&phys[i], is_src,
1950 						       event_line, is_log,
1951 						       first_phy_user))
1952 					goto found_log;
1953 			}
1954 		}
1955 	}
1956 	return -EINVAL;
1957 
1958 found_log:
1959 	d40c->phy_chan = &phys[i];
1960 	d40c->log_num = log_num;
1961 out:
1962 
1963 	if (is_log)
1964 		d40c->base->lookup_log_chans[d40c->log_num] = d40c;
1965 	else
1966 		d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
1967 
1968 	return 0;
1969 
1970 }
1971 
1972 static int d40_config_memcpy(struct d40_chan *d40c)
1973 {
1974 	dma_cap_mask_t cap = d40c->chan.device->cap_mask;
1975 
1976 	if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
1977 		d40c->dma_cfg = dma40_memcpy_conf_log;
1978 		d40c->dma_cfg.dev_type = dma40_memcpy_channels[d40c->chan.chan_id];
1979 
1980 		d40_log_cfg(&d40c->dma_cfg,
1981 			    &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
1982 
1983 	} else if (dma_has_cap(DMA_MEMCPY, cap) &&
1984 		   dma_has_cap(DMA_SLAVE, cap)) {
1985 		d40c->dma_cfg = dma40_memcpy_conf_phy;
1986 
1987 		/* Generate interrrupt at end of transfer or relink. */
1988 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_TIM_POS);
1989 
1990 		/* Generate interrupt on error. */
1991 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
1992 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
1993 
1994 	} else {
1995 		chan_err(d40c, "No memcpy\n");
1996 		return -EINVAL;
1997 	}
1998 
1999 	return 0;
2000 }
2001 
2002 static int d40_free_dma(struct d40_chan *d40c)
2003 {
2004 
2005 	int res = 0;
2006 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2007 	struct d40_phy_res *phy = d40c->phy_chan;
2008 	bool is_src;
2009 
2010 	/* Terminate all queued and active transfers */
2011 	d40_term_all(d40c);
2012 
2013 	if (phy == NULL) {
2014 		chan_err(d40c, "phy == null\n");
2015 		return -EINVAL;
2016 	}
2017 
2018 	if (phy->allocated_src == D40_ALLOC_FREE &&
2019 	    phy->allocated_dst == D40_ALLOC_FREE) {
2020 		chan_err(d40c, "channel already free\n");
2021 		return -EINVAL;
2022 	}
2023 
2024 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2025 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM)
2026 		is_src = false;
2027 	else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2028 		is_src = true;
2029 	else {
2030 		chan_err(d40c, "Unknown direction\n");
2031 		return -EINVAL;
2032 	}
2033 
2034 	pm_runtime_get_sync(d40c->base->dev);
2035 	res = d40_channel_execute_command(d40c, D40_DMA_STOP);
2036 	if (res) {
2037 		chan_err(d40c, "stop failed\n");
2038 		goto out;
2039 	}
2040 
2041 	d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
2042 
2043 	if (chan_is_logical(d40c))
2044 		d40c->base->lookup_log_chans[d40c->log_num] = NULL;
2045 	else
2046 		d40c->base->lookup_phy_chans[phy->num] = NULL;
2047 
2048 	if (d40c->busy) {
2049 		pm_runtime_mark_last_busy(d40c->base->dev);
2050 		pm_runtime_put_autosuspend(d40c->base->dev);
2051 	}
2052 
2053 	d40c->busy = false;
2054 	d40c->phy_chan = NULL;
2055 	d40c->configured = false;
2056 out:
2057 
2058 	pm_runtime_mark_last_busy(d40c->base->dev);
2059 	pm_runtime_put_autosuspend(d40c->base->dev);
2060 	return res;
2061 }
2062 
2063 static bool d40_is_paused(struct d40_chan *d40c)
2064 {
2065 	void __iomem *chanbase = chan_base(d40c);
2066 	bool is_paused = false;
2067 	unsigned long flags;
2068 	void __iomem *active_reg;
2069 	u32 status;
2070 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2071 
2072 	spin_lock_irqsave(&d40c->lock, flags);
2073 
2074 	if (chan_is_physical(d40c)) {
2075 		if (d40c->phy_chan->num % 2 == 0)
2076 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
2077 		else
2078 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
2079 
2080 		status = (readl(active_reg) &
2081 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
2082 			D40_CHAN_POS(d40c->phy_chan->num);
2083 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
2084 			is_paused = true;
2085 
2086 		goto _exit;
2087 	}
2088 
2089 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2090 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
2091 		status = readl(chanbase + D40_CHAN_REG_SDLNK);
2092 	} else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
2093 		status = readl(chanbase + D40_CHAN_REG_SSLNK);
2094 	} else {
2095 		chan_err(d40c, "Unknown direction\n");
2096 		goto _exit;
2097 	}
2098 
2099 	status = (status & D40_EVENTLINE_MASK(event)) >>
2100 		D40_EVENTLINE_POS(event);
2101 
2102 	if (status != D40_DMA_RUN)
2103 		is_paused = true;
2104 _exit:
2105 	spin_unlock_irqrestore(&d40c->lock, flags);
2106 	return is_paused;
2107 
2108 }
2109 
2110 static u32 stedma40_residue(struct dma_chan *chan)
2111 {
2112 	struct d40_chan *d40c =
2113 		container_of(chan, struct d40_chan, chan);
2114 	u32 bytes_left;
2115 	unsigned long flags;
2116 
2117 	spin_lock_irqsave(&d40c->lock, flags);
2118 	bytes_left = d40_residue(d40c);
2119 	spin_unlock_irqrestore(&d40c->lock, flags);
2120 
2121 	return bytes_left;
2122 }
2123 
2124 static int
2125 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
2126 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2127 		unsigned int sg_len, dma_addr_t src_dev_addr,
2128 		dma_addr_t dst_dev_addr)
2129 {
2130 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2131 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2132 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2133 	int ret;
2134 
2135 	ret = d40_log_sg_to_lli(sg_src, sg_len,
2136 				src_dev_addr,
2137 				desc->lli_log.src,
2138 				chan->log_def.lcsp1,
2139 				src_info->data_width,
2140 				dst_info->data_width);
2141 
2142 	ret = d40_log_sg_to_lli(sg_dst, sg_len,
2143 				dst_dev_addr,
2144 				desc->lli_log.dst,
2145 				chan->log_def.lcsp3,
2146 				dst_info->data_width,
2147 				src_info->data_width);
2148 
2149 	return ret < 0 ? ret : 0;
2150 }
2151 
2152 static int
2153 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2154 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2155 		unsigned int sg_len, dma_addr_t src_dev_addr,
2156 		dma_addr_t dst_dev_addr)
2157 {
2158 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2159 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2160 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2161 	unsigned long flags = 0;
2162 	int ret;
2163 
2164 	if (desc->cyclic)
2165 		flags |= LLI_CYCLIC | LLI_TERM_INT;
2166 
2167 	ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2168 				desc->lli_phy.src,
2169 				virt_to_phys(desc->lli_phy.src),
2170 				chan->src_def_cfg,
2171 				src_info, dst_info, flags);
2172 
2173 	ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2174 				desc->lli_phy.dst,
2175 				virt_to_phys(desc->lli_phy.dst),
2176 				chan->dst_def_cfg,
2177 				dst_info, src_info, flags);
2178 
2179 	dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2180 				   desc->lli_pool.size, DMA_TO_DEVICE);
2181 
2182 	return ret < 0 ? ret : 0;
2183 }
2184 
2185 static struct d40_desc *
2186 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2187 	      unsigned int sg_len, unsigned long dma_flags)
2188 {
2189 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2190 	struct d40_desc *desc;
2191 	int ret;
2192 
2193 	desc = d40_desc_get(chan);
2194 	if (!desc)
2195 		return NULL;
2196 
2197 	desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2198 					cfg->dst_info.data_width);
2199 	if (desc->lli_len < 0) {
2200 		chan_err(chan, "Unaligned size\n");
2201 		goto err;
2202 	}
2203 
2204 	ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2205 	if (ret < 0) {
2206 		chan_err(chan, "Could not allocate lli\n");
2207 		goto err;
2208 	}
2209 
2210 	desc->lli_current = 0;
