xref: /openbmc/linux/drivers/dma/ste_dma40.c (revision 4f6cce39)
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 set_current;
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  set_current:
929 	desc->lli_current = lli_current;
930 }
931 
932 static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
933 {
934 	if (chan_is_physical(d40c)) {
935 		d40_phy_lli_load(d40c, d40d);
936 		d40d->lli_current = d40d->lli_len;
937 	} else
938 		d40_log_lli_to_lcxa(d40c, d40d);
939 }
940 
941 static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
942 {
943 	return list_first_entry_or_null(&d40c->active, struct d40_desc, node);
944 }
945 
946 /* remove desc from current queue and add it to the pending_queue */
947 static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
948 {
949 	d40_desc_remove(desc);
950 	desc->is_in_client_list = false;
951 	list_add_tail(&desc->node, &d40c->pending_queue);
952 }
953 
954 static struct d40_desc *d40_first_pending(struct d40_chan *d40c)
955 {
956 	return list_first_entry_or_null(&d40c->pending_queue, struct d40_desc,
957 					node);
958 }
959 
960 static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
961 {
962 	return list_first_entry_or_null(&d40c->queue, struct d40_desc, node);
963 }
964 
965 static struct d40_desc *d40_first_done(struct d40_chan *d40c)
966 {
967 	return list_first_entry_or_null(&d40c->done, struct d40_desc, node);
968 }
969 
970 static int d40_psize_2_burst_size(bool is_log, int psize)
971 {
972 	if (is_log) {
973 		if (psize == STEDMA40_PSIZE_LOG_1)
974 			return 1;
975 	} else {
976 		if (psize == STEDMA40_PSIZE_PHY_1)
977 			return 1;
978 	}
979 
980 	return 2 << psize;
981 }
982 
983 /*
984  * The dma only supports transmitting packages up to
985  * STEDMA40_MAX_SEG_SIZE * data_width, where data_width is stored in Bytes.
986  *
987  * Calculate the total number of dma elements required to send the entire sg list.
988  */
989 static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2)
990 {
991 	int dmalen;
992 	u32 max_w = max(data_width1, data_width2);
993 	u32 min_w = min(data_width1, data_width2);
994 	u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE * min_w, max_w);
995 
996 	if (seg_max > STEDMA40_MAX_SEG_SIZE)
997 		seg_max -= max_w;
998 
999 	if (!IS_ALIGNED(size, max_w))
1000 		return -EINVAL;
1001 
1002 	if (size <= seg_max)
1003 		dmalen = 1;
1004 	else {
1005 		dmalen = size / seg_max;
1006 		if (dmalen * seg_max < size)
1007 			dmalen++;
1008 	}
1009 	return dmalen;
1010 }
1011 
1012 static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len,
1013 			   u32 data_width1, u32 data_width2)
1014 {
1015 	struct scatterlist *sg;
1016 	int i;
1017 	int len = 0;
1018 	int ret;
1019 
1020 	for_each_sg(sgl, sg, sg_len, i) {
1021 		ret = d40_size_2_dmalen(sg_dma_len(sg),
1022 					data_width1, data_width2);
1023 		if (ret < 0)
1024 			return ret;
1025 		len += ret;
1026 	}
1027 	return len;
1028 }
1029 
1030 static int __d40_execute_command_phy(struct d40_chan *d40c,
1031 				     enum d40_command command)
1032 {
1033 	u32 status;
1034 	int i;
1035 	void __iomem *active_reg;
1036 	int ret = 0;
1037 	unsigned long flags;
1038 	u32 wmask;
1039 
1040 	if (command == D40_DMA_STOP) {
1041 		ret = __d40_execute_command_phy(d40c, D40_DMA_SUSPEND_REQ);
1042 		if (ret)
1043 			return ret;
1044 	}
1045 
1046 	spin_lock_irqsave(&d40c->base->execmd_lock, flags);
1047 
1048 	if (d40c->phy_chan->num % 2 == 0)
1049 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1050 	else
1051 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1052 
1053 	if (command == D40_DMA_SUSPEND_REQ) {
1054 		status = (readl(active_reg) &
1055 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1056 			D40_CHAN_POS(d40c->phy_chan->num);
1057 
1058 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1059 			goto unlock;
1060 	}
1061 
1062 	wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
1063 	writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
1064 	       active_reg);
1065 
1066 	if (command == D40_DMA_SUSPEND_REQ) {
1067 
1068 		for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
1069 			status = (readl(active_reg) &
1070 				  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1071 				D40_CHAN_POS(d40c->phy_chan->num);
1072 
1073 			cpu_relax();
1074 			/*
1075 			 * Reduce the number of bus accesses while
1076 			 * waiting for the DMA to suspend.
1077 			 */
1078 			udelay(3);
1079 
1080 			if (status == D40_DMA_STOP ||
1081 			    status == D40_DMA_SUSPENDED)
1082 				break;
1083 		}
1084 
1085 		if (i == D40_SUSPEND_MAX_IT) {
1086 			chan_err(d40c,
1087 				"unable to suspend the chl %d (log: %d) status %x\n",
1088 				d40c->phy_chan->num, d40c->log_num,
1089 				status);
1090 			dump_stack();
1091 			ret = -EBUSY;
1092 		}
1093 
1094 	}
1095  unlock:
1096 	spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
1097 	return ret;
1098 }
1099 
1100 static void d40_term_all(struct d40_chan *d40c)
1101 {
1102 	struct d40_desc *d40d;
1103 	struct d40_desc *_d;
1104 
1105 	/* Release completed descriptors */
1106 	while ((d40d = d40_first_done(d40c))) {
1107 		d40_desc_remove(d40d);
1108 		d40_desc_free(d40c, d40d);
1109 	}
1110 
1111 	/* Release active descriptors */
1112 	while ((d40d = d40_first_active_get(d40c))) {
1113 		d40_desc_remove(d40d);
1114 		d40_desc_free(d40c, d40d);
1115 	}
1116 
1117 	/* Release queued descriptors waiting for transfer */
1118 	while ((d40d = d40_first_queued(d40c))) {
1119 		d40_desc_remove(d40d);
1120 		d40_desc_free(d40c, d40d);
1121 	}
1122 
1123 	/* Release pending descriptors */
1124 	while ((d40d = d40_first_pending(d40c))) {
1125 		d40_desc_remove(d40d);
1126 		d40_desc_free(d40c, d40d);
1127 	}
1128 
1129 	/* Release client owned descriptors */
1130 	if (!list_empty(&d40c->client))
1131 		list_for_each_entry_safe(d40d, _d, &d40c->client, node) {
1132 			d40_desc_remove(d40d);
1133 			d40_desc_free(d40c, d40d);
1134 		}
1135 
1136 	/* Release descriptors in prepare queue */
1137 	if (!list_empty(&d40c->prepare_queue))
1138 		list_for_each_entry_safe(d40d, _d,
1139 					 &d40c->prepare_queue, node) {
1140 			d40_desc_remove(d40d);
1141 			d40_desc_free(d40c, d40d);
1142 		}
1143 
1144 	d40c->pending_tx = 0;
1145 }
1146 
1147 static void __d40_config_set_event(struct d40_chan *d40c,
1148 				   enum d40_events event_type, u32 event,
1149 				   int reg)
1150 {
1151 	void __iomem *addr = chan_base(d40c) + reg;
1152 	int tries;
1153 	u32 status;
1154 
1155 	switch (event_type) {
1156 
1157 	case D40_DEACTIVATE_EVENTLINE:
1158 
1159 		writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
1160 		       | ~D40_EVENTLINE_MASK(event), addr);
1161 		break;
1162 
1163 	case D40_SUSPEND_REQ_EVENTLINE:
1164 		status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1165 			  D40_EVENTLINE_POS(event);
1166 
1167 		if (status == D40_DEACTIVATE_EVENTLINE ||
1168 		    status == D40_SUSPEND_REQ_EVENTLINE)
1169 			break;
1170 
1171 		writel((D40_SUSPEND_REQ_EVENTLINE << D40_EVENTLINE_POS(event))
1172 		       | ~D40_EVENTLINE_MASK(event), addr);
1173 
1174 		for (tries = 0 ; tries < D40_SUSPEND_MAX_IT; tries++) {
1175 
1176 			status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1177 				  D40_EVENTLINE_POS(event);
1178 
1179 			cpu_relax();
1180 			/*
1181 			 * Reduce the number of bus accesses while
1182 			 * waiting for the DMA to suspend.
1183 			 */
1184 			udelay(3);
1185 
1186 			if (status == D40_DEACTIVATE_EVENTLINE)
1187 				break;
1188 		}
1189 
1190 		if (tries == D40_SUSPEND_MAX_IT) {
1191 			chan_err(d40c,
1192 				"unable to stop the event_line chl %d (log: %d)"
1193 				"status %x\n", d40c->phy_chan->num,
1194 				 d40c->log_num, status);
1195 		}
1196 		break;
1197 
1198 	case D40_ACTIVATE_EVENTLINE:
1199 	/*
1200 	 * The hardware sometimes doesn't register the enable when src and dst
1201 	 * event lines are active on the same logical channel.  Retry to ensure
1202 	 * it does.  Usually only one retry is sufficient.