2211 	desc->txd.flags = dma_flags;
2212 	desc->txd.tx_submit = d40_tx_submit;
2213 
2214 	dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2215 
2216 	return desc;
2217 
2218 err:
2219 	d40_desc_free(chan, desc);
2220 	return NULL;
2221 }
2222 
2223 static struct dma_async_tx_descriptor *
2224 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2225 	    struct scatterlist *sg_dst, unsigned int sg_len,
2226 	    enum dma_transfer_direction direction, unsigned long dma_flags)
2227 {
2228 	struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2229 	dma_addr_t src_dev_addr = 0;
2230 	dma_addr_t dst_dev_addr = 0;
2231 	struct d40_desc *desc;
2232 	unsigned long flags;
2233 	int ret;
2234 
2235 	if (!chan->phy_chan) {
2236 		chan_err(chan, "Cannot prepare unallocated channel\n");
2237 		return NULL;
2238 	}
2239 
2240 	spin_lock_irqsave(&chan->lock, flags);
2241 
2242 	desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2243 	if (desc == NULL)
2244 		goto err;
2245 
2246 	if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2247 		desc->cyclic = true;
2248 
2249 	if (direction == DMA_DEV_TO_MEM)
2250 		src_dev_addr = chan->runtime_addr;
2251 	else if (direction == DMA_MEM_TO_DEV)
2252 		dst_dev_addr = chan->runtime_addr;
2253 
2254 	if (chan_is_logical(chan))
2255 		ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2256 				      sg_len, src_dev_addr, dst_dev_addr);
2257 	else
2258 		ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2259 				      sg_len, src_dev_addr, dst_dev_addr);
2260 
2261 	if (ret) {
2262 		chan_err(chan, "Failed to prepare %s sg job: %d\n",
2263 			 chan_is_logical(chan) ? "log" : "phy", ret);
2264 		goto err;
2265 	}
2266 
2267 	/*
2268 	 * add descriptor to the prepare queue in order to be able
2269 	 * to free them later in terminate_all
2270 	 */
2271 	list_add_tail(&desc->node, &chan->prepare_queue);
2272 
2273 	spin_unlock_irqrestore(&chan->lock, flags);
2274 
2275 	return &desc->txd;
2276 
2277 err:
2278 	if (desc)
2279 		d40_desc_free(chan, desc);
2280 	spin_unlock_irqrestore(&chan->lock, flags);
2281 	return NULL;
2282 }
2283 
2284 bool stedma40_filter(struct dma_chan *chan, void *data)
2285 {
2286 	struct stedma40_chan_cfg *info = data;
2287 	struct d40_chan *d40c =
2288 		container_of(chan, struct d40_chan, chan);
2289 	int err;
2290 
2291 	if (data) {
2292 		err = d40_validate_conf(d40c, info);
2293 		if (!err)
2294 			d40c->dma_cfg = *info;
2295 	} else
2296 		err = d40_config_memcpy(d40c);
2297 
2298 	if (!err)
2299 		d40c->configured = true;
2300 
2301 	return err == 0;
2302 }
2303 EXPORT_SYMBOL(stedma40_filter);
2304 
2305 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2306 {
2307 	bool realtime = d40c->dma_cfg.realtime;
2308 	bool highprio = d40c->dma_cfg.high_priority;
2309 	u32 rtreg;
2310 	u32 event = D40_TYPE_TO_EVENT(dev_type);
2311 	u32 group = D40_TYPE_TO_GROUP(dev_type);
2312 	u32 bit = BIT(event);
2313 	u32 prioreg;
2314 	struct d40_gen_dmac *dmac = &d40c->base->gen_dmac;
2315 
2316 	rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear;
2317 	/*
2318 	 * Due to a hardware bug, in some cases a logical channel triggered by
2319 	 * a high priority destination event line can generate extra packet
2320 	 * transactions.
2321 	 *
2322 	 * The workaround is to not set the high priority level for the
2323 	 * destination event lines that trigger logical channels.
2324 	 */
2325 	if (!src && chan_is_logical(d40c))
2326 		highprio = false;
2327 
2328 	prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear;
2329 
2330 	/* Destination event lines are stored in the upper halfword */
2331 	if (!src)
2332 		bit <<= 16;
2333 
2334 	writel(bit, d40c->base->virtbase + prioreg + group * 4);
2335 	writel(bit, d40c->base->virtbase + rtreg + group * 4);
2336 }
2337 
2338 static void d40_set_prio_realtime(struct d40_chan *d40c)
2339 {
2340 	if (d40c->base->rev < 3)
2341 		return;
2342 
2343 	if ((d40c->dma_cfg.dir ==  DMA_DEV_TO_MEM) ||
2344 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2345 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, true);
2346 
2347 	if ((d40c->dma_cfg.dir ==  DMA_MEM_TO_DEV) ||
2348 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2349 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, false);
2350 }
2351 
2352 #define D40_DT_FLAGS_MODE(flags)       ((flags >> 0) & 0x1)
2353 #define D40_DT_FLAGS_DIR(flags)        ((flags >> 1) & 0x1)
2354 #define D40_DT_FLAGS_BIG_ENDIAN(flags) ((flags >> 2) & 0x1)
2355 #define D40_DT_FLAGS_FIXED_CHAN(flags) ((flags >> 3) & 0x1)
2356 #define D40_DT_FLAGS_HIGH_PRIO(flags)  ((flags >> 4) & 0x1)
2357 
2358 static struct dma_chan *d40_xlate(struct of_phandle_args *dma_spec,
2359 				  struct of_dma *ofdma)
2360 {
2361 	struct stedma40_chan_cfg cfg;
2362 	dma_cap_mask_t cap;
2363 	u32 flags;
2364 
2365 	memset(&cfg, 0, sizeof(struct stedma40_chan_cfg));
2366 
2367 	dma_cap_zero(cap);
2368 	dma_cap_set(DMA_SLAVE, cap);
2369 
2370 	cfg.dev_type = dma_spec->args[0];
2371 	flags = dma_spec->args[2];
2372 
2373 	switch (D40_DT_FLAGS_MODE(flags)) {
2374 	case 0: cfg.mode = STEDMA40_MODE_LOGICAL; break;
2375 	case 1: cfg.mode = STEDMA40_MODE_PHYSICAL; break;
2376 	}
2377 
2378 	switch (D40_DT_FLAGS_DIR(flags)) {
2379 	case 0:
2380 		cfg.dir = DMA_MEM_TO_DEV;
2381 		cfg.dst_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2382 		break;
2383 	case 1:
2384 		cfg.dir = DMA_DEV_TO_MEM;
2385 		cfg.src_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2386 		break;
2387 	}
2388 
2389 	if (D40_DT_FLAGS_FIXED_CHAN(flags)) {
2390 		cfg.phy_channel = dma_spec->args[1];
2391 		cfg.use_fixed_channel = true;
2392 	}
2393 
2394 	if (D40_DT_FLAGS_HIGH_PRIO(flags))
2395 		cfg.high_priority = true;
2396 
2397 	return dma_request_channel(cap, stedma40_filter, &cfg);
2398 }
2399 
2400 /* DMA ENGINE functions */
2401 static int d40_alloc_chan_resources(struct dma_chan *chan)
2402 {
2403 	int err;
2404 	unsigned long flags;
2405 	struct d40_chan *d40c =
2406 		container_of(chan, struct d40_chan, chan);
2407 	bool is_free_phy;
2408 	spin_lock_irqsave(&d40c->lock, flags);
2409 
2410 	dma_cookie_init(chan);
2411 
2412 	/* If no dma configuration is set use default configuration (memcpy) */
2413 	if (!d40c->configured) {
2414 		err = d40_config_memcpy(d40c);
2415 		if (err) {
2416 			chan_err(d40c, "Failed to configure memcpy channel\n");
2417 			goto fail;
2418 		}
2419 	}
2420 
2421 	err = d40_allocate_channel(d40c, &is_free_phy);
2422 	if (err) {
2423 		chan_err(d40c, "Failed to allocate channel\n");
2424 		d40c->configured = false;
2425 		goto fail;
2426 	}
2427 
2428 	pm_runtime_get_sync(d40c->base->dev);
2429 
2430 	d40_set_prio_realtime(d40c);
2431 
2432 	if (chan_is_logical(d40c)) {
2433 		if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2434 			d40c->lcpa = d40c->base->lcpa_base +
2435 				d40c->dma_cfg.dev_type * D40_LCPA_CHAN_SIZE;
2436 		else
2437 			d40c->lcpa = d40c->base->lcpa_base +
2438 				d40c->dma_cfg.dev_type *
2439 				D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2440 
2441 		/* Unmask the Global Interrupt Mask. */
2442 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2443 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2444 	}
2445 
2446 	dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2447 		 chan_is_logical(d40c) ? "logical" : "physical",
2448 		 d40c->phy_chan->num,
2449 		 d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2450 
2451 
2452 	/*
2453 	 * Only write channel configuration to the DMA if the physical
2454 	 * resource is free. In case of multiple logical channels
2455 	 * on the same physical resource, only the first write is necessary.
2456 	 */
2457 	if (is_free_phy)
2458 		d40_config_write(d40c);
2459 fail:
2460 	pm_runtime_mark_last_busy(d40c->base->dev);
2461 	pm_runtime_put_autosuspend(d40c->base->dev);
2462 	spin_unlock_irqrestore(&d40c->lock, flags);
2463 	return err;
2464 }
2465 
2466 static void d40_free_chan_resources(struct dma_chan *chan)
2467 {
2468 	struct d40_chan *d40c =
2469 		container_of(chan, struct d40_chan, chan);
2470 	int err;
2471 	unsigned long flags;
2472 
2473 	if (d40c->phy_chan == NULL) {
2474 		chan_err(d40c, "Cannot free unallocated channel\n");
2475 		return;
2476 	}
2477 
2478 	spin_lock_irqsave(&d40c->lock, flags);
2479 
2480 	err = d40_free_dma(d40c);
2481 
2482 	if (err)
2483 		chan_err(d40c, "Failed to free channel\n");
2484 	spin_unlock_irqrestore(&d40c->lock, flags);
2485 }
2486 
2487 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2488 						       dma_addr_t dst,
2489 						       dma_addr_t src,
2490 						       size_t size,
2491 						       unsigned long dma_flags)
2492 {
2493 	struct scatterlist dst_sg;
2494 	struct scatterlist src_sg;
2495 
2496 	sg_init_table(&dst_sg, 1);
2497 	sg_init_table(&src_sg, 1);
2498 
2499 	sg_dma_address(&dst_sg) = dst;
2500 	sg_dma_address(&src_sg) = src;
2501 
2502 	sg_dma_len(&dst_sg) = size;
2503 	sg_dma_len(&src_sg) = size;
2504 
2505 	return d40_prep_sg(chan, &src_sg, &dst_sg, 1, DMA_NONE, dma_flags);
2506 }
2507 
2508 static struct dma_async_tx_descriptor *
2509 d40_prep_memcpy_sg(struct dma_chan *chan,
2510 		   struct scatterlist *dst_sg, unsigned int dst_nents,
2511 		   struct scatterlist *src_sg, unsigned int src_nents,
2512 		   unsigned long dma_flags)
2513 {
2514 	if (dst_nents != src_nents)
2515 		return NULL;
2516 
2517 	return d40_prep_sg(chan, src_sg, dst_sg, src_nents, DMA_NONE, dma_flags);
2518 }
2519 
2520 static struct dma_async_tx_descriptor *
2521 d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2522 		  unsigned int sg_len, enum dma_transfer_direction direction,
2523 		  unsigned long dma_flags, void *context)
2524 {
2525 	if (!is_slave_direction(direction))
2526 		return NULL;
2527 
2528 	return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2529 }
2530 
2531 static struct dma_async_tx_descriptor *
2532 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2533 		     size_t buf_len, size_t period_len,
2534 		     enum dma_transfer_direction direction, unsigned long flags)
2535 {
2536 	unsigned int periods = buf_len / period_len;
2537 	struct dma_async_tx_descriptor *txd;
2538 	struct scatterlist *sg;
2539 	int i;
2540 
2541 	sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2542 	if (!sg)
2543 		return NULL;
2544 
2545 	for (i = 0; i < periods; i++) {
2546 		sg_dma_address(&sg[i]) = dma_addr;
2547 		sg_dma_len(&sg[i]) = period_len;
2548 		dma_addr += period_len;
2549 	}
2550 
2551 	sg[periods].offset = 0;
2552 	sg_dma_len(&sg[periods]) = 0;
2553 	sg[periods].page_link =
2554 		((unsigned long)sg | 0x01) & ~0x02;
2555 
2556 	txd = d40_prep_sg(chan, sg, sg, periods, direction,
2557 			  DMA_PREP_INTERRUPT);
2558 
2559 	kfree(sg);
2560 
2561 	return txd;
2562 }
2563 
2564 static enum dma_status d40_tx_status(struct dma_chan *chan,
2565 				     dma_cookie_t cookie,
2566 				     struct dma_tx_state *txstate)
2567 {
2568 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2569 	enum dma_status ret;
2570 
2571 	if (d40c->phy_chan == NULL) {
2572 		chan_err(d40c, "Cannot read status of unallocated channel\n");
2573 		return -EINVAL;
2574 	}
2575 
2576 	ret = dma_cookie_status(chan, cookie, txstate);
2577 	if (ret != DMA_COMPLETE)
2578 		dma_set_residue(txstate, stedma40_residue(chan));
2579 
2580 	if (d40_is_paused(d40c))
2581 		ret = DMA_PAUSED;
2582 
2583 	return ret;
2584 }
2585 
2586 static void d40_issue_pending(struct dma_chan *chan)
2587 {
2588 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2589 	unsigned long flags;
2590 
2591 	if (d40c->phy_chan == NULL) {
2592 		chan_err(d40c, "Channel is not allocated!\n");
2593 		return;
2594 	}
2595 
2596 	spin_lock_irqsave(&d40c->lock, flags);
2597 
2598 	list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2599 
2600 	/* Busy means that queued jobs are already being processed */
2601 	if (!d40c->busy)
2602 		(void) d40_queue_start(d40c);
2603 
2604 	spin_unlock_irqrestore(&d40c->lock, flags);
2605 }
2606 
2607 static void d40_terminate_all(struct dma_chan *chan)
2608 {
2609 	unsigned long flags;
2610 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2611 	int ret;
2612 
2613 	spin_lock_irqsave(&d40c->lock, flags);
2614 
2615 	pm_runtime_get_sync(d40c->base->dev);
2616 	ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2617 	if (ret)
2618 		chan_err(d40c, "Failed to stop channel\n");
2619 
2620 	d40_term_all(d40c);
2621 	pm_runtime_mark_last_busy(d40c->base->dev);
2622 	pm_runtime_put_autosuspend(d40c->base->dev);
2623 	if (d40c->busy) {
2624 		pm_runtime_mark_last_busy(d40c->base->dev);
2625 		pm_runtime_put_autosuspend(d40c->base->dev);
2626 	}
2627 	d40c->busy = false;
2628 
2629 	spin_unlock_irqrestore(&d40c->lock, flags);
2630 }
2631 
2632 static int
2633 dma40_config_to_halfchannel(struct d40_chan *d40c,
2634 			    struct stedma40_half_channel_info *info,
2635 			    u32 maxburst)
2636 {
2637 	int psize;
2638 
2639 	if (chan_is_logical(d40c)) {
2640 		if (maxburst >= 16)
2641 			psize = STEDMA40_PSIZE_LOG_16;
2642 		else if (maxburst >= 8)
2643 			psize = STEDMA40_PSIZE_LOG_8;
2644 		else if (maxburst >= 4)
2645 			psize = STEDMA40_PSIZE_LOG_4;
2646 		else
2647 			psize = STEDMA40_PSIZE_LOG_1;
2648 	} else {
2649 		if (maxburst >= 16)
2650 			psize = STEDMA40_PSIZE_PHY_16;
2651 		else if (maxburst >= 8)
2652 			psize = STEDMA40_PSIZE_PHY_8;
2653 		else if (maxburst >= 4)
2654 			psize = STEDMA40_PSIZE_PHY_4;
2655 		else
2656 			psize = STEDMA40_PSIZE_PHY_1;
2657 	}
2658 
2659 	info->psize = psize;
2660 	info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2661 
2662 	return 0;
2663 }
2664 
2665 /* Runtime reconfiguration extension */
2666 static int d40_set_runtime_config(struct dma_chan *chan,
2667 				  struct dma_slave_config *config)
2668 {
2669 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2670 	struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2671 	enum dma_slave_buswidth src_addr_width, dst_addr_width;
2672 	dma_addr_t config_addr;
2673 	u32 src_maxburst, dst_maxburst;
2674 	int ret;
2675 
2676 	src_addr_width = config->src_addr_width;
2677 	src_maxburst = config->src_maxburst;
2678 	dst_addr_width = config->dst_addr_width;
2679 	dst_maxburst = config->dst_maxburst;
2680 
2681 	if (config->direction == DMA_DEV_TO_MEM) {
2682 		config_addr = config->src_addr;
2683 
2684 		if (cfg->dir != DMA_DEV_TO_MEM)
2685 			dev_dbg(d40c->base->dev,
2686 				"channel was not configured for peripheral "
2687 				"to memory transfer (%d) overriding\n",
2688 				cfg->dir);
2689 		cfg->dir = DMA_DEV_TO_MEM;
2690 
2691 		/* Configure the memory side */