1203 	 */
1204 		tries = 100;
1205 		while (--tries) {
1206 			writel((D40_ACTIVATE_EVENTLINE <<
1207 				D40_EVENTLINE_POS(event)) |
1208 				~D40_EVENTLINE_MASK(event), addr);
1209 
1210 			if (readl(addr) & D40_EVENTLINE_MASK(event))
1211 				break;
1212 		}
1213 
1214 		if (tries != 99)
1215 			dev_dbg(chan2dev(d40c),
1216 				"[%s] workaround enable S%cLNK (%d tries)\n",
1217 				__func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D',
1218 				100 - tries);
1219 
1220 		WARN_ON(!tries);
1221 		break;
1222 
1223 	case D40_ROUND_EVENTLINE:
1224 		BUG();
1225 		break;
1226 
1227 	}
1228 }
1229 
1230 static void d40_config_set_event(struct d40_chan *d40c,
1231 				 enum d40_events event_type)
1232 {
1233 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
1234 
1235 	/* Enable event line connected to device (or memcpy) */
1236 	if ((d40c->dma_cfg.dir == DMA_DEV_TO_MEM) ||
1237 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
1238 		__d40_config_set_event(d40c, event_type, event,
1239 				       D40_CHAN_REG_SSLNK);
1240 
1241 	if (d40c->dma_cfg.dir !=  DMA_DEV_TO_MEM)
1242 		__d40_config_set_event(d40c, event_type, event,
1243 				       D40_CHAN_REG_SDLNK);
1244 }
1245 
1246 static u32 d40_chan_has_events(struct d40_chan *d40c)
1247 {
1248 	void __iomem *chanbase = chan_base(d40c);
1249 	u32 val;
1250 
1251 	val = readl(chanbase + D40_CHAN_REG_SSLNK);
1252 	val |= readl(chanbase + D40_CHAN_REG_SDLNK);
1253 
1254 	return val;
1255 }
1256 
1257 static int
1258 __d40_execute_command_log(struct d40_chan *d40c, enum d40_command command)
1259 {
1260 	unsigned long flags;
1261 	int ret = 0;
1262 	u32 active_status;
1263 	void __iomem *active_reg;
1264 
1265 	if (d40c->phy_chan->num % 2 == 0)
1266 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1267 	else
1268 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1269 
1270 
1271 	spin_lock_irqsave(&d40c->phy_chan->lock, flags);
1272 
1273 	switch (command) {
1274 	case D40_DMA_STOP:
1275 	case D40_DMA_SUSPEND_REQ:
1276 
1277 		active_status = (readl(active_reg) &
1278 				 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1279 				 D40_CHAN_POS(d40c->phy_chan->num);
1280 
1281 		if (active_status == D40_DMA_RUN)
1282 			d40_config_set_event(d40c, D40_SUSPEND_REQ_EVENTLINE);
1283 		else
1284 			d40_config_set_event(d40c, D40_DEACTIVATE_EVENTLINE);
1285 
1286 		if (!d40_chan_has_events(d40c) && (command == D40_DMA_STOP))
1287 			ret = __d40_execute_command_phy(d40c, command);
1288 
1289 		break;
1290 
1291 	case D40_DMA_RUN:
1292 
1293 		d40_config_set_event(d40c, D40_ACTIVATE_EVENTLINE);
1294 		ret = __d40_execute_command_phy(d40c, command);
1295 		break;
1296 
1297 	case D40_DMA_SUSPENDED:
1298 		BUG();
1299 		break;
1300 	}
1301 
1302 	spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
1303 	return ret;
1304 }
1305 
1306 static int d40_channel_execute_command(struct d40_chan *d40c,
1307 				       enum d40_command command)
1308 {
1309 	if (chan_is_logical(d40c))
1310 		return __d40_execute_command_log(d40c, command);
1311 	else
1312 		return __d40_execute_command_phy(d40c, command);
1313 }
1314 
1315 static u32 d40_get_prmo(struct d40_chan *d40c)
1316 {
1317 	static const unsigned int phy_map[] = {
1318 		[STEDMA40_PCHAN_BASIC_MODE]
1319 			= D40_DREG_PRMO_PCHAN_BASIC,
1320 		[STEDMA40_PCHAN_MODULO_MODE]
1321 			= D40_DREG_PRMO_PCHAN_MODULO,
1322 		[STEDMA40_PCHAN_DOUBLE_DST_MODE]
1323 			= D40_DREG_PRMO_PCHAN_DOUBLE_DST,
1324 	};
1325 	static const unsigned int log_map[] = {
1326 		[STEDMA40_LCHAN_SRC_PHY_DST_LOG]
1327 			= D40_DREG_PRMO_LCHAN_SRC_PHY_DST_LOG,
1328 		[STEDMA40_LCHAN_SRC_LOG_DST_PHY]
1329 			= D40_DREG_PRMO_LCHAN_SRC_LOG_DST_PHY,
1330 		[STEDMA40_LCHAN_SRC_LOG_DST_LOG]
1331 			= D40_DREG_PRMO_LCHAN_SRC_LOG_DST_LOG,
1332 	};
1333 
1334 	if (chan_is_physical(d40c))
1335 		return phy_map[d40c->dma_cfg.mode_opt];
1336 	else
1337 		return log_map[d40c->dma_cfg.mode_opt];
1338 }
1339 
1340 static void d40_config_write(struct d40_chan *d40c)
1341 {
1342 	u32 addr_base;
1343 	u32 var;
1344 
1345 	/* Odd addresses are even addresses + 4 */
1346 	addr_base = (d40c->phy_chan->num % 2) * 4;
1347 	/* Setup channel mode to logical or physical */
1348 	var = ((u32)(chan_is_logical(d40c)) + 1) <<
1349 		D40_CHAN_POS(d40c->phy_chan->num);
1350 	writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
1351 
1352 	/* Setup operational mode option register */
1353 	var = d40_get_prmo(d40c) << D40_CHAN_POS(d40c->phy_chan->num);
1354 
1355 	writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
1356 
1357 	if (chan_is_logical(d40c)) {
1358 		int lidx = (d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS)
1359 			   & D40_SREG_ELEM_LOG_LIDX_MASK;
1360 		void __iomem *chanbase = chan_base(d40c);
1361 
1362 		/* Set default config for CFG reg */
1363 		writel(d40c->src_def_cfg, chanbase + D40_CHAN_REG_SSCFG);
1364 		writel(d40c->dst_def_cfg, chanbase + D40_CHAN_REG_SDCFG);
1365 
1366 		/* Set LIDX for lcla */
1367 		writel(lidx, chanbase + D40_CHAN_REG_SSELT);
1368 		writel(lidx, chanbase + D40_CHAN_REG_SDELT);
1369 
1370 		/* Clear LNK which will be used by d40_chan_has_events() */
1371 		writel(0, chanbase + D40_CHAN_REG_SSLNK);
1372 		writel(0, chanbase + D40_CHAN_REG_SDLNK);
1373 	}
1374 }
1375 
1376 static u32 d40_residue(struct d40_chan *d40c)
1377 {
1378 	u32 num_elt;
1379 
1380 	if (chan_is_logical(d40c))
1381 		num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
1382 			>> D40_MEM_LCSP2_ECNT_POS;
1383 	else {
1384 		u32 val = readl(chan_base(d40c) + D40_CHAN_REG_SDELT);
1385 		num_elt = (val & D40_SREG_ELEM_PHY_ECNT_MASK)
1386 			  >> D40_SREG_ELEM_PHY_ECNT_POS;
1387 	}
1388 
1389 	return num_elt * d40c->dma_cfg.dst_info.data_width;
1390 }
1391 
1392 static bool d40_tx_is_linked(struct d40_chan *d40c)
1393 {
1394 	bool is_link;
1395 
1396 	if (chan_is_logical(d40c))
1397 		is_link = readl(&d40c->lcpa->lcsp3) &  D40_MEM_LCSP3_DLOS_MASK;
1398 	else
1399 		is_link = readl(chan_base(d40c) + D40_CHAN_REG_SDLNK)
1400 			  & D40_SREG_LNK_PHYS_LNK_MASK;
1401 
1402 	return is_link;
1403 }
1404 
1405 static int d40_pause(struct dma_chan *chan)
1406 {
1407 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
1408 	int res = 0;
1409 	unsigned long flags;
1410 
1411 	if (d40c->phy_chan == NULL) {
1412 		chan_err(d40c, "Channel is not allocated!\n");
1413 		return -EINVAL;
1414 	}
1415 
1416 	if (!d40c->busy)
1417 		return 0;
1418 
1419 	spin_lock_irqsave(&d40c->lock, flags);
1420 	pm_runtime_get_sync(d40c->base->dev);
1421 
1422 	res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1423 
1424 	pm_runtime_mark_last_busy(d40c->base->dev);
1425 	pm_runtime_put_autosuspend(d40c->base->dev);
1426 	spin_unlock_irqrestore(&d40c->lock, flags);
1427 	return res;
1428 }
1429 
1430 static int d40_resume(struct dma_chan *chan)
1431 {
1432 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
1433 	int res = 0;
1434 	unsigned long flags;
1435 
1436 	if (d40c->phy_chan == NULL) {
1437 		chan_err(d40c, "Channel is not allocated!\n");
1438 		return -EINVAL;
1439 	}
1440 
1441 	if (!d40c->busy)
1442 		return 0;
1443 
1444 	spin_lock_irqsave(&d40c->lock, flags);
1445 	pm_runtime_get_sync(d40c->base->dev);
1446 
1447 	/* If bytes left to transfer or linked tx resume job */
1448 	if (d40_residue(d40c) || d40_tx_is_linked(d40c))
1449 		res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1450 
1451 	pm_runtime_mark_last_busy(d40c->base->dev);
1452 	pm_runtime_put_autosuspend(d40c->base->dev);
1453 	spin_unlock_irqrestore(&d40c->lock, flags);
1454 	return res;
1455 }
1456 
1457 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
1458 {
1459 	struct d40_chan *d40c = container_of(tx->chan,
1460 					     struct d40_chan,
1461 					     chan);
1462 	struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
1463 	unsigned long flags;
1464 	dma_cookie_t cookie;
1465 
1466 	spin_lock_irqsave(&d40c->lock, flags);
1467 	cookie = dma_cookie_assign(tx);
1468 	d40_desc_queue(d40c, d40d);
1469 	spin_unlock_irqrestore(&d40c->lock, flags);
1470 
1471 	return cookie;
1472 }
1473 
1474 static int d40_start(struct d40_chan *d40c)
1475 {
1476 	return d40_channel_execute_command(d40c, D40_DMA_RUN);
1477 }
1478 
1479 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
1480 {
1481 	struct d40_desc *d40d;
1482 	int err;
1483 
1484 	/* Start queued jobs, if any */
1485 	d40d = d40_first_queued(d40c);
1486 
1487 	if (d40d != NULL) {
1488 		if (!d40c->busy) {
1489 			d40c->busy = true;
1490 			pm_runtime_get_sync(d40c->base->dev);
1491 		}
1492 
1493 		/* Remove from queue */
1494 		d40_desc_remove(d40d);
1495 
1496 		/* Add to active queue */
1497 		d40_desc_submit(d40c, d40d);
1498 
1499 		/* Initiate DMA job */
1500 		d40_desc_load(d40c, d40d);
1501 
1502 		/* Start dma job */
1503 		err = d40_start(d40c);
1504 
1505 		if (err)
1506 			return NULL;
1507 	}
1508 
1509 	return d40d;
1510 }
1511 
1512 /* called from interrupt context */
1513 static void dma_tc_handle(struct d40_chan *d40c)
1514 {
1515 	struct d40_desc *d40d;
1516 
1517 	/* Get first active entry from list */
1518 	d40d = d40_first_active_get(d40c);
1519 
1520 	if (d40d == NULL)
1521 		return;
1522 
1523 	if (d40d->cyclic) {
1524 		/*
1525 		 * If this was a paritially loaded list, we need to reloaded
1526 		 * it, and only when the list is completed.  We need to check
1527 		 * for done because the interrupt will hit for every link, and
1528 		 * not just the last one.
1529 		 */
1530 		if (d40d->lli_current < d40d->lli_len
1531 		    && !d40_tx_is_linked(d40c)
1532 		    && !d40_residue(d40c)) {
1533 			d40_lcla_free_all(d40c, d40d);
1534 			d40_desc_load(d40c, d40d);
1535 			(void) d40_start(d40c);
1536 
1537 			if (d40d->lli_current == d40d->lli_len)
1538 				d40d->lli_current = 0;
1539 		}
1540 	} else {
1541 		d40_lcla_free_all(d40c, d40d);
1542 
1543 		if (d40d->lli_current < d40d->lli_len) {
1544 			d40_desc_load(d40c, d40d);
1545 			/* Start dma job */
1546 			(void) d40_start(d40c);
1547 			return;
1548 		}
1549 
1550 		if (d40_queue_start(d40c) == NULL) {
1551 			d40c->busy = false;
1552 
1553 			pm_runtime_mark_last_busy(d40c->base->dev);
1554 			pm_runtime_put_autosuspend(d40c->base->dev);
1555 		}
1556 
1557 		d40_desc_remove(d40d);
1558 		d40_desc_done(d40c, d40d);
1559 	}
1560 
1561 	d40c->pending_tx++;
1562 	tasklet_schedule(&d40c->tasklet);
1563 
1564 }
1565 
1566 static void dma_tasklet(unsigned long data)
1567 {
1568 	struct d40_chan *d40c = (struct d40_chan *) data;
1569 	struct d40_desc *d40d;
1570 	unsigned long flags;
1571 	bool callback_active;
1572 	struct dmaengine_desc_callback cb;
1573 
1574 	spin_lock_irqsave(&d40c->lock, flags);
1575 
1576 	/* Get first entry from the done list */
1577 	d40d = d40_first_done(d40c);
1578 	if (d40d == NULL) {
1579 		/* Check if we have reached here for cyclic job */
1580 		d40d = d40_first_active_get(d40c);
1581 		if (d40d == NULL || !d40d->cyclic)
1582 			goto check_pending_tx;
1583 	}
1584 
1585 	if (!d40d->cyclic)
1586 		dma_cookie_complete(&d40d->txd);
1587 
1588 	/*
1589 	 * If terminating a channel pending_tx is set to zero.
1590 	 * This prevents any finished active jobs to return to the client.