2692 		if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2693 			dst_addr_width = src_addr_width;
2694 		if (dst_maxburst == 0)
2695 			dst_maxburst = src_maxburst;
2696 
2697 	} else if (config->direction == DMA_MEM_TO_DEV) {
2698 		config_addr = config->dst_addr;
2699 
2700 		if (cfg->dir != DMA_MEM_TO_DEV)
2701 			dev_dbg(d40c->base->dev,
2702 				"channel was not configured for memory "
2703 				"to peripheral transfer (%d) overriding\n",
2704 				cfg->dir);
2705 		cfg->dir = DMA_MEM_TO_DEV;
2706 
2707 		/* Configure the memory side */
2708 		if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2709 			src_addr_width = dst_addr_width;
2710 		if (src_maxburst == 0)
2711 			src_maxburst = dst_maxburst;
2712 	} else {
2713 		dev_err(d40c->base->dev,
2714 			"unrecognized channel direction %d\n",
2715 			config->direction);
2716 		return -EINVAL;
2717 	}
2718 
2719 	if (config_addr <= 0) {
2720 		dev_err(d40c->base->dev, "no address supplied\n");
2721 		return -EINVAL;
2722 	}
2723 
2724 	if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2725 		dev_err(d40c->base->dev,
2726 			"src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2727 			src_maxburst,
2728 			src_addr_width,
2729 			dst_maxburst,
2730 			dst_addr_width);
2731 		return -EINVAL;
2732 	}
2733 
2734 	if (src_maxburst > 16) {
2735 		src_maxburst = 16;
2736 		dst_maxburst = src_maxburst * src_addr_width / dst_addr_width;
2737 	} else if (dst_maxburst > 16) {
2738 		dst_maxburst = 16;
2739 		src_maxburst = dst_maxburst * dst_addr_width / src_addr_width;
2740 	}
2741 
2742 	/* Only valid widths are; 1, 2, 4 and 8. */
2743 	if (src_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2744 	    src_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2745 	    dst_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2746 	    dst_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2747 	    !is_power_of_2(src_addr_width) ||
2748 	    !is_power_of_2(dst_addr_width))
2749 		return -EINVAL;
2750 
2751 	cfg->src_info.data_width = src_addr_width;
2752 	cfg->dst_info.data_width = dst_addr_width;
2753 
2754 	ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2755 					  src_maxburst);
2756 	if (ret)
2757 		return ret;
2758 
2759 	ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2760 					  dst_maxburst);
2761 	if (ret)
2762 		return ret;
2763 
2764 	/* Fill in register values */
2765 	if (chan_is_logical(d40c))
2766 		d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2767 	else
2768 		d40_phy_cfg(cfg, &d40c->src_def_cfg, &d40c->dst_def_cfg);
2769 
2770 	/* These settings will take precedence later */
2771 	d40c->runtime_addr = config_addr;
2772 	d40c->runtime_direction = config->direction;
2773 	dev_dbg(d40c->base->dev,
2774 		"configured channel %s for %s, data width %d/%d, "
2775 		"maxburst %d/%d elements, LE, no flow control\n",
2776 		dma_chan_name(chan),
2777 		(config->direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2778 		src_addr_width, dst_addr_width,
2779 		src_maxburst, dst_maxburst);
2780 
2781 	return 0;
2782 }
2783 
2784 static int d40_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
2785 		       unsigned long arg)
2786 {
2787 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2788 
2789 	if (d40c->phy_chan == NULL) {
2790 		chan_err(d40c, "Channel is not allocated!\n");
2791 		return -EINVAL;
2792 	}
2793 
2794 	switch (cmd) {
2795 	case DMA_TERMINATE_ALL:
2796 		d40_terminate_all(chan);
2797 		return 0;
2798 	case DMA_PAUSE:
2799 		return d40_pause(d40c);
2800 	case DMA_RESUME:
2801 		return d40_resume(d40c);
2802 	case DMA_SLAVE_CONFIG:
2803 		return d40_set_runtime_config(chan,
2804 			(struct dma_slave_config *) arg);
2805 	default:
2806 		break;
2807 	}
2808 
2809 	/* Other commands are unimplemented */
2810 	return -ENXIO;
2811 }
2812 
2813 /* Initialization functions */
2814 
2815 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2816 				 struct d40_chan *chans, int offset,
2817 				 int num_chans)
2818 {
2819 	int i = 0;
2820 	struct d40_chan *d40c;
2821 
2822 	INIT_LIST_HEAD(&dma->channels);
2823 
2824 	for (i = offset; i < offset + num_chans; i++) {
2825 		d40c = &chans[i];
2826 		d40c->base = base;
2827 		d40c->chan.device = dma;
2828 
2829 		spin_lock_init(&d40c->lock);
2830 
2831 		d40c->log_num = D40_PHY_CHAN;
2832 
2833 		INIT_LIST_HEAD(&d40c->done);
2834 		INIT_LIST_HEAD(&d40c->active);
2835 		INIT_LIST_HEAD(&d40c->queue);
2836 		INIT_LIST_HEAD(&d40c->pending_queue);
2837 		INIT_LIST_HEAD(&d40c->client);
2838 		INIT_LIST_HEAD(&d40c->prepare_queue);
2839 
2840 		tasklet_init(&d40c->tasklet, dma_tasklet,
2841 			     (unsigned long) d40c);
2842 
2843 		list_add_tail(&d40c->chan.device_node,
2844 			      &dma->channels);
2845 	}
2846 }
2847 
2848 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2849 {
2850 	if (dma_has_cap(DMA_SLAVE, dev->cap_mask))
2851 		dev->device_prep_slave_sg = d40_prep_slave_sg;
2852 
2853 	if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2854 		dev->device_prep_dma_memcpy = d40_prep_memcpy;
2855 
2856 		/*
2857 		 * This controller can only access address at even
2858 		 * 32bit boundaries, i.e. 2^2
2859 		 */
2860 		dev->copy_align = 2;
2861 	}
2862 
2863 	if (dma_has_cap(DMA_SG, dev->cap_mask))
2864 		dev->device_prep_dma_sg = d40_prep_memcpy_sg;
2865 
2866 	if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2867 		dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2868 
2869 	dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2870 	dev->device_free_chan_resources = d40_free_chan_resources;
2871 	dev->device_issue_pending = d40_issue_pending;
2872 	dev->device_tx_status = d40_tx_status;
2873 	dev->device_control = d40_control;
2874 	dev->dev = base->dev;
2875 }
2876 
2877 static int __init d40_dmaengine_init(struct d40_base *base,
2878 				     int num_reserved_chans)
2879 {
2880 	int err ;
2881 
2882 	d40_chan_init(base, &base->dma_slave, base->log_chans,
2883 		      0, base->num_log_chans);
2884 
2885 	dma_cap_zero(base->dma_slave.cap_mask);
2886 	dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2887 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2888 
2889 	d40_ops_init(base, &base->dma_slave);
2890 
2891 	err = dma_async_device_register(&base->dma_slave);
2892 
2893 	if (err) {
2894 		d40_err(base->dev, "Failed to register slave channels\n");
2895 		goto failure1;
2896 	}
2897 
2898 	d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2899 		      base->num_log_chans, base->num_memcpy_chans);
2900 
2901 	dma_cap_zero(base->dma_memcpy.cap_mask);
2902 	dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2903 	dma_cap_set(DMA_SG, base->dma_memcpy.cap_mask);
2904 
2905 	d40_ops_init(base, &base->dma_memcpy);
2906 
2907 	err = dma_async_device_register(&base->dma_memcpy);
2908 
2909 	if (err) {
2910 		d40_err(base->dev,
2911 			"Failed to regsiter memcpy only channels\n");
2912 		goto failure2;
2913 	}
2914 
2915 	d40_chan_init(base, &base->dma_both, base->phy_chans,
2916 		      0, num_reserved_chans);
2917 
2918 	dma_cap_zero(base->dma_both.cap_mask);
2919 	dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2920 	dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2921 	dma_cap_set(DMA_SG, base->dma_both.cap_mask);
2922 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2923 
2924 	d40_ops_init(base, &base->dma_both);
2925 	err = dma_async_device_register(&base->dma_both);
2926 
2927 	if (err) {
2928 		d40_err(base->dev,
2929 			"Failed to register logical and physical capable channels\n");
2930 		goto failure3;
2931 	}
2932 	return 0;
2933 failure3:
2934 	dma_async_device_unregister(&base->dma_memcpy);
2935 failure2:
2936 	dma_async_device_unregister(&base->dma_slave);
2937 failure1:
2938 	return err;
2939 }
2940 
2941 /* Suspend resume functionality */
2942 #ifdef CONFIG_PM_SLEEP
2943 static int dma40_suspend(struct device *dev)
2944 {
2945 	struct platform_device *pdev = to_platform_device(dev);
2946 	struct d40_base *base = platform_get_drvdata(pdev);
2947 	int ret;
2948 
2949 	ret = pm_runtime_force_suspend(dev);
2950 	if (ret)
2951 		return ret;
2952 
2953 	if (base->lcpa_regulator)
2954 		ret = regulator_disable(base->lcpa_regulator);
2955 	return ret;
2956 }
2957 
2958 static int dma40_resume(struct device *dev)
2959 {
2960 	struct platform_device *pdev = to_platform_device(dev);
2961 	struct d40_base *base = platform_get_drvdata(pdev);
2962 	int ret = 0;
2963 
2964 	if (base->lcpa_regulator) {
2965 		ret = regulator_enable(base->lcpa_regulator);
2966 		if (ret)
2967 			return ret;
2968 	}
2969 
2970 	return pm_runtime_force_resume(dev);
2971 }
2972 #endif
2973 
2974 #ifdef CONFIG_PM
2975 static void dma40_backup(void __iomem *baseaddr, u32 *backup,
2976 			 u32 *regaddr, int num, bool save)
2977 {
2978 	int i;
2979 
2980 	for (i = 0; i < num; i++) {
2981 		void __iomem *addr = baseaddr + regaddr[i];
2982 
2983 		if (save)
2984 			backup[i] = readl_relaxed(addr);
2985 		else
2986 			writel_relaxed(backup[i], addr);
2987 	}
2988 }
2989 
2990 static void d40_save_restore_registers(struct d40_base *base, bool save)
2991 {
2992 	int i;
2993 
2994 	/* Save/Restore channel specific registers */
2995 	for (i = 0; i < base->num_phy_chans; i++) {
2996 		void __iomem *addr;
2997 		int idx;
2998 
2999 		if (base->phy_res[i].reserved)
3000 			continue;
3001 
3002 		addr = base->virtbase + D40_DREG_PCBASE + i * D40_DREG_PCDELTA;
3003 		idx = i * ARRAY_SIZE(d40_backup_regs_chan);
3004 
3005 		dma40_backup(addr, &base->reg_val_backup_chan[idx],
3006 			     d40_backup_regs_chan,
3007 			     ARRAY_SIZE(d40_backup_regs_chan),
3008 			     save);
3009 	}
3010 
3011 	/* Save/Restore global registers */
3012 	dma40_backup(base->virtbase, base->reg_val_backup,
3013 		     d40_backup_regs, ARRAY_SIZE(d40_backup_regs),
3014 		     save);
3015 
3016 	/* Save/Restore registers only existing on dma40 v3 and later */
3017 	if (base->gen_dmac.backup)
3018 		dma40_backup(base->virtbase, base->reg_val_backup_v4,
3019 			     base->gen_dmac.backup,
3020 			base->gen_dmac.backup_size,
3021 			save);
3022 }
3023 
3024 static int dma40_runtime_suspend(struct device *dev)
3025 {
3026 	struct platform_device *pdev = to_platform_device(dev);
3027 	struct d40_base *base = platform_get_drvdata(pdev);
3028 
3029 	d40_save_restore_registers(base, true);
3030 
3031 	/* Don't disable/enable clocks for v1 due to HW bugs */
3032 	if (base->rev != 1)
3033 		writel_relaxed(base->gcc_pwr_off_mask,
3034 			       base->virtbase + D40_DREG_GCC);
3035 
3036 	return 0;
3037 }
3038 
3039 static int dma40_runtime_resume(struct device *dev)
3040 {
3041 	struct platform_device *pdev = to_platform_device(dev);
3042 	struct d40_base *base = platform_get_drvdata(pdev);
3043 
3044 	d40_save_restore_registers(base, false);
3045 
3046 	writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
3047 		       base->virtbase + D40_DREG_GCC);
3048 	return 0;
3049 }
3050 #endif
3051 
3052 static const struct dev_pm_ops dma40_pm_ops = {
3053 	SET_LATE_SYSTEM_SLEEP_PM_OPS(dma40_suspend, dma40_resume)
3054 	SET_PM_RUNTIME_PM_OPS(dma40_runtime_suspend,
3055 				dma40_runtime_resume,
3056 				NULL)
3057 };
3058 
3059 /* Initialization functions. */
3060 
3061 static int __init d40_phy_res_init(struct d40_base *base)
3062 {
3063 	int i;
3064 	int num_phy_chans_avail = 0;
3065 	u32 val[2];
3066 	int odd_even_bit = -2;
3067 	int gcc = D40_DREG_GCC_ENA;
3068 
3069 	val[0] = readl(base->virtbase + D40_DREG_PRSME);
3070 	val[1] = readl(base->virtbase + D40_DREG_PRSMO);
3071 
3072 	for (i = 0; i < base->num_phy_chans; i++) {
3073 		base->phy_res[i].num = i;
3074 		odd_even_bit += 2 * ((i % 2) == 0);
3075 		if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
3076 			/* Mark security only channels as occupied */
3077 			base->phy_res[i].allocated_src = D40_ALLOC_PHY;
3078 			base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
3079 			base->phy_res[i].reserved = true;
3080 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3081 						       D40_DREG_GCC_SRC);
3082 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3083 						       D40_DREG_GCC_DST);
3084 
3085 
3086 		} else {
3087 			base->phy_res[i].allocated_src = D40_ALLOC_FREE;
3088 			base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
3089 			base->phy_res[i].reserved = false;
3090 			num_phy_chans_avail++;
3091 		}
3092 		spin_lock_init(&base->phy_res[i].lock);
3093 	}
3094 
3095 	/* Mark disabled channels as occupied */
3096 	for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
3097 		int chan = base->plat_data->disabled_channels[i];
3098 
3099 		base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
3100 		base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
3101 		base->phy_res[chan].reserved = true;
3102 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3103 					       D40_DREG_GCC_SRC);
3104 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3105 					       D40_DREG_GCC_DST);
3106 		num_phy_chans_avail--;
3107 	}
3108 
3109 	/* Mark soft_lli channels */
3110 	for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) {
3111 		int chan = base->plat_data->soft_lli_chans[i];
3112 
3113 		base->phy_res[chan].use_soft_lli = true;
3114 	}
3115 
3116 	dev_info(base->dev, "%d of %d physical DMA channels available\n",
3117 		 num_phy_chans_avail, base->num_phy_chans);
3118 
3119 	/* Verify settings extended vs standard */
3120 	val[0] = readl(base->virtbase + D40_DREG_PRTYP);
3121 
3122 	for (i = 0; i < base->num_phy_chans; i++) {
3123 
3124 		if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
3125 		    (val[0] & 0x3) != 1)
3126 			dev_info(base->dev,
3127 				 "[%s] INFO: channel %d is misconfigured (%d)\n",
3128 				 __func__, i, val[0] & 0x3);
3129 
3130 		val[0] = val[0] >> 2;
3131 	}
3132 
3133 	/*
3134 	 * To keep things simple, Enable all clocks initially.