1591 	 */
1592 	if (d40c->pending_tx == 0) {
1593 		spin_unlock_irqrestore(&d40c->lock, flags);
1594 		return;
1595 	}
1596 
1597 	/* Callback to client */
1598 	callback_active = !!(d40d->txd.flags & DMA_PREP_INTERRUPT);
1599 	dmaengine_desc_get_callback(&d40d->txd, &cb);
1600 
1601 	if (!d40d->cyclic) {
1602 		if (async_tx_test_ack(&d40d->txd)) {
1603 			d40_desc_remove(d40d);
1604 			d40_desc_free(d40c, d40d);
1605 		} else if (!d40d->is_in_client_list) {
1606 			d40_desc_remove(d40d);
1607 			d40_lcla_free_all(d40c, d40d);
1608 			list_add_tail(&d40d->node, &d40c->client);
1609 			d40d->is_in_client_list = true;
1610 		}
1611 	}
1612 
1613 	d40c->pending_tx--;
1614 
1615 	if (d40c->pending_tx)
1616 		tasklet_schedule(&d40c->tasklet);
1617 
1618 	spin_unlock_irqrestore(&d40c->lock, flags);
1619 
1620 	if (callback_active)
1621 		dmaengine_desc_callback_invoke(&cb, NULL);
1622 
1623 	return;
1624  check_pending_tx:
1625 	/* Rescue manouver if receiving double interrupts */
1626 	if (d40c->pending_tx > 0)
1627 		d40c->pending_tx--;
1628 	spin_unlock_irqrestore(&d40c->lock, flags);
1629 }
1630 
1631 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1632 {
1633 	int i;
1634 	u32 idx;
1635 	u32 row;
1636 	long chan = -1;
1637 	struct d40_chan *d40c;
1638 	unsigned long flags;
1639 	struct d40_base *base = data;
1640 	u32 regs[base->gen_dmac.il_size];
1641 	struct d40_interrupt_lookup *il = base->gen_dmac.il;
1642 	u32 il_size = base->gen_dmac.il_size;
1643 
1644 	spin_lock_irqsave(&base->interrupt_lock, flags);
1645 
1646 	/* Read interrupt status of both logical and physical channels */
1647 	for (i = 0; i < il_size; i++)
1648 		regs[i] = readl(base->virtbase + il[i].src);
1649 
1650 	for (;;) {
1651 
1652 		chan = find_next_bit((unsigned long *)regs,
1653 				     BITS_PER_LONG * il_size, chan + 1);
1654 
1655 		/* No more set bits found? */
1656 		if (chan == BITS_PER_LONG * il_size)
1657 			break;
1658 
1659 		row = chan / BITS_PER_LONG;
1660 		idx = chan & (BITS_PER_LONG - 1);
1661 
1662 		if (il[row].offset == D40_PHY_CHAN)
1663 			d40c = base->lookup_phy_chans[idx];
1664 		else
1665 			d40c = base->lookup_log_chans[il[row].offset + idx];
1666 
1667 		if (!d40c) {
1668 			/*
1669 			 * No error because this can happen if something else
1670 			 * in the system is using the channel.
1671 			 */
1672 			continue;
1673 		}
1674 
1675 		/* ACK interrupt */
1676 		writel(BIT(idx), base->virtbase + il[row].clr);
1677 
1678 		spin_lock(&d40c->lock);
1679 
1680 		if (!il[row].is_error)
1681 			dma_tc_handle(d40c);
1682 		else
1683 			d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1684 				chan, il[row].offset, idx);
1685 
1686 		spin_unlock(&d40c->lock);
1687 	}
1688 
1689 	spin_unlock_irqrestore(&base->interrupt_lock, flags);
1690 
1691 	return IRQ_HANDLED;
1692 }
1693 
1694 static int d40_validate_conf(struct d40_chan *d40c,
1695 			     struct stedma40_chan_cfg *conf)
1696 {
1697 	int res = 0;
1698 	bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1699 
1700 	if (!conf->dir) {
1701 		chan_err(d40c, "Invalid direction.\n");
1702 		res = -EINVAL;
1703 	}
1704 
1705 	if ((is_log && conf->dev_type > d40c->base->num_log_chans)  ||
1706 	    (!is_log && conf->dev_type > d40c->base->num_phy_chans) ||
1707 	    (conf->dev_type < 0)) {
1708 		chan_err(d40c, "Invalid device type (%d)\n", conf->dev_type);
1709 		res = -EINVAL;
1710 	}
1711 
1712 	if (conf->dir == DMA_DEV_TO_DEV) {
1713 		/*
1714 		 * DMAC HW supports it. Will be added to this driver,
1715 		 * in case any dma client requires it.
1716 		 */
1717 		chan_err(d40c, "periph to periph not supported\n");
1718 		res = -EINVAL;
1719 	}
1720 
1721 	if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1722 	    conf->src_info.data_width !=
1723 	    d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1724 	    conf->dst_info.data_width) {
1725 		/*
1726 		 * The DMAC hardware only supports
1727 		 * src (burst x width) == dst (burst x width)
1728 		 */
1729 
1730 		chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1731 		res = -EINVAL;
1732 	}
1733 
1734 	return res;
1735 }
1736 
1737 static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1738 			       bool is_src, int log_event_line, bool is_log,
1739 			       bool *first_user)
1740 {
1741 	unsigned long flags;
1742 	spin_lock_irqsave(&phy->lock, flags);
1743 
1744 	*first_user = ((phy->allocated_src | phy->allocated_dst)
1745 			== D40_ALLOC_FREE);
1746 
1747 	if (!is_log) {
1748 		/* Physical interrupts are masked per physical full channel */
1749 		if (phy->allocated_src == D40_ALLOC_FREE &&
1750 		    phy->allocated_dst == D40_ALLOC_FREE) {
1751 			phy->allocated_dst = D40_ALLOC_PHY;
1752 			phy->allocated_src = D40_ALLOC_PHY;
1753 			goto found_unlock;
1754 		} else
1755 			goto not_found_unlock;
1756 	}
1757 
1758 	/* Logical channel */
1759 	if (is_src) {
1760 		if (phy->allocated_src == D40_ALLOC_PHY)
1761 			goto not_found_unlock;
1762 
1763 		if (phy->allocated_src == D40_ALLOC_FREE)
1764 			phy->allocated_src = D40_ALLOC_LOG_FREE;
1765 
1766 		if (!(phy->allocated_src & BIT(log_event_line))) {
1767 			phy->allocated_src |= BIT(log_event_line);
1768 			goto found_unlock;
1769 		} else
1770 			goto not_found_unlock;
1771 	} else {
1772 		if (phy->allocated_dst == D40_ALLOC_PHY)
1773 			goto not_found_unlock;
1774 
1775 		if (phy->allocated_dst == D40_ALLOC_FREE)
1776 			phy->allocated_dst = D40_ALLOC_LOG_FREE;
1777 
1778 		if (!(phy->allocated_dst & BIT(log_event_line))) {
1779 			phy->allocated_dst |= BIT(log_event_line);
1780 			goto found_unlock;
1781 		}
1782 	}
1783  not_found_unlock:
1784 	spin_unlock_irqrestore(&phy->lock, flags);
1785 	return false;
1786  found_unlock:
1787 	spin_unlock_irqrestore(&phy->lock, flags);
1788 	return true;
1789 }
1790 
1791 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1792 			       int log_event_line)
1793 {
1794 	unsigned long flags;
1795 	bool is_free = false;
1796 
1797 	spin_lock_irqsave(&phy->lock, flags);
1798 	if (!log_event_line) {
1799 		phy->allocated_dst = D40_ALLOC_FREE;
1800 		phy->allocated_src = D40_ALLOC_FREE;
1801 		is_free = true;
1802 		goto unlock;
1803 	}
1804 
1805 	/* Logical channel */
1806 	if (is_src) {
1807 		phy->allocated_src &= ~BIT(log_event_line);
1808 		if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1809 			phy->allocated_src = D40_ALLOC_FREE;
1810 	} else {
1811 		phy->allocated_dst &= ~BIT(log_event_line);
1812 		if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1813 			phy->allocated_dst = D40_ALLOC_FREE;
1814 	}
1815 
1816 	is_free = ((phy->allocated_src | phy->allocated_dst) ==
1817 		   D40_ALLOC_FREE);
1818  unlock:
1819 	spin_unlock_irqrestore(&phy->lock, flags);
1820 
1821 	return is_free;
1822 }
1823 
1824 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1825 {
1826 	int dev_type = d40c->dma_cfg.dev_type;
1827 	int event_group;
1828 	int event_line;
1829 	struct d40_phy_res *phys;
1830 	int i;
1831 	int j;
1832 	int log_num;
1833 	int num_phy_chans;
1834 	bool is_src;
1835 	bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1836 
1837 	phys = d40c->base->phy_res;
1838 	num_phy_chans = d40c->base->num_phy_chans;
1839 
1840 	if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
1841 		log_num = 2 * dev_type;
1842 		is_src = true;
1843 	} else if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
1844 		   d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1845 		/* dst event lines are used for logical memcpy */
1846 		log_num = 2 * dev_type + 1;
1847 		is_src = false;
1848 	} else
1849 		return -EINVAL;
1850 
1851 	event_group = D40_TYPE_TO_GROUP(dev_type);
1852 	event_line = D40_TYPE_TO_EVENT(dev_type);
1853 
1854 	if (!is_log) {
1855 		if (d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1856 			/* Find physical half channel */
1857 			if (d40c->dma_cfg.use_fixed_channel) {
1858 				i = d40c->dma_cfg.phy_channel;
1859 				if (d40_alloc_mask_set(&phys[i], is_src,
1860 						       0, is_log,
1861 						       first_phy_user))
1862 					goto found_phy;
1863 			} else {
1864 				for (i = 0; i < num_phy_chans; i++) {
1865 					if (d40_alloc_mask_set(&phys[i], is_src,
1866 						       0, is_log,
1867 						       first_phy_user))
1868 						goto found_phy;
1869 				}
1870 			}
1871 		} else
1872 			for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1873 				int phy_num = j  + event_group * 2;
1874 				for (i = phy_num; i < phy_num + 2; i++) {
1875 					if (d40_alloc_mask_set(&phys[i],
1876 							       is_src,
1877 							       0,
1878 							       is_log,
1879 							       first_phy_user))
1880 						goto found_phy;
1881 				}
1882 			}
1883 		return -EINVAL;
1884 found_phy:
1885 		d40c->phy_chan = &phys[i];
1886 		d40c->log_num = D40_PHY_CHAN;
1887 		goto out;
1888 	}
1889 	if (dev_type == -1)
1890 		return -EINVAL;
1891 
1892 	/* Find logical channel */
1893 	for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1894 		int phy_num = j + event_group * 2;
1895 
1896 		if (d40c->dma_cfg.use_fixed_channel) {
1897 			i = d40c->dma_cfg.phy_channel;
1898 
1899 			if ((i != phy_num) && (i != phy_num + 1)) {
1900 				dev_err(chan2dev(d40c),
1901 					"invalid fixed phy channel %d\n", i);
1902 				return -EINVAL;
1903 			}
1904 
1905 			if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1906 					       is_log, first_phy_user))
1907 				goto found_log;
1908 
1909 			dev_err(chan2dev(d40c),
1910 				"could not allocate fixed phy channel %d\n", i);
1911 			return -EINVAL;
1912 		}
1913 
1914 		/*
1915 		 * Spread logical channels across all available physical rather
1916 		 * than pack every logical channel at the first available phy
1917 		 * channels.