3135 	 * The clocks will get managed later post channel allocation.
3136 	 * The clocks for the event lines on which reserved channels exists
3137 	 * are not managed here.
3138 	 */
3139 	writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3140 	base->gcc_pwr_off_mask = gcc;
3141 
3142 	return num_phy_chans_avail;
3143 }
3144 
3145 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
3146 {
3147 	struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3148 	struct clk *clk = NULL;
3149 	void __iomem *virtbase = NULL;
3150 	struct resource *res = NULL;
3151 	struct d40_base *base = NULL;
3152 	int num_log_chans = 0;
3153 	int num_phy_chans;
3154 	int num_memcpy_chans;
3155 	int clk_ret = -EINVAL;
3156 	int i;
3157 	u32 pid;
3158 	u32 cid;
3159 	u8 rev;
3160 
3161 	clk = clk_get(&pdev->dev, NULL);
3162 	if (IS_ERR(clk)) {
3163 		d40_err(&pdev->dev, "No matching clock found\n");
3164 		goto failure;
3165 	}
3166 
3167 	clk_ret = clk_prepare_enable(clk);
3168 	if (clk_ret) {
3169 		d40_err(&pdev->dev, "Failed to prepare/enable clock\n");
3170 		goto failure;
3171 	}
3172 
3173 	/* Get IO for DMAC base address */
3174 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
3175 	if (!res)
3176 		goto failure;
3177 
3178 	if (request_mem_region(res->start, resource_size(res),
3179 			       D40_NAME " I/O base") == NULL)
3180 		goto failure;
3181 
3182 	virtbase = ioremap(res->start, resource_size(res));
3183 	if (!virtbase)
3184 		goto failure;
3185 
3186 	/* This is just a regular AMBA PrimeCell ID actually */
3187 	for (pid = 0, i = 0; i < 4; i++)
3188 		pid |= (readl(virtbase + resource_size(res) - 0x20 + 4 * i)
3189 			& 255) << (i * 8);
3190 	for (cid = 0, i = 0; i < 4; i++)
3191 		cid |= (readl(virtbase + resource_size(res) - 0x10 + 4 * i)
3192 			& 255) << (i * 8);
3193 
3194 	if (cid != AMBA_CID) {
3195 		d40_err(&pdev->dev, "Unknown hardware! No PrimeCell ID\n");
3196 		goto failure;
3197 	}
3198 	if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
3199 		d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
3200 			AMBA_MANF_BITS(pid),
3201 			AMBA_VENDOR_ST);
3202 		goto failure;
3203 	}
3204 	/*
3205 	 * HW revision:
3206 	 * DB8500ed has revision 0
3207 	 * ? has revision 1
3208 	 * DB8500v1 has revision 2
3209 	 * DB8500v2 has revision 3
3210 	 * AP9540v1 has revision 4
3211 	 * DB8540v1 has revision 4
3212 	 */
3213 	rev = AMBA_REV_BITS(pid);
3214 	if (rev < 2) {
3215 		d40_err(&pdev->dev, "hardware revision: %d is not supported", rev);
3216 		goto failure;
3217 	}
3218 
3219 	/* The number of physical channels on this HW */
3220 	if (plat_data->num_of_phy_chans)
3221 		num_phy_chans = plat_data->num_of_phy_chans;
3222 	else
3223 		num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
3224 
3225 	/* The number of channels used for memcpy */
3226 	if (plat_data->num_of_memcpy_chans)
3227 		num_memcpy_chans = plat_data->num_of_memcpy_chans;
3228 	else
3229 		num_memcpy_chans = ARRAY_SIZE(dma40_memcpy_channels);
3230 
3231 	num_log_chans = num_phy_chans * D40_MAX_LOG_CHAN_PER_PHY;
3232 
3233 	dev_info(&pdev->dev,
3234 		 "hardware rev: %d @ %pa with %d physical and %d logical channels\n",
3235 		 rev, &res->start, num_phy_chans, num_log_chans);
3236 
3237 	base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
3238 		       (num_phy_chans + num_log_chans + num_memcpy_chans) *
3239 		       sizeof(struct d40_chan), GFP_KERNEL);
3240 
3241 	if (base == NULL) {
3242 		d40_err(&pdev->dev, "Out of memory\n");
3243 		goto failure;
3244 	}
3245 
3246 	base->rev = rev;
3247 	base->clk = clk;
3248 	base->num_memcpy_chans = num_memcpy_chans;
3249 	base->num_phy_chans = num_phy_chans;
3250 	base->num_log_chans = num_log_chans;
3251 	base->phy_start = res->start;
3252 	base->phy_size = resource_size(res);
3253 	base->virtbase = virtbase;
3254 	base->plat_data = plat_data;
3255 	base->dev = &pdev->dev;
3256 	base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3257 	base->log_chans = &base->phy_chans[num_phy_chans];
3258 
3259 	if (base->plat_data->num_of_phy_chans == 14) {
3260 		base->gen_dmac.backup = d40_backup_regs_v4b;
3261 		base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
3262 		base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
3263 		base->gen_dmac.interrupt_clear = D40_DREG_CPCICR;
3264 		base->gen_dmac.realtime_en = D40_DREG_CRSEG1;
3265 		base->gen_dmac.realtime_clear = D40_DREG_CRCEG1;
3266 		base->gen_dmac.high_prio_en = D40_DREG_CPSEG1;
3267 		base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1;
3268 		base->gen_dmac.il = il_v4b;
3269 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4b);
3270 		base->gen_dmac.init_reg = dma_init_reg_v4b;
3271 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
3272 	} else {
3273 		if (base->rev >= 3) {
3274 			base->gen_dmac.backup = d40_backup_regs_v4a;
3275 			base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
3276 		}
3277 		base->gen_dmac.interrupt_en = D40_DREG_PCMIS;
3278 		base->gen_dmac.interrupt_clear = D40_DREG_PCICR;
3279 		base->gen_dmac.realtime_en = D40_DREG_RSEG1;
3280 		base->gen_dmac.realtime_clear = D40_DREG_RCEG1;
3281 		base->gen_dmac.high_prio_en = D40_DREG_PSEG1;
3282 		base->gen_dmac.high_prio_clear = D40_DREG_PCEG1;
3283 		base->gen_dmac.il = il_v4a;
3284 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4a);
3285 		base->gen_dmac.init_reg = dma_init_reg_v4a;
3286 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a);
3287 	}
3288 
3289 	base->phy_res = kzalloc(num_phy_chans * sizeof(struct d40_phy_res),
3290 				GFP_KERNEL);
3291 	if (!base->phy_res)
3292 		goto failure;
3293 
3294 	base->lookup_phy_chans = kzalloc(num_phy_chans *
3295 					 sizeof(struct d40_chan *),
3296 					 GFP_KERNEL);
3297 	if (!base->lookup_phy_chans)
3298 		goto failure;
3299 
3300 	base->lookup_log_chans = kzalloc(num_log_chans *
3301 					 sizeof(struct d40_chan *),
3302 					 GFP_KERNEL);
3303 	if (!base->lookup_log_chans)
3304 		goto failure;
3305 
3306 	base->reg_val_backup_chan = kmalloc(base->num_phy_chans *
3307 					    sizeof(d40_backup_regs_chan),
3308 					    GFP_KERNEL);
3309 	if (!base->reg_val_backup_chan)
3310 		goto failure;
3311 
3312 	base->lcla_pool.alloc_map =
3313 		kzalloc(num_phy_chans * sizeof(struct d40_desc *)
3314 			* D40_LCLA_LINK_PER_EVENT_GRP, GFP_KERNEL);
3315 	if (!base->lcla_pool.alloc_map)
3316 		goto failure;
3317 
3318 	base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3319 					    0, SLAB_HWCACHE_ALIGN,
3320 					    NULL);
3321 	if (base->desc_slab == NULL)
3322 		goto failure;
3323 