1918 		 */
1919 		if (is_src) {
1920 			for (i = phy_num; i < phy_num + 2; i++) {
1921 				if (d40_alloc_mask_set(&phys[i], is_src,
1922 						       event_line, is_log,
1923 						       first_phy_user))
1924 					goto found_log;
1925 			}
1926 		} else {
1927 			for (i = phy_num + 1; i >= phy_num; i--) {
1928 				if (d40_alloc_mask_set(&phys[i], is_src,
1929 						       event_line, is_log,
1930 						       first_phy_user))
1931 					goto found_log;
1932 			}
1933 		}
1934 	}
1935 	return -EINVAL;
1936 
1937 found_log:
1938 	d40c->phy_chan = &phys[i];
1939 	d40c->log_num = log_num;
1940 out:
1941 
1942 	if (is_log)
1943 		d40c->base->lookup_log_chans[d40c->log_num] = d40c;
1944 	else
1945 		d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
1946 
1947 	return 0;
1948 
1949 }
1950 
1951 static int d40_config_memcpy(struct d40_chan *d40c)
1952 {
1953 	dma_cap_mask_t cap = d40c->chan.device->cap_mask;
1954 
1955 	if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
1956 		d40c->dma_cfg = dma40_memcpy_conf_log;
1957 		d40c->dma_cfg.dev_type = dma40_memcpy_channels[d40c->chan.chan_id];
1958 
1959 		d40_log_cfg(&d40c->dma_cfg,
1960 			    &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
1961 
1962 	} else if (dma_has_cap(DMA_MEMCPY, cap) &&
1963 		   dma_has_cap(DMA_SLAVE, cap)) {
1964 		d40c->dma_cfg = dma40_memcpy_conf_phy;
1965 
1966 		/* Generate interrrupt at end of transfer or relink. */
1967 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_TIM_POS);
1968 
1969 		/* Generate interrupt on error. */
1970 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
1971 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
1972 
1973 	} else {
1974 		chan_err(d40c, "No memcpy\n");
1975 		return -EINVAL;
1976 	}
1977 
1978 	return 0;
1979 }
1980 
1981 static int d40_free_dma(struct d40_chan *d40c)
1982 {
1983 
1984 	int res = 0;
1985 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
1986 	struct d40_phy_res *phy = d40c->phy_chan;
1987 	bool is_src;
1988 
1989 	/* Terminate all queued and active transfers */
1990 	d40_term_all(d40c);
1991 
1992 	if (phy == NULL) {
1993 		chan_err(d40c, "phy == null\n");
1994 		return -EINVAL;
1995 	}
1996 
1997 	if (phy->allocated_src == D40_ALLOC_FREE &&
1998 	    phy->allocated_dst == D40_ALLOC_FREE) {
1999 		chan_err(d40c, "channel already free\n");
2000 		return -EINVAL;
2001 	}
2002 
2003 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2004 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM)
2005 		is_src = false;
2006 	else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2007 		is_src = true;
2008 	else {
2009 		chan_err(d40c, "Unknown direction\n");
2010 		return -EINVAL;
2011 	}
2012 
2013 	pm_runtime_get_sync(d40c->base->dev);
2014 	res = d40_channel_execute_command(d40c, D40_DMA_STOP);
2015 	if (res) {
2016 		chan_err(d40c, "stop failed\n");
2017 		goto mark_last_busy;
2018 	}
2019 
2020 	d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
2021 
2022 	if (chan_is_logical(d40c))
2023 		d40c->base->lookup_log_chans[d40c->log_num] = NULL;
2024 	else
2025 		d40c->base->lookup_phy_chans[phy->num] = NULL;
2026 
2027 	if (d40c->busy) {
2028 		pm_runtime_mark_last_busy(d40c->base->dev);
2029 		pm_runtime_put_autosuspend(d40c->base->dev);
2030 	}
2031 
2032 	d40c->busy = false;
2033 	d40c->phy_chan = NULL;
2034 	d40c->configured = false;
2035  mark_last_busy:
2036 	pm_runtime_mark_last_busy(d40c->base->dev);
2037 	pm_runtime_put_autosuspend(d40c->base->dev);
2038 	return res;
2039 }
2040 
2041 static bool d40_is_paused(struct d40_chan *d40c)
2042 {
2043 	void __iomem *chanbase = chan_base(d40c);
2044 	bool is_paused = false;
2045 	unsigned long flags;
2046 	void __iomem *active_reg;
2047 	u32 status;
2048 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2049 
2050 	spin_lock_irqsave(&d40c->lock, flags);
2051 
2052 	if (chan_is_physical(d40c)) {
2053 		if (d40c->phy_chan->num % 2 == 0)
2054 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
2055 		else
2056 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
2057 
2058 		status = (readl(active_reg) &
2059 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
2060 			D40_CHAN_POS(d40c->phy_chan->num);
2061 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
2062 			is_paused = true;
2063 		goto unlock;
2064 	}
2065 
2066 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2067 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
2068 		status = readl(chanbase + D40_CHAN_REG_SDLNK);
2069 	} else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
2070 		status = readl(chanbase + D40_CHAN_REG_SSLNK);
2071 	} else {
2072 		chan_err(d40c, "Unknown direction\n");
2073 		goto unlock;
2074 	}
2075 
2076 	status = (status & D40_EVENTLINE_MASK(event)) >>
2077 		D40_EVENTLINE_POS(event);
2078 
2079 	if (status != D40_DMA_RUN)
2080 		is_paused = true;
2081  unlock:
2082 	spin_unlock_irqrestore(&d40c->lock, flags);
2083 	return is_paused;
2084 
2085 }
2086 
2087 static u32 stedma40_residue(struct dma_chan *chan)
2088 {
2089 	struct d40_chan *d40c =
2090 		container_of(chan, struct d40_chan, chan);
2091 	u32 bytes_left;
2092 	unsigned long flags;
2093 
2094 	spin_lock_irqsave(&d40c->lock, flags);
2095 	bytes_left = d40_residue(d40c);
2096 	spin_unlock_irqrestore(&d40c->lock, flags);
2097 
2098 	return bytes_left;
2099 }
2100 
2101 static int
2102 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
2103 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2104 		unsigned int sg_len, dma_addr_t src_dev_addr,
2105 		dma_addr_t dst_dev_addr)
2106 {
2107 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2108 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2109 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2110 	int ret;
2111 
2112 	ret = d40_log_sg_to_lli(sg_src, sg_len,
2113 				src_dev_addr,
2114 				desc->lli_log.src,
2115 				chan->log_def.lcsp1,
2116 				src_info->data_width,
2117 				dst_info->data_width);
2118 
2119 	ret = d40_log_sg_to_lli(sg_dst, sg_len,
2120 				dst_dev_addr,
2121 				desc->lli_log.dst,
2122 				chan->log_def.lcsp3,
2123 				dst_info->data_width,
2124 				src_info->data_width);
2125 
2126 	return ret < 0 ? ret : 0;
2127 }
2128 
2129 static int
2130 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2131 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2132 		unsigned int sg_len, dma_addr_t src_dev_addr,
2133 		dma_addr_t dst_dev_addr)
2134 {
2135 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2136 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2137 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2138 	unsigned long flags = 0;
2139 	int ret;
2140 
2141 	if (desc->cyclic)
2142 		flags |= LLI_CYCLIC | LLI_TERM_INT;
2143 
2144 	ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2145 				desc->lli_phy.src,
2146 				virt_to_phys(desc->lli_phy.src),
2147 				chan->src_def_cfg,
2148 				src_info, dst_info, flags);
2149 
2150 	ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2151 				desc->lli_phy.dst,
2152 				virt_to_phys(desc->lli_phy.dst),
2153 				chan->dst_def_cfg,
2154 				dst_info, src_info, flags);
2155 
2156 	dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2157 				   desc->lli_pool.size, DMA_TO_DEVICE);
2158 
2159 	return ret < 0 ? ret : 0;
2160 }
2161 
2162 static struct d40_desc *
2163 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2164 	      unsigned int sg_len, unsigned long dma_flags)
2165 {
2166 	struct stedma40_chan_cfg *cfg;
2167 	struct d40_desc *desc;
2168 	int ret;
2169 
2170 	desc = d40_desc_get(chan);
2171 	if (!desc)
2172 		return NULL;
2173 
2174 	cfg = &chan->dma_cfg;
2175 	desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2176 					cfg->dst_info.data_width);
2177 	if (desc->lli_len < 0) {
2178 		chan_err(chan, "Unaligned size\n");
2179 		goto free_desc;
2180 	}
2181 
2182 	ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2183 	if (ret < 0) {
2184 		chan_err(chan, "Could not allocate lli\n");
2185 		goto free_desc;
2186 	}
2187 
2188 	desc->lli_current = 0;
2189 	desc->txd.flags = dma_flags;
2190 	desc->txd.tx_submit = d40_tx_submit;
2191 
2192 	dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2193 
2194 	return desc;
2195  free_desc:
2196 	d40_desc_free(chan, desc);
2197 	return NULL;
2198 }
2199 
2200 static struct dma_async_tx_descriptor *
2201 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2202 	    struct scatterlist *sg_dst, unsigned int sg_len,
2203 	    enum dma_transfer_direction direction, unsigned long dma_flags)
2204 {
2205 	struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2206 	dma_addr_t src_dev_addr;
2207 	dma_addr_t dst_dev_addr;
2208 	struct d40_desc *desc;
2209 	unsigned long flags;
2210 	int ret;
2211 
2212 	if (!chan->phy_chan) {
2213 		chan_err(chan, "Cannot prepare unallocated channel\n");
2214 		return NULL;
2215 	}
2216 
2217 	spin_lock_irqsave(&chan->lock, flags);
2218 
2219 	desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2220 	if (desc == NULL)
2221 		goto unlock;
2222 
2223 	if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2224 		desc->cyclic = true;
2225 
2226 	src_dev_addr = 0;
2227 	dst_dev_addr = 0;
2228 	if (direction == DMA_DEV_TO_MEM)
2229 		src_dev_addr = chan->runtime_addr;
2230 	else if (direction == DMA_MEM_TO_DEV)
2231 		dst_dev_addr = chan->runtime_addr;
2232 
2233 	if (chan_is_logical(chan))
2234 		ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2235 				      sg_len, src_dev_addr, dst_dev_addr);
2236 	else
2237 		ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2238 				      sg_len, src_dev_addr, dst_dev_addr);
2239 
2240 	if (ret) {
2241 		chan_err(chan, "Failed to prepare %s sg job: %d\n",
2242 			 chan_is_logical(chan) ? "log" : "phy", ret);
2243 		goto free_desc;
2244 	}
2245 
2246 	/*
2247 	 * add descriptor to the prepare queue in order to be able
2248 	 * to free them later in terminate_all
2249 	 */
2250 	list_add_tail(&desc->node, &chan->prepare_queue);
2251 
2252 	spin_unlock_irqrestore(&chan->lock, flags);
2253 
2254 	return &desc->txd;
2255  free_desc:
2256 	d40_desc_free(chan, desc);
2257  unlock:
2258 	spin_unlock_irqrestore(&chan->lock, flags);
2259 	return NULL;
2260 }
2261 
2262 bool stedma40_filter(struct dma_chan *chan, void *data)
2263 {
2264 	struct stedma40_chan_cfg *info = data;
2265 	struct d40_chan *d40c =
2266 		container_of(chan, struct d40_chan, chan);
2267 	int err;
2268 
2269 	if (data) {
2270 		err = d40_validate_conf(d40c, info);
2271 		if (!err)
2272 			d40c->dma_cfg = *info;
2273 	} else
2274 		err = d40_config_memcpy(d40c);
2275 
2276 	if (!err)
2277 		d40c->configured = true;
2278 
2279 	return err == 0;
2280 }
2281 EXPORT_SYMBOL(stedma40_filter);
2282 
2283 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2284 {
2285 	bool realtime = d40c->dma_cfg.realtime;
2286 	bool highprio = d40c->dma_cfg.high_priority;
2287 	u32 rtreg;
2288 	u32 event = D40_TYPE_TO_EVENT(dev_type);
2289 	u32 group = D40_TYPE_TO_GROUP(dev_type);
2290 	u32 bit = BIT(event);
2291 	u32 prioreg;
2292 	struct d40_gen_dmac *dmac = &d40c->base->gen_dmac;
2293 
2294 	rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear;
2295 	/*
2296 	 * Due to a hardware bug, in some cases a logical channel triggered by
2297 	 * a high priority destination event line can generate extra packet
2298 	 * transactions.
2299 	 *
2300 	 * The workaround is to not set the high priority level for the
2301 	 * destination event lines that trigger logical channels.
2302 	 */
2303 	if (!src && chan_is_logical(d40c))
2304 		highprio = false;
2305 
2306 	prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear;
2307 
2308 	/* Destination event lines are stored in the upper halfword */
2309 	if (!src)
2310 		bit <<= 16;
2311 
2312 	writel(bit, d40c->base->virtbase + prioreg + group * 4);
2313 	writel(bit, d40c->base->virtbase + rtreg + group * 4);
2314 }
2315 
2316 static void d40_set_prio_realtime(struct d40_chan *d40c)
2317 {
2318 	if (d40c->base->rev < 3)
2319 		return;
2320 
2321 	if ((d40c->dma_cfg.dir ==  DMA_DEV_TO_MEM) ||
2322 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2323 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, true);
2324 
2325 	if ((d40c->dma_cfg.dir ==  DMA_MEM_TO_DEV) ||
2326 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2327 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, false);
2328 }
2329 
2330 #define D40_DT_FLAGS_MODE(flags)       ((flags >> 0) & 0x1)
2331 #define D40_DT_FLAGS_DIR(flags)        ((flags >> 1) & 0x1)
2332 #define D40_DT_FLAGS_BIG_ENDIAN(flags) ((flags >> 2) & 0x1)
2333 #define D40_DT_FLAGS_FIXED_CHAN(flags) ((flags >> 3) & 0x1)
2334 #define D40_DT_FLAGS_HIGH_PRIO(flags)  ((flags >> 4) & 0x1)
2335 
2336 static struct dma_chan *d40_xlate(struct of_phandle_args *dma_spec,
2337 				  struct of_dma *ofdma)
2338 {
2339 	struct stedma40_chan_cfg cfg;
2340 	dma_cap_mask_t cap;
2341 	u32 flags;
2342 
2343 	memset(&cfg, 0, sizeof(struct stedma40_chan_cfg));
2344 
2345 	dma_cap_zero(cap);
2346 	dma_cap_set(DMA_SLAVE, cap);
2347 
2348 	cfg.dev_type = dma_spec->args[0];
2349 	flags = dma_spec->args[2];
2350 
2351 	switch (D40_DT_FLAGS_MODE(flags)) {
2352 	case 0: cfg.mode = STEDMA40_MODE_LOGICAL; break;
2353 	case 1: cfg.mode = STEDMA40_MODE_PHYSICAL; break;
2354 	}
2355 
2356 	switch (D40_DT_FLAGS_DIR(flags)) {
2357 	case 0:
2358 		cfg.dir = DMA_MEM_TO_DEV;
2359 		cfg.dst_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2360 		break;
2361 	case 1:
2362 		cfg.dir = DMA_DEV_TO_MEM;
2363 		cfg.src_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2364 		break;
2365 	}
2366 
2367 	if (D40_DT_FLAGS_FIXED_CHAN(flags)) {
2368 		cfg.phy_channel = dma_spec->args[1];
2369 		cfg.use_fixed_channel = true;
2370 	}
2371 
2372 	if (D40_DT_FLAGS_HIGH_PRIO(flags))
2373 		cfg.high_priority = true;
2374 
2375 	return dma_request_channel(cap, stedma40_filter, &cfg);
2376 }
2377 
2378 /* DMA ENGINE functions */
2379 static int d40_alloc_chan_resources(struct dma_chan *chan)
2380 {
2381 	int err;
2382 	unsigned long flags;
2383 	struct d40_chan *d40c =
2384 		container_of(chan, struct d40_chan, chan);
2385 	bool is_free_phy;
2386 	spin_lock_irqsave(&d40c->lock, flags);
2387 
2388 	dma_cookie_init(chan);
2389 
2390 	/* If no dma configuration is set use default configuration (memcpy) */
2391 	if (!d40c->configured) {
2392 		err = d40_config_memcpy(d40c);
2393 		if (err) {
2394 			chan_err(d40c, "Failed to configure memcpy channel\n");
2395 			goto mark_last_busy;
2396 		}
2397 	}
2398 
2399 	err = d40_allocate_channel(d40c, &is_free_phy);
2400 	if (err) {
2401 		chan_err(d40c, "Failed to allocate channel\n");
2402 		d40c->configured = false;
2403 		goto mark_last_busy;
2404 	}
2405 
2406 	pm_runtime_get_sync(d40c->base->dev);
2407 
2408 	d40_set_prio_realtime(d40c);
2409 
2410 	if (chan_is_logical(d40c)) {
2411 		if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2412 			d40c->lcpa = d40c->base->lcpa_base +
2413 				d40c->dma_cfg.dev_type * D40_LCPA_CHAN_SIZE;
2414 		else
2415 			d40c->lcpa = d40c->base->lcpa_base +
2416 				d40c->dma_cfg.dev_type *
2417 				D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2418 
2419 		/* Unmask the Global Interrupt Mask. */
2420 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2421 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2422 	}
2423 
2424 	dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2425 		 chan_is_logical(d40c) ? "logical" : "physical",
2426 		 d40c->phy_chan->num,
2427 		 d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2428 
2429 
2430 	/*
2431 	 * Only write channel configuration to the DMA if the physical
2432 	 * resource is free. In case of multiple logical channels
2433 	 * on the same physical resource, only the first write is necessary.
2434 	 */
2435 	if (is_free_phy)
2436 		d40_config_write(d40c);
2437  mark_last_busy:
2438 	pm_runtime_mark_last_busy(d40c->base->dev);
2439 	pm_runtime_put_autosuspend(d40c->base->dev);
2440 	spin_unlock_irqrestore(&d40c->lock, flags);
2441 	return err;
2442 }
2443 
2444 static void d40_free_chan_resources(struct dma_chan *chan)
2445 {
2446 	struct d40_chan *d40c =
2447 		container_of(chan, struct d40_chan, chan);
2448 	int err;
2449 	unsigned long flags;
2450 
2451 	if (d40c->phy_chan == NULL) {
2452 		chan_err(d40c, "Cannot free unallocated channel\n");
2453 		return;
2454 	}
2455 
2456 	spin_lock_irqsave(&d40c->lock, flags);
2457 
2458 	err = d40_free_dma(d40c);
2459 
2460 	if (err)
2461 		chan_err(d40c, "Failed to free channel\n");
2462 	spin_unlock_irqrestore(&d40c->lock, flags);
2463 }
2464 
2465 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2466 						       dma_addr_t dst,
2467 						       dma_addr_t src,
2468 						       size_t size,
2469 						       unsigned long dma_flags)
2470 {
2471 	struct scatterlist dst_sg;
2472 	struct scatterlist src_sg;
2473 
2474 	sg_init_table(&dst_sg, 1);
2475 	sg_init_table(&src_sg, 1);
2476 
2477 	sg_dma_address(&dst_sg) = dst;
2478 	sg_dma_address(&src_sg) = src;
2479 
2480 	sg_dma_len(&dst_sg) = size;
2481 	sg_dma_len(&src_sg) = size;
2482 
2483 	return d40_prep_sg(chan, &src_sg, &dst_sg, 1,
2484 			   DMA_MEM_TO_MEM, dma_flags);
2485 }
2486 
2487 static struct dma_async_tx_descriptor *
2488 d40_prep_memcpy_sg(struct dma_chan *chan,
2489 		   struct scatterlist *dst_sg, unsigned int dst_nents,
2490 		   struct scatterlist *src_sg, unsigned int src_nents,
2491 		   unsigned long dma_flags)
2492 {
2493 	if (dst_nents != src_nents)
2494 		return NULL;
2495 
2496 	return d40_prep_sg(chan, src_sg, dst_sg, src_nents,
2497 			   DMA_MEM_TO_MEM, dma_flags);
2498 }
2499 
2500 static struct dma_async_tx_descriptor *
2501 d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2502 		  unsigned int sg_len, enum dma_transfer_direction direction,
2503 		  unsigned long dma_flags, void *context)
2504 {
2505 	if (!is_slave_direction(direction))
2506 		return NULL;
2507 
2508 	return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2509 }
2510 
2511 static struct dma_async_tx_descriptor *
2512 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2513 		     size_t buf_len, size_t period_len,
2514 		     enum dma_transfer_direction direction, unsigned long flags)
2515 {
2516 	unsigned int periods = buf_len / period_len;
2517 	struct dma_async_tx_descriptor *txd;
2518 	struct scatterlist *sg;
2519 	int i;
2520 
2521 	sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2522 	if (!sg)
2523 		return NULL;
2524 
2525 	for (i = 0; i < periods; i++) {
2526 		sg_dma_address(&sg[i]) = dma_addr;
2527 		sg_dma_len(&sg[i]) = period_len;
2528 		dma_addr += period_len;
2529 	}
2530 
2531 	sg[periods].offset = 0;
2532 	sg_dma_len(&sg[periods]) = 0;
2533 	sg[periods].page_link =
2534 		((unsigned long)sg | 0x01) & ~0x02;
2535 
2536 	txd = d40_prep_sg(chan, sg, sg, periods, direction,
2537 			  DMA_PREP_INTERRUPT);
2538 
2539 	kfree(sg);
2540 
2541 	return txd;
2542 }
2543 
2544 static enum dma_status d40_tx_status(struct dma_chan *chan,
2545 				     dma_cookie_t cookie,
2546 				     struct dma_tx_state *txstate)
2547 {
2548 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2549 	enum dma_status ret;
2550 
2551 	if (d40c->phy_chan == NULL) {
2552 		chan_err(d40c, "Cannot read status of unallocated channel\n");
2553 		return -EINVAL;
2554 	}
2555 
2556 	ret = dma_cookie_status(chan, cookie, txstate);
2557 	if (ret != DMA_COMPLETE && txstate)
2558 		dma_set_residue(txstate, stedma40_residue(chan));
2559 
2560 	if (d40_is_paused(d40c))
2561 		ret = DMA_PAUSED;
2562 
2563 	return ret;
2564 }
2565 
2566 static void d40_issue_pending(struct dma_chan *chan)
2567 {
2568 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2569 	unsigned long flags;
2570 
2571 	if (d40c->phy_chan == NULL) {
2572 		chan_err(d40c, "Channel is not allocated!\n");
2573 		return;
2574 	}
2575 
2576 	spin_lock_irqsave(&d40c->lock, flags);
2577 
2578 	list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2579 
2580 	/* Busy means that queued jobs are already being processed */
2581 	if (!d40c->busy)
2582 		(void) d40_queue_start(d40c);
2583 
2584 	spin_unlock_irqrestore(&d40c->lock, flags);
2585 }
2586 
2587 static int d40_terminate_all(struct dma_chan *chan)
2588 {
2589 	unsigned long flags;
2590 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2591 	int ret;
2592 
2593 	if (d40c->phy_chan == NULL) {
2594 		chan_err(d40c, "Channel is not allocated!\n");
2595 		return -EINVAL;
2596 	}
2597 
2598 	spin_lock_irqsave(&d40c->lock, flags);
2599 
2600 	pm_runtime_get_sync(d40c->base->dev);
2601 	ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2602 	if (ret)
2603 		chan_err(d40c, "Failed to stop channel\n");
2604 
2605 	d40_term_all(d40c);
2606 	pm_runtime_mark_last_busy(d40c->base->dev);
2607 	pm_runtime_put_autosuspend(d40c->base->dev);
2608 	if (d40c->busy) {
2609 		pm_runtime_mark_last_busy(d40c->base->dev);
2610 		pm_runtime_put_autosuspend(d40c->base->dev);
2611 	}
2612 	d40c->busy = false;
2613 
2614 	spin_unlock_irqrestore(&d40c->lock, flags);
2615 	return 0;
2616 }
2617 
2618 static int
2619 dma40_config_to_halfchannel(struct d40_chan *d40c,
2620 			    struct stedma40_half_channel_info *info,
2621 			    u32 maxburst)
2622 {
2623 	int psize;
2624 
2625 	if (chan_is_logical(d40c)) {
2626 		if (maxburst >= 16)
2627 			psize = STEDMA40_PSIZE_LOG_16;
2628 		else if (maxburst >= 8)
2629 			psize = STEDMA40_PSIZE_LOG_8;
2630 		else if (maxburst >= 4)
2631 			psize = STEDMA40_PSIZE_LOG_4;
2632 		else
2633 			psize = STEDMA40_PSIZE_LOG_1;
2634 	} else {
2635 		if (maxburst >= 16)
2636 			psize = STEDMA40_PSIZE_PHY_16;
2637 		else if (maxburst >= 8)
2638 			psize = STEDMA40_PSIZE_PHY_8;
2639 		else if (maxburst >= 4)
2640 			psize = STEDMA40_PSIZE_PHY_4;
2641 		else
2642 			psize = STEDMA40_PSIZE_PHY_1;
2643 	}
2644 
2645 	info->psize = psize;
2646 	info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2647 
2648 	return 0;
2649 }
2650 
2651 /* Runtime reconfiguration extension */
2652 static int d40_set_runtime_config(struct dma_chan *chan,
2653 				  struct dma_slave_config *config)
2654 {
2655 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2656 	struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2657 	enum dma_slave_buswidth src_addr_width, dst_addr_width;
2658 	dma_addr_t config_addr;
2659 	u32 src_maxburst, dst_maxburst;
2660 	int ret;
2661 
2662 	if (d40c->phy_chan == NULL) {
2663 		chan_err(d40c, "Channel is not allocated!\n");
2664 		return -EINVAL;
2665 	}
2666 
2667 	src_addr_width = config->src_addr_width;
2668 	src_maxburst = config->src_maxburst;
2669 	dst_addr_width = config->dst_addr_width;
2670 	dst_maxburst = config->dst_maxburst;
2671 
2672 	if (config->direction == DMA_DEV_TO_MEM) {
2673 		config_addr = config->src_addr;
2674 
2675 		if (cfg->dir != DMA_DEV_TO_MEM)
2676 			dev_dbg(d40c->base->dev,
2677 				"channel was not configured for peripheral "
2678 				"to memory transfer (%d) overriding\n",