3324 	return base;
3325 
3326 failure:
3327 	if (!clk_ret)
3328 		clk_disable_unprepare(clk);
3329 	if (!IS_ERR(clk))
3330 		clk_put(clk);
3331 	if (virtbase)
3332 		iounmap(virtbase);
3333 	if (res)
3334 		release_mem_region(res->start,
3335 				   resource_size(res));
3336 	if (virtbase)
3337 		iounmap(virtbase);
3338 
3339 	if (base) {
3340 		kfree(base->lcla_pool.alloc_map);
3341 		kfree(base->reg_val_backup_chan);
3342 		kfree(base->lookup_log_chans);
3343 		kfree(base->lookup_phy_chans);
3344 		kfree(base->phy_res);
3345 		kfree(base);
3346 	}
3347 
3348 	return NULL;
3349 }
3350 
3351 static void __init d40_hw_init(struct d40_base *base)
3352 {
3353 
3354 	int i;
3355 	u32 prmseo[2] = {0, 0};
3356 	u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3357 	u32 pcmis = 0;
3358 	u32 pcicr = 0;
3359 	struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg;
3360 	u32 reg_size = base->gen_dmac.init_reg_size;
3361 
3362 	for (i = 0; i < reg_size; i++)
3363 		writel(dma_init_reg[i].val,
3364 		       base->virtbase + dma_init_reg[i].reg);
3365 
3366 	/* Configure all our dma channels to default settings */
3367 	for (i = 0; i < base->num_phy_chans; i++) {
3368 
3369 		activeo[i % 2] = activeo[i % 2] << 2;
3370 
3371 		if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3372 		    == D40_ALLOC_PHY) {
3373 			activeo[i % 2] |= 3;
3374 			continue;
3375 		}
3376 
3377 		/* Enable interrupt # */
3378 		pcmis = (pcmis << 1) | 1;
3379 
3380 		/* Clear interrupt # */
3381 		pcicr = (pcicr << 1) | 1;
3382 
3383 		/* Set channel to physical mode */
3384 		prmseo[i % 2] = prmseo[i % 2] << 2;
3385 		prmseo[i % 2] |= 1;
3386 
3387 	}
3388 
3389 	writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3390 	writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3391 	writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3392 	writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3393 
3394 	/* Write which interrupt to enable */
3395 	writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en);
3396 
3397 	/* Write which interrupt to clear */
3398 	writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear);
3399 
3400 	/* These are __initdata and cannot be accessed after init */
3401 	base->gen_dmac.init_reg = NULL;
3402 	base->gen_dmac.init_reg_size = 0;
3403 }
3404 
3405 static int __init d40_lcla_allocate(struct d40_base *base)
3406 {
3407 	struct d40_lcla_pool *pool = &base->lcla_pool;
3408 	unsigned long *page_list;
3409 	int i, j;
3410 	int ret = 0;
3411 
3412 	/*
3413 	 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3414 	 * To full fill this hardware requirement without wasting 256 kb
3415 	 * we allocate pages until we get an aligned one.
3416 	 */
3417 	page_list = kmalloc(sizeof(unsigned long) * MAX_LCLA_ALLOC_ATTEMPTS,
3418 			    GFP_KERNEL);
3419 
3420 	if (!page_list) {
3421 		ret = -ENOMEM;
3422 		goto failure;
3423 	}
3424 
3425 	/* Calculating how many pages that are required */
3426 	base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3427 
3428 	for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3429 		page_list[i] = __get_free_pages(GFP_KERNEL,
3430 						base->lcla_pool.pages);
3431 		if (!page_list[i]) {
3432 
3433 			d40_err(base->dev, "Failed to allocate %d pages.\n",
3434 				base->lcla_pool.pages);
3435 
3436 			for (j = 0; j < i; j++)
3437 				free_pages(page_list[j], base->lcla_pool.pages);
3438 			goto failure;
3439 		}
3440 
3441 		if ((virt_to_phys((void *)page_list[i]) &
3442 		     (LCLA_ALIGNMENT - 1)) == 0)
3443 			break;
3444 	}
3445 
3446 	for (j = 0; j < i; j++)
3447 		free_pages(page_list[j], base->lcla_pool.pages);
3448 
3449 	if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3450 		base->lcla_pool.base = (void *)page_list[i];
3451 	} else {
3452 		/*
3453 		 * After many attempts and no succees with finding the correct
3454 		 * alignment, try with allocating a big buffer.
3455 		 */
3456 		dev_warn(base->dev,
3457 			 "[%s] Failed to get %d pages @ 18 bit align.\n",
3458 			 __func__, base->lcla_pool.pages);
3459 		base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3460 							 base->num_phy_chans +
3461 							 LCLA_ALIGNMENT,
3462 							 GFP_KERNEL);
3463 		if (!base->lcla_pool.base_unaligned) {
3464 			ret = -ENOMEM;
3465 			goto failure;
3466 		}
3467 
3468 		base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3469 						 LCLA_ALIGNMENT);
3470 	}
3471 
3472 	pool->dma_addr = dma_map_single(base->dev, pool->base,
3473 					SZ_1K * base->num_phy_chans,
3474 					DMA_TO_DEVICE);
3475 	if (dma_mapping_error(base->dev, pool->dma_addr)) {
3476 		pool->dma_addr = 0;
3477 		ret = -ENOMEM;
3478 		goto failure;
3479 	}
3480 
3481 	writel(virt_to_phys(base->lcla_pool.base),
3482 	       base->virtbase + D40_DREG_LCLA);
3483 failure:
3484 	kfree(page_list);
3485 	return ret;
3486 }
3487 
3488 static int __init d40_of_probe(struct platform_device *pdev,
3489 			       struct device_node *np)
3490 {
3491 	struct stedma40_platform_data *pdata;
3492 	int num_phy = 0, num_memcpy = 0, num_disabled = 0;
3493 	const __be32 *list;
3494 
3495 	pdata = devm_kzalloc(&pdev->dev,
3496 			     sizeof(struct stedma40_platform_data),
3497 			     GFP_KERNEL);
3498 	if (!pdata)
3499 		return -ENOMEM;
3500 
3501 	/* If absent this value will be obtained from h/w. */
3502 	of_property_read_u32(np, "dma-channels", &num_phy);
3503 	if (num_phy > 0)
3504 		pdata->num_of_phy_chans = num_phy;
3505 
3506 	list = of_get_property(np, "memcpy-channels", &num_memcpy);
3507 	num_memcpy /= sizeof(*list);
3508 
3509 	if (num_memcpy > D40_MEMCPY_MAX_CHANS || num_memcpy <= 0) {
3510 		d40_err(&pdev->dev,
3511 			"Invalid number of memcpy channels specified (%d)\n",
3512 			num_memcpy);
3513 		return -EINVAL;
3514 	}
3515 	pdata->num_of_memcpy_chans = num_memcpy;
3516 
3517 	of_property_read_u32_array(np, "memcpy-channels",
3518 				   dma40_memcpy_channels,
3519 				   num_memcpy);
3520 
3521 	list = of_get_property(np, "disabled-channels", &num_disabled);
3522 	num_disabled /= sizeof(*list);
3523 
3524 	if (num_disabled >= STEDMA40_MAX_PHYS || num_disabled < 0) {
3525 		d40_err(&pdev->dev,
3526 			"Invalid number of disabled channels specified (%d)\n",
3527 			num_disabled);
3528 		return -EINVAL;
3529 	}
3530 
3531 	of_property_read_u32_array(np, "disabled-channels",
3532 				   pdata->disabled_channels,
3533 				   num_disabled);
3534 	pdata->disabled_channels[num_disabled] = -1;
3535 
3536 	pdev->dev.platform_data = pdata;
3537 
3538 	return 0;
3539 }
3540 
3541 static int __init d40_probe(struct platform_device *pdev)