2679 				cfg->dir);
2680 		cfg->dir = DMA_DEV_TO_MEM;
2681 
2682 		/* Configure the memory side */
2683 		if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2684 			dst_addr_width = src_addr_width;
2685 		if (dst_maxburst == 0)
2686 			dst_maxburst = src_maxburst;
2687 
2688 	} else if (config->direction == DMA_MEM_TO_DEV) {
2689 		config_addr = config->dst_addr;
2690 
2691 		if (cfg->dir != DMA_MEM_TO_DEV)
2692 			dev_dbg(d40c->base->dev,
2693 				"channel was not configured for memory "
2694 				"to peripheral transfer (%d) overriding\n",
2695 				cfg->dir);
2696 		cfg->dir = DMA_MEM_TO_DEV;
2697 
2698 		/* Configure the memory side */
2699 		if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2700 			src_addr_width = dst_addr_width;
2701 		if (src_maxburst == 0)
2702 			src_maxburst = dst_maxburst;
2703 	} else {
2704 		dev_err(d40c->base->dev,
2705 			"unrecognized channel direction %d\n",
2706 			config->direction);
2707 		return -EINVAL;
2708 	}
2709 
2710 	if (config_addr <= 0) {
2711 		dev_err(d40c->base->dev, "no address supplied\n");
2712 		return -EINVAL;
2713 	}
2714 
2715 	if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2716 		dev_err(d40c->base->dev,
2717 			"src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2718 			src_maxburst,
2719 			src_addr_width,
2720 			dst_maxburst,
2721 			dst_addr_width);
2722 		return -EINVAL;
2723 	}
2724 
2725 	if (src_maxburst > 16) {
2726 		src_maxburst = 16;
2727 		dst_maxburst = src_maxburst * src_addr_width / dst_addr_width;
2728 	} else if (dst_maxburst > 16) {
2729 		dst_maxburst = 16;
2730 		src_maxburst = dst_maxburst * dst_addr_width / src_addr_width;
2731 	}
2732 
2733 	/* Only valid widths are; 1, 2, 4 and 8. */
2734 	if (src_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2735 	    src_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2736 	    dst_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2737 	    dst_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2738 	    !is_power_of_2(src_addr_width) ||
2739 	    !is_power_of_2(dst_addr_width))
2740 		return -EINVAL;
2741 
2742 	cfg->src_info.data_width = src_addr_width;
2743 	cfg->dst_info.data_width = dst_addr_width;
2744 
2745 	ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2746 					  src_maxburst);
2747 	if (ret)
2748 		return ret;
2749 
2750 	ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2751 					  dst_maxburst);
2752 	if (ret)
2753 		return ret;
2754 
2755 	/* Fill in register values */
2756 	if (chan_is_logical(d40c))
2757 		d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2758 	else
2759 		d40_phy_cfg(cfg, &d40c->src_def_cfg, &d40c->dst_def_cfg);
2760 
2761 	/* These settings will take precedence later */
2762 	d40c->runtime_addr = config_addr;
2763 	d40c->runtime_direction = config->direction;
2764 	dev_dbg(d40c->base->dev,
2765 		"configured channel %s for %s, data width %d/%d, "
2766 		"maxburst %d/%d elements, LE, no flow control\n",
2767 		dma_chan_name(chan),
2768 		(config->direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2769 		src_addr_width, dst_addr_width,
2770 		src_maxburst, dst_maxburst);
2771 
2772 	return 0;
2773 }
2774 
2775 /* Initialization functions */
2776 
2777 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2778 				 struct d40_chan *chans, int offset,
2779 				 int num_chans)
2780 {
2781 	int i = 0;
2782 	struct d40_chan *d40c;
2783 
2784 	INIT_LIST_HEAD(&dma->channels);
2785 
2786 	for (i = offset; i < offset + num_chans; i++) {
2787 		d40c = &chans[i];
2788 		d40c->base = base;
2789 		d40c->chan.device = dma;
2790 
2791 		spin_lock_init(&d40c->lock);
2792 
2793 		d40c->log_num = D40_PHY_CHAN;
2794 
2795 		INIT_LIST_HEAD(&d40c->done);
2796 		INIT_LIST_HEAD(&d40c->active);
2797 		INIT_LIST_HEAD(&d40c->queue);
2798 		INIT_LIST_HEAD(&d40c->pending_queue);
2799 		INIT_LIST_HEAD(&d40c->client);
2800 		INIT_LIST_HEAD(&d40c->prepare_queue);
2801 
2802 		tasklet_init(&d40c->tasklet, dma_tasklet,
2803 			     (unsigned long) d40c);
2804 
2805 		list_add_tail(&d40c->chan.device_node,
2806 			      &dma->channels);
2807 	}
2808 }
2809 
2810 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2811 {
2812 	if (dma_has_cap(DMA_SLAVE, dev->cap_mask)) {
2813 		dev->device_prep_slave_sg = d40_prep_slave_sg;
2814 		dev->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
2815 	}
2816 
2817 	if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2818 		dev->device_prep_dma_memcpy = d40_prep_memcpy;
2819 		dev->directions = BIT(DMA_MEM_TO_MEM);
2820 		/*
2821 		 * This controller can only access address at even
2822 		 * 32bit boundaries, i.e. 2^2
2823 		 */
2824 		dev->copy_align = DMAENGINE_ALIGN_4_BYTES;
2825 	}
2826 
2827 	if (dma_has_cap(DMA_SG, dev->cap_mask))
2828 		dev->device_prep_dma_sg = d40_prep_memcpy_sg;
2829 
2830 	if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2831 		dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2832 
2833 	dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2834 	dev->device_free_chan_resources = d40_free_chan_resources;
2835 	dev->device_issue_pending = d40_issue_pending;
2836 	dev->device_tx_status = d40_tx_status;
2837 	dev->device_config = d40_set_runtime_config;
2838 	dev->device_pause = d40_pause;
2839 	dev->device_resume = d40_resume;
2840 	dev->device_terminate_all = d40_terminate_all;
2841 	dev->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
2842 	dev->dev = base->dev;
2843 }
2844 
2845 static int __init d40_dmaengine_init(struct d40_base *base,
2846 				     int num_reserved_chans)
2847 {
2848 	int err ;
2849 
2850 	d40_chan_init(base, &base->dma_slave, base->log_chans,
2851 		      0, base->num_log_chans);
2852 
2853 	dma_cap_zero(base->dma_slave.cap_mask);
2854 	dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2855 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2856 
2857 	d40_ops_init(base, &base->dma_slave);
2858 
2859 	err = dma_async_device_register(&base->dma_slave);
2860 
2861 	if (err) {
2862 		d40_err(base->dev, "Failed to register slave channels\n");
2863 		goto exit;
2864 	}
2865 
2866 	d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2867 		      base->num_log_chans, base->num_memcpy_chans);
2868 
2869 	dma_cap_zero(base->dma_memcpy.cap_mask);
2870 	dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2871 	dma_cap_set(DMA_SG, base->dma_memcpy.cap_mask);
2872 
2873 	d40_ops_init(base, &base->dma_memcpy);
2874 
2875 	err = dma_async_device_register(&base->dma_memcpy);
2876 
2877 	if (err) {
2878 		d40_err(base->dev,
2879 			"Failed to register memcpy only channels\n");
2880 		goto unregister_slave;
2881 	}
2882 
2883 	d40_chan_init(base, &base->dma_both, base->phy_chans,
2884 		      0, num_reserved_chans);
2885 
2886 	dma_cap_zero(base->dma_both.cap_mask);
2887 	dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2888 	dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2889 	dma_cap_set(DMA_SG, base->dma_both.cap_mask);
2890 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2891 
2892 	d40_ops_init(base, &base->dma_both);
2893 	err = dma_async_device_register(&base->dma_both);
2894 
2895 	if (err) {
2896 		d40_err(base->dev,
2897 			"Failed to register logical and physical capable channels\n");
2898 		goto unregister_memcpy;
2899 	}
2900 	return 0;
2901  unregister_memcpy:
2902 	dma_async_device_unregister(&base->dma_memcpy);
2903  unregister_slave:
2904 	dma_async_device_unregister(&base->dma_slave);
2905  exit:
2906 	return err;
2907 }
2908 
2909 /* Suspend resume functionality */
2910 #ifdef CONFIG_PM_SLEEP
2911 static int dma40_suspend(struct device *dev)
2912 {
2913 	struct platform_device *pdev = to_platform_device(dev);
2914 	struct d40_base *base = platform_get_drvdata(pdev);
2915 	int ret;
2916 
2917 	ret = pm_runtime_force_suspend(dev);
2918 	if (ret)
2919 		return ret;
2920 
2921 	if (base->lcpa_regulator)
2922 		ret = regulator_disable(base->lcpa_regulator);
2923 	return ret;
2924 }
2925 
2926 static int dma40_resume(struct device *dev)
2927 {
2928 	struct platform_device *pdev = to_platform_device(dev);
2929 	struct d40_base *base = platform_get_drvdata(pdev);
2930 	int ret = 0;
2931 
2932 	if (base->lcpa_regulator) {
2933 		ret = regulator_enable(base->lcpa_regulator);
2934 		if (ret)
2935 			return ret;
2936 	}
2937 
2938 	return pm_runtime_force_resume(dev);
2939 }
2940 #endif
2941 
2942 #ifdef CONFIG_PM
2943 static void dma40_backup(void __iomem *baseaddr, u32 *backup,
2944 			 u32 *regaddr, int num, bool save)
2945 {
2946 	int i;
2947 
2948 	for (i = 0; i < num; i++) {
2949 		void __iomem *addr = baseaddr + regaddr[i];
2950 
2951 		if (save)
2952 			backup[i] = readl_relaxed(addr);
2953 		else
2954 			writel_relaxed(backup[i], addr);
2955 	}
2956 }
2957 
2958 static void d40_save_restore_registers(struct d40_base *base, bool save)
2959 {
2960 	int i;
2961 
2962 	/* Save/Restore channel specific registers */
2963 	for (i = 0; i < base->num_phy_chans; i++) {
2964 		void __iomem *addr;
2965 		int idx;
2966 
2967 		if (base->phy_res[i].reserved)
2968 			continue;
2969 
2970 		addr = base->virtbase + D40_DREG_PCBASE + i * D40_DREG_PCDELTA;
2971 		idx = i * ARRAY_SIZE(d40_backup_regs_chan);
2972 
2973 		dma40_backup(addr, &base->reg_val_backup_chan[idx],
2974 			     d40_backup_regs_chan,
2975 			     ARRAY_SIZE(d40_backup_regs_chan),
2976 			     save);
2977 	}
2978 
2979 	/* Save/Restore global registers */
2980 	dma40_backup(base->virtbase, base->reg_val_backup,
2981 		     d40_backup_regs, ARRAY_SIZE(d40_backup_regs),
2982 		     save);
2983 
2984 	/* Save/Restore registers only existing on dma40 v3 and later */
2985 	if (base->gen_dmac.backup)
2986 		dma40_backup(base->virtbase, base->reg_val_backup_v4,
2987 			     base->gen_dmac.backup,
2988 			base->gen_dmac.backup_size,
2989 			save);
2990 }
2991 
2992 static int dma40_runtime_suspend(struct device *dev)
2993 {
2994 	struct platform_device *pdev = to_platform_device(dev);
2995 	struct d40_base *base = platform_get_drvdata(pdev);
2996 
2997 	d40_save_restore_registers(base, true);
2998 
2999 	/* Don't disable/enable clocks for v1 due to HW bugs */
3000 	if (base->rev != 1)
3001 		writel_relaxed(base->gcc_pwr_off_mask,
3002 			       base->virtbase + D40_DREG_GCC);
3003 
3004 	return 0;
3005 }
3006 
3007 static int dma40_runtime_resume(struct device *dev)
3008 {
3009 	struct platform_device *pdev = to_platform_device(dev);
3010 	struct d40_base *base = platform_get_drvdata(pdev);
3011 
3012 	d40_save_restore_registers(base, false);
3013 
3014 	writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
3015 		       base->virtbase + D40_DREG_GCC);
3016 	return 0;
3017 }
3018 #endif
3019 
3020 static const struct dev_pm_ops dma40_pm_ops = {
3021 	SET_LATE_SYSTEM_SLEEP_PM_OPS(dma40_suspend, dma40_resume)
3022 	SET_RUNTIME_PM_OPS(dma40_runtime_suspend,
3023 				dma40_runtime_resume,
3024 				NULL)
3025 };
3026 
3027 /* Initialization functions. */
3028 
3029 static int __init d40_phy_res_init(struct d40_base *base)
3030 {
3031 	int i;
3032 	int num_phy_chans_avail = 0;
3033 	u32 val[2];
3034 	int odd_even_bit = -2;
3035 	int gcc = D40_DREG_GCC_ENA;
3036 
3037 	val[0] = readl(base->virtbase + D40_DREG_PRSME);
3038 	val[1] = readl(base->virtbase + D40_DREG_PRSMO);
3039 
3040 	for (i = 0; i < base->num_phy_chans; i++) {
3041 		base->phy_res[i].num = i;
3042 		odd_even_bit += 2 * ((i % 2) == 0);
3043 		if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
3044 			/* Mark security only channels as occupied */
3045 			base->phy_res[i].allocated_src = D40_ALLOC_PHY;
3046 			base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
3047 			base->phy_res[i].reserved = true;
3048 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3049 						       D40_DREG_GCC_SRC);
3050 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3051 						       D40_DREG_GCC_DST);
3052 
3053 
3054 		} else {
3055 			base->phy_res[i].allocated_src = D40_ALLOC_FREE;
3056 			base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
3057 			base->phy_res[i].reserved = false;
3058 			num_phy_chans_avail++;
3059 		}
3060 		spin_lock_init(&base->phy_res[i].lock);
3061 	}
3062 
3063 	/* Mark disabled channels as occupied */
3064 	for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
3065 		int chan = base->plat_data->disabled_channels[i];
3066 
3067 		base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
3068 		base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
3069 		base->phy_res[chan].reserved = true;
3070 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3071 					       D40_DREG_GCC_SRC);
3072 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3073 					       D40_DREG_GCC_DST);
3074 		num_phy_chans_avail--;
3075 	}
3076 
3077 	/* Mark soft_lli channels */
3078 	for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) {
3079 		int chan = base->plat_data->soft_lli_chans[i];
3080 
3081 		base->phy_res[chan].use_soft_lli = true;
3082 	}
3083 
3084 	dev_info(base->dev, "%d of %d physical DMA channels available\n",
3085 		 num_phy_chans_avail, base->num_phy_chans);
3086 
3087 	/* Verify settings extended vs standard */
3088 	val[0] = readl(base->virtbase + D40_DREG_PRTYP);
3089 
3090 	for (i = 0; i < base->num_phy_chans; i++) {
3091 
3092 		if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
3093 		    (val[0] & 0x3) != 1)
3094 			dev_info(base->dev,
3095 				 "[%s] INFO: channel %d is misconfigured (%d)\n",
3096 				 __func__, i, val[0] & 0x3);
3097 
3098 		val[0] = val[0] >> 2;
3099 	}
3100 
3101 	/*
3102 	 * To keep things simple, Enable all clocks initially.
3103 	 * The clocks will get managed later post channel allocation.
3104 	 * The clocks for the event lines on which reserved channels exists
3105 	 * are not managed here.
3106 	 */
3107 	writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3108 	base->gcc_pwr_off_mask = gcc;
3109 
3110 	return num_phy_chans_avail;
3111 }
3112 
3113 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
3114 {
3115 	struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3116 	struct clk *clk;
3117 	void __iomem *virtbase;
3118 	struct resource *res;
3119 	struct d40_base *base;
3120 	int num_log_chans;
3121 	int num_phy_chans;
3122 	int num_memcpy_chans;
3123 	int clk_ret = -EINVAL;
3124 	int i;
3125 	u32 pid;
3126 	u32 cid;
3127 	u8 rev;
3128 
3129 	clk = clk_get(&pdev->dev, NULL);
3130 	if (IS_ERR(clk)) {
3131 		d40_err(&pdev->dev, "No matching clock found\n");
3132 		goto check_prepare_enabled;
3133 	}
3134 
3135 	clk_ret = clk_prepare_enable(clk);
3136 	if (clk_ret) {
3137 		d40_err(&pdev->dev, "Failed to prepare/enable clock\n");
3138 		goto disable_unprepare;
3139 	}
3140 
3141 	/* Get IO for DMAC base address */
3142 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
3143 	if (!res)
3144 		goto disable_unprepare;
3145 
3146 	if (request_mem_region(res->start, resource_size(res),
3147 			       D40_NAME " I/O base") == NULL)
3148 		goto release_region;
3149 
3150 	virtbase = ioremap(res->start, resource_size(res));
3151 	if (!virtbase)
3152 		goto release_region;
3153 
3154 	/* This is just a regular AMBA PrimeCell ID actually */
3155 	for (pid = 0, i = 0; i < 4; i++)
3156 		pid |= (readl(virtbase + resource_size(res) - 0x20 + 4 * i)
3157 			& 255) << (i * 8);
3158 	for (cid = 0, i = 0; i < 4; i++)
3159 		cid |= (readl(virtbase + resource_size(res) - 0x10 + 4 * i)
3160 			& 255) << (i * 8);
3161 
3162 	if (cid != AMBA_CID) {
3163 		d40_err(&pdev->dev, "Unknown hardware! No PrimeCell ID\n");
3164 		goto unmap_io;
3165 	}
3166 	if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
3167 		d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
3168 			AMBA_MANF_BITS(pid),
3169 			AMBA_VENDOR_ST);
3170 		goto unmap_io;
3171 	}
3172 	/*
3173 	 * HW revision:
3174 	 * DB8500ed has revision 0
3175 	 * ? has revision 1
3176 	 * DB8500v1 has revision 2
3177 	 * DB8500v2 has revision 3
3178 	 * AP9540v1 has revision 4
3179 	 * DB8540v1 has revision 4
3180 	 */
3181 	rev = AMBA_REV_BITS(pid);
3182 	if (rev < 2) {
3183 		d40_err(&pdev->dev, "hardware revision: %d is not supported", rev);
3184 		goto unmap_io;
3185 	}
3186 
3187 	/* The number of physical channels on this HW */
3188 	if (plat_data->num_of_phy_chans)
3189 		num_phy_chans = plat_data->num_of_phy_chans;
3190 	else
3191 		num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
3192 
3193 	/* The number of channels used for memcpy */
3194 	if (plat_data->num_of_memcpy_chans)
3195 		num_memcpy_chans = plat_data->num_of_memcpy_chans;
3196 	else
3197 		num_memcpy_chans = ARRAY_SIZE(dma40_memcpy_channels);
3198 
3199 	num_log_chans = num_phy_chans * D40_MAX_LOG_CHAN_PER_PHY;
3200 
3201 	dev_info(&pdev->dev,
3202 		 "hardware rev: %d @ %pa with %d physical and %d logical channels\n",
3203 		 rev, &res->start, num_phy_chans, num_log_chans);
3204 
3205 	base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
3206 		       (num_phy_chans + num_log_chans + num_memcpy_chans) *
3207 		       sizeof(struct d40_chan), GFP_KERNEL);
3208 
3209 	if (base == NULL)
3210 		goto unmap_io;
3211 
3212 	base->rev = rev;
3213 	base->clk = clk;
3214 	base->num_memcpy_chans = num_memcpy_chans;
3215 	base->num_phy_chans = num_phy_chans;
3216 	base->num_log_chans = num_log_chans;
3217 	base->phy_start = res->start;
3218 	base->phy_size = resource_size(res);
3219 	base->virtbase = virtbase;
3220 	base->plat_data = plat_data;
3221 	base->dev = &pdev->dev;
3222 	base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3223 	base->log_chans = &base->phy_chans[num_phy_chans];
3224 
3225 	if (base->plat_data->num_of_phy_chans == 14) {
3226 		base->gen_dmac.backup = d40_backup_regs_v4b;
3227 		base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
3228 		base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
3229 		base->gen_dmac.interrupt_clear = D40_DREG_CPCICR;
3230 		base->gen_dmac.realtime_en = D40_DREG_CRSEG1;
3231 		base->gen_dmac.realtime_clear = D40_DREG_CRCEG1;
3232 		base->gen_dmac.high_prio_en = D40_DREG_CPSEG1;
3233 		base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1;
3234 		base->gen_dmac.il = il_v4b;
3235 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4b);
3236 		base->gen_dmac.init_reg = dma_init_reg_v4b;
3237 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
3238 	} else {
3239 		if (base->rev >= 3) {
3240 			base->gen_dmac.backup = d40_backup_regs_v4a;
3241 			base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
3242 		}
3243 		base->gen_dmac.interrupt_en = D40_DREG_PCMIS;
3244 		base->gen_dmac.interrupt_clear = D40_DREG_PCICR;
3245 		base->gen_dmac.realtime_en = D40_DREG_RSEG1;
3246 		base->gen_dmac.realtime_clear = D40_DREG_RCEG1;
3247 		base->gen_dmac.high_prio_en = D40_DREG_PSEG1;
3248 		base->gen_dmac.high_prio_clear = D40_DREG_PCEG1;
3249 		base->gen_dmac.il = il_v4a;
3250 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4a);
3251 		base->gen_dmac.init_reg = dma_init_reg_v4a;
3252 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a);
3253 	}
3254 
3255 	base->phy_res = kcalloc(num_phy_chans,
3256 				sizeof(*base->phy_res),
3257 				GFP_KERNEL);
3258 	if (!base->phy_res)
3259 		goto free_base;
3260 
3261 	base->lookup_phy_chans = kcalloc(num_phy_chans,
3262 					 sizeof(*base->lookup_phy_chans),
3263 					 GFP_KERNEL);
3264 	if (!base->lookup_phy_chans)
3265 		goto free_phy_res;
3266 
3267 	base->lookup_log_chans = kcalloc(num_log_chans,
3268 					 sizeof(*base->lookup_log_chans),
3269 					 GFP_KERNEL);
3270 	if (!base->lookup_log_chans)
3271 		goto free_phy_chans;
3272 
3273 	base->reg_val_backup_chan = kmalloc_array(base->num_phy_chans,
3274 						  sizeof(d40_backup_regs_chan),
3275 						  GFP_KERNEL);
3276 	if (!base->reg_val_backup_chan)
3277 		goto free_log_chans;
3278 
3279 	base->lcla_pool.alloc_map = kcalloc(num_phy_chans
3280 					    * D40_LCLA_LINK_PER_EVENT_GRP,
3281 					    sizeof(*base->lcla_pool.alloc_map),
3282 					    GFP_KERNEL);
3283 	if (!base->lcla_pool.alloc_map)
3284 		goto free_backup_chan;
3285 
3286 	base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3287 					    0, SLAB_HWCACHE_ALIGN,
3288 					    NULL);
3289 	if (base->desc_slab == NULL)
3290 		goto free_map;
3291 
3292 	return base;
3293  free_map:
3294 	kfree(base->lcla_pool.alloc_map);
3295  free_backup_chan:
3296 	kfree(base->reg_val_backup_chan);
3297  free_log_chans:
3298 	kfree(base->lookup_log_chans);
3299  free_phy_chans:
3300 	kfree(base->lookup_phy_chans);
3301  free_phy_res:
3302 	kfree(base->phy_res);
3303  free_base:
3304 	kfree(base);
3305  unmap_io:
3306 	iounmap(virtbase);
3307  release_region:
3308 	release_mem_region(res->start, resource_size(res));
3309  check_prepare_enabled:
3310 	if (!clk_ret)
3311  disable_unprepare:
3312 		clk_disable_unprepare(clk);
3313 	if (!IS_ERR(clk))
3314 		clk_put(clk);
3315 	return NULL;
3316 }
3317 
3318 static void __init d40_hw_init(struct d40_base *base)
3319 {
3320 
3321 	int i;
3322 	u32 prmseo[2] = {0, 0};
3323 	u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3324 	u32 pcmis = 0;
3325 	u32 pcicr = 0;
3326 	struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg;
3327 	u32 reg_size = base->gen_dmac.init_reg_size;
3328 
3329 	for (i = 0; i < reg_size; i++)
3330 		writel(dma_init_reg[i].val,
3331 		       base->virtbase + dma_init_reg[i].reg);
3332 
3333 	/* Configure all our dma channels to default settings */
3334 	for (i = 0; i < base->num_phy_chans; i++) {
3335 
3336 		activeo[i % 2] = activeo[i % 2] << 2;
3337 
3338 		if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3339 		    == D40_ALLOC_PHY) {
3340 			activeo[i % 2] |= 3;
3341 			continue;
3342 		}
3343 
3344 		/* Enable interrupt # */
3345 		pcmis = (pcmis << 1) | 1;
3346 
3347 		/* Clear interrupt # */
3348 		pcicr = (pcicr << 1) | 1;
3349 
3350 		/* Set channel to physical mode */
3351 		prmseo[i % 2] = prmseo[i % 2] << 2;
3352 		prmseo[i % 2] |= 1;
3353 
3354 	}
3355 
3356 	writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3357 	writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3358 	writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3359 	writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3360 
3361 	/* Write which interrupt to enable */
3362 	writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en);
3363 
3364 	/* Write which interrupt to clear */
3365 	writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear);
3366 
3367 	/* These are __initdata and cannot be accessed after init */
3368 	base->gen_dmac.init_reg = NULL;
3369 	base->gen_dmac.init_reg_size = 0;
3370 }
3371 
3372 static int __init d40_lcla_allocate(struct d40_base *base)
3373 {
3374 	struct d40_lcla_pool *pool = &base->lcla_pool;
3375 	unsigned long *page_list;
3376 	int i, j;
3377 	int ret;
3378 
3379 	/*
3380 	 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3381 	 * To full fill this hardware requirement without wasting 256 kb
3382 	 * we allocate pages until we get an aligned one.