3542 {
3543 	struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3544 	struct device_node *np = pdev->dev.of_node;
3545 	int ret = -ENOENT;
3546 	struct d40_base *base = NULL;
3547 	struct resource *res = NULL;
3548 	int num_reserved_chans;
3549 	u32 val;
3550 
3551 	if (!plat_data) {
3552 		if (np) {
3553 			if(d40_of_probe(pdev, np)) {
3554 				ret = -ENOMEM;
3555 				goto failure;
3556 			}
3557 		} else {
3558 			d40_err(&pdev->dev, "No pdata or Device Tree provided\n");
3559 			goto failure;
3560 		}
3561 	}
3562 
3563 	base = d40_hw_detect_init(pdev);
3564 	if (!base)
3565 		goto failure;
3566 
3567 	num_reserved_chans = d40_phy_res_init(base);
3568 
3569 	platform_set_drvdata(pdev, base);
3570 
3571 	spin_lock_init(&base->interrupt_lock);
3572 	spin_lock_init(&base->execmd_lock);
3573 
3574 	/* Get IO for logical channel parameter address */
3575 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
3576 	if (!res) {
3577 		ret = -ENOENT;
3578 		d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
3579 		goto failure;
3580 	}
3581 	base->lcpa_size = resource_size(res);
3582 	base->phy_lcpa = res->start;
3583 
3584 	if (request_mem_region(res->start, resource_size(res),
3585 			       D40_NAME " I/O lcpa") == NULL) {
3586 		ret = -EBUSY;
3587 		d40_err(&pdev->dev, "Failed to request LCPA region %pR\n", res);
3588 		goto failure;
3589 	}
3590 
3591 	/* We make use of ESRAM memory for this. */
3592 	val = readl(base->virtbase + D40_DREG_LCPA);
3593 	if (res->start != val && val != 0) {
3594 		dev_warn(&pdev->dev,
3595 			 "[%s] Mismatch LCPA dma 0x%x, def %pa\n",
3596 			 __func__, val, &res->start);
3597 	} else
3598 		writel(res->start, base->virtbase + D40_DREG_LCPA);
3599 
3600 	base->lcpa_base = ioremap(res->start, resource_size(res));
3601 	if (!base->lcpa_base) {
3602 		ret = -ENOMEM;
3603 		d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
3604 		goto failure;
3605 	}
3606 	/* If lcla has to be located in ESRAM we don't need to allocate */
3607 	if (base->plat_data->use_esram_lcla) {
3608 		res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3609 							"lcla_esram");
3610 		if (!res) {
3611 			ret = -ENOENT;
3612 			d40_err(&pdev->dev,
3613 				"No \"lcla_esram\" memory resource\n");
3614 			goto failure;
3615 		}
3616 		base->lcla_pool.base = ioremap(res->start,
3617 						resource_size(res));
3618 		if (!base->lcla_pool.base) {
3619 			ret = -ENOMEM;
3620 			d40_err(&pdev->dev, "Failed to ioremap LCLA region\n");
3621 			goto failure;
3622 		}
3623 		writel(res->start, base->virtbase + D40_DREG_LCLA);
3624 
3625 	} else {
3626 		ret = d40_lcla_allocate(base);
3627 		if (ret) {
3628 			d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
3629 			goto failure;
3630 		}
3631 	}
3632 
3633 	spin_lock_init(&base->lcla_pool.lock);
3634 
3635 	base->irq = platform_get_irq(pdev, 0);
3636 
3637 	ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3638 	if (ret) {
3639 		d40_err(&pdev->dev, "No IRQ defined\n");
3640 		goto failure;
3641 	}
3642 
3643 	if (base->plat_data->use_esram_lcla) {
3644 
3645 		base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3646 		if (IS_ERR(base->lcpa_regulator)) {
3647 			d40_err(&pdev->dev, "Failed to get lcpa_regulator\n");
3648 			ret = PTR_ERR(base->lcpa_regulator);
3649 			base->lcpa_regulator = NULL;
3650 			goto failure;
3651 		}
3652 
3653 		ret = regulator_enable(base->lcpa_regulator);
3654 		if (ret) {
3655 			d40_err(&pdev->dev,
3656 				"Failed to enable lcpa_regulator\n");
3657 			regulator_put(base->lcpa_regulator);
3658 			base->lcpa_regulator = NULL;
3659 			goto failure;
3660 		}
3661 	}
3662 
3663 	writel_relaxed(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3664 
3665 	pm_runtime_irq_safe(base->dev);
3666 	pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3667 	pm_runtime_use_autosuspend(base->dev);
3668 	pm_runtime_mark_last_busy(base->dev);
3669 	pm_runtime_set_active(base->dev);
3670 	pm_runtime_enable(base->dev);
3671 
3672 	ret = d40_dmaengine_init(base, num_reserved_chans);
3673 	if (ret)
3674 		goto failure;
3675 
3676 	base->dev->dma_parms = &base->dma_parms;
3677 	ret = dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
3678 	if (ret) {
3679 		d40_err(&pdev->dev, "Failed to set dma max seg size\n");
3680 		goto failure;
3681 	}
3682 
3683 	d40_hw_init(base);
3684 
3685 	if (np) {
3686 		ret = of_dma_controller_register(np, d40_xlate, NULL);
3687 		if (ret)
3688 			dev_err(&pdev->dev,
3689 				"could not register of_dma_controller\n");
3690 	}
3691 
3692 	dev_info(base->dev, "initialized\n");
3693 	return 0;
3694 
3695 failure:
3696 	if (base) {
3697 		if (base->desc_slab)
3698 			kmem_cache_destroy(base->desc_slab);
3699 		if (base->virtbase)
3700 			iounmap(base->virtbase);
3701 
3702 		if (base->lcla_pool.base && base->plat_data->use_esram_lcla) {
3703 			iounmap(base->lcla_pool.base);
3704 			base->lcla_pool.base = NULL;
3705 		}
3706 
3707 		if (base->lcla_pool.dma_addr)
3708 			dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3709 					 SZ_1K * base->num_phy_chans,
3710 					 DMA_TO_DEVICE);
3711 
3712 		if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3713 			free_pages((unsigned long)base->lcla_pool.base,
3714 				   base->lcla_pool.pages);
3715 
3716 		kfree(base->lcla_pool.base_unaligned);
3717 
3718 		if (base->phy_lcpa)
3719 			release_mem_region(base->phy_lcpa,
3720 					   base->lcpa_size);
3721 		if (base->phy_start)
3722 			release_mem_region(base->phy_start,
3723 					   base->phy_size);
3724 		if (base->clk) {
3725 			clk_disable_unprepare(base->clk);
3726 			clk_put(base->clk);
3727 		}
3728 
3729 		if (base->lcpa_regulator) {
3730 			regulator_disable(base->lcpa_regulator);
3731 			regulator_put(base->lcpa_regulator);
3732 		}
3733 
3734 		kfree(base->lcla_pool.alloc_map);
3735 		kfree(base->lookup_log_chans);
3736 		kfree(base->lookup_phy_chans);
3737 		kfree(base->phy_res);
3738 		kfree(base);
3739 	}
3740 
3741 	d40_err(&pdev->dev, "probe failed\n");
3742 	return ret;
3743 }
3744 
3745 static const struct of_device_id d40_match[] = {
3746         { .compatible = "stericsson,dma40", },
3747         {}
3748 };
3749 
3750 static struct platform_driver d40_driver = {
3751 	.driver = {
3752 		.owner = THIS_MODULE,
3753 		.name  = D40_NAME,
3754 		.pm = &dma40_pm_ops,
3755 		.of_match_table = d40_match,
3756 	},
3757 };
3758 
3759 static int __init stedma40_init(void)
3760 {
3761 	return platform_driver_probe(&d40_driver, d40_probe);
3762 }
3763 subsys_initcall(stedma40_init);
3764