3383 	 */
3384 	page_list = kmalloc_array(MAX_LCLA_ALLOC_ATTEMPTS,
3385 				  sizeof(*page_list),
3386 				  GFP_KERNEL);
3387 	if (!page_list)
3388 		return -ENOMEM;
3389 
3390 	/* Calculating how many pages that are required */
3391 	base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3392 
3393 	for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3394 		page_list[i] = __get_free_pages(GFP_KERNEL,
3395 						base->lcla_pool.pages);
3396 		if (!page_list[i]) {
3397 
3398 			d40_err(base->dev, "Failed to allocate %d pages.\n",
3399 				base->lcla_pool.pages);
3400 			ret = -ENOMEM;
3401 
3402 			for (j = 0; j < i; j++)
3403 				free_pages(page_list[j], base->lcla_pool.pages);
3404 			goto free_page_list;
3405 		}
3406 
3407 		if ((virt_to_phys((void *)page_list[i]) &
3408 		     (LCLA_ALIGNMENT - 1)) == 0)
3409 			break;
3410 	}
3411 
3412 	for (j = 0; j < i; j++)
3413 		free_pages(page_list[j], base->lcla_pool.pages);
3414 
3415 	if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3416 		base->lcla_pool.base = (void *)page_list[i];
3417 	} else {
3418 		/*
3419 		 * After many attempts and no succees with finding the correct
3420 		 * alignment, try with allocating a big buffer.
3421 		 */
3422 		dev_warn(base->dev,
3423 			 "[%s] Failed to get %d pages @ 18 bit align.\n",
3424 			 __func__, base->lcla_pool.pages);
3425 		base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3426 							 base->num_phy_chans +
3427 							 LCLA_ALIGNMENT,
3428 							 GFP_KERNEL);
3429 		if (!base->lcla_pool.base_unaligned) {
3430 			ret = -ENOMEM;
3431 			goto free_page_list;
3432 		}
3433 
3434 		base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3435 						 LCLA_ALIGNMENT);
3436 	}
3437 
3438 	pool->dma_addr = dma_map_single(base->dev, pool->base,
3439 					SZ_1K * base->num_phy_chans,
3440 					DMA_TO_DEVICE);
3441 	if (dma_mapping_error(base->dev, pool->dma_addr)) {
3442 		pool->dma_addr = 0;
3443 		ret = -ENOMEM;
3444 		goto free_page_list;
3445 	}
3446 
3447 	writel(virt_to_phys(base->lcla_pool.base),
3448 	       base->virtbase + D40_DREG_LCLA);
3449 	ret = 0;
3450  free_page_list:
3451 	kfree(page_list);
3452 	return ret;
3453 }
3454 
3455 static int __init d40_of_probe(struct platform_device *pdev,
3456 			       struct device_node *np)
3457 {
3458 	struct stedma40_platform_data *pdata;
3459 	int num_phy = 0, num_memcpy = 0, num_disabled = 0;
3460 	const __be32 *list;
3461 
3462 	pdata = devm_kzalloc(&pdev->dev, sizeof(*pdata), GFP_KERNEL);
3463 	if (!pdata)
3464 		return -ENOMEM;
3465 
3466 	/* If absent this value will be obtained from h/w. */
3467 	of_property_read_u32(np, "dma-channels", &num_phy);
3468 	if (num_phy > 0)
3469 		pdata->num_of_phy_chans = num_phy;
3470 
3471 	list = of_get_property(np, "memcpy-channels", &num_memcpy);
3472 	num_memcpy /= sizeof(*list);
3473 
3474 	if (num_memcpy > D40_MEMCPY_MAX_CHANS || num_memcpy <= 0) {
3475 		d40_err(&pdev->dev,
3476 			"Invalid number of memcpy channels specified (%d)\n",
3477 			num_memcpy);
3478 		return -EINVAL;
3479 	}
3480 	pdata->num_of_memcpy_chans = num_memcpy;
3481 
3482 	of_property_read_u32_array(np, "memcpy-channels",
3483 				   dma40_memcpy_channels,
3484 				   num_memcpy);
3485 
3486 	list = of_get_property(np, "disabled-channels", &num_disabled);
3487 	num_disabled /= sizeof(*list);
3488 
3489 	if (num_disabled >= STEDMA40_MAX_PHYS || num_disabled < 0) {
3490 		d40_err(&pdev->dev,
3491 			"Invalid number of disabled channels specified (%d)\n",
3492 			num_disabled);
3493 		return -EINVAL;
3494 	}
3495 
3496 	of_property_read_u32_array(np, "disabled-channels",
3497 				   pdata->disabled_channels,
3498 				   num_disabled);
3499 	pdata->disabled_channels[num_disabled] = -1;
3500 
3501 	pdev->dev.platform_data = pdata;
3502 
3503 	return 0;
3504 }
3505 
3506 static int __init d40_probe(struct platform_device *pdev)
3507 {
3508 	struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3509 	struct device_node *np = pdev->dev.of_node;
3510 	int ret = -ENOENT;
3511 	struct d40_base *base;
3512 	struct resource *res;
3513 	int num_reserved_chans;
3514 	u32 val;
3515 
3516 	if (!plat_data) {
3517 		if (np) {
3518 			if (d40_of_probe(pdev, np)) {
3519 				ret = -ENOMEM;
3520 				goto report_failure;
3521 			}
3522 		} else {
3523 			d40_err(&pdev->dev, "No pdata or Device Tree provided\n");
3524 			goto report_failure;
3525 		}
3526 	}
3527 
3528 	base = d40_hw_detect_init(pdev);
3529 	if (!base)
3530 		goto report_failure;
3531 
3532 	num_reserved_chans = d40_phy_res_init(base);
3533 
3534 	platform_set_drvdata(pdev, base);
3535 
3536 	spin_lock_init(&base->interrupt_lock);
3537 	spin_lock_init(&base->execmd_lock);
3538 
3539 	/* Get IO for logical channel parameter address */
3540 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
3541 	if (!res) {
3542 		ret = -ENOENT;
3543 		d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
3544 		goto destroy_cache;
3545 	}
3546 	base->lcpa_size = resource_size(res);
3547 	base->phy_lcpa = res->start;
3548 
3549 	if (request_mem_region(res->start, resource_size(res),
3550 			       D40_NAME " I/O lcpa") == NULL) {
3551 		ret = -EBUSY;
3552 		d40_err(&pdev->dev, "Failed to request LCPA region %pR\n", res);
3553 		goto destroy_cache;
3554 	}
3555 
3556 	/* We make use of ESRAM memory for this. */
3557 	val = readl(base->virtbase + D40_DREG_LCPA);
3558 	if (res->start != val && val != 0) {
3559 		dev_warn(&pdev->dev,
3560 			 "[%s] Mismatch LCPA dma 0x%x, def %pa\n",
3561 			 __func__, val, &res->start);
3562 	} else
3563 		writel(res->start, base->virtbase + D40_DREG_LCPA);
3564 
3565 	base->lcpa_base = ioremap(res->start, resource_size(res));
3566 	if (!base->lcpa_base) {
3567 		ret = -ENOMEM;
3568 		d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
3569 		goto destroy_cache;
3570 	}
3571 	/* If lcla has to be located in ESRAM we don't need to allocate */
3572 	if (base->plat_data->use_esram_lcla) {
3573 		res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3574 							"lcla_esram");
3575 		if (!res) {
3576 			ret = -ENOENT;
3577 			d40_err(&pdev->dev,
3578 				"No \"lcla_esram\" memory resource\n");
3579 			goto destroy_cache;
3580 		}
3581 		base->lcla_pool.base = ioremap(res->start,
3582 						resource_size(res));
3583 		if (!base->lcla_pool.base) {
3584 			ret = -ENOMEM;
3585 			d40_err(&pdev->dev, "Failed to ioremap LCLA region\n");
3586 			goto destroy_cache;
3587 		}
3588 		writel(res->start, base->virtbase + D40_DREG_LCLA);
3589 
3590 	} else {
3591 		ret = d40_lcla_allocate(base);
3592 		if (ret) {
3593 			d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
3594 			goto destroy_cache;
3595 		}
3596 	}
3597 
3598 	spin_lock_init(&base->lcla_pool.lock);
3599 
3600 	base->irq = platform_get_irq(pdev, 0);
3601 
3602 	ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3603 	if (ret) {
3604 		d40_err(&pdev->dev, "No IRQ defined\n");
3605 		goto destroy_cache;
3606 	}
3607 
3608 	if (base->plat_data->use_esram_lcla) {
3609 
3610 		base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3611 		if (IS_ERR(base->lcpa_regulator)) {
3612 			d40_err(&pdev->dev, "Failed to get lcpa_regulator\n");
3613 			ret = PTR_ERR(base->lcpa_regulator);
3614 			base->lcpa_regulator = NULL;
3615 			goto destroy_cache;
3616 		}
3617 
3618 		ret = regulator_enable(base->lcpa_regulator);
3619 		if (ret) {
3620 			d40_err(&pdev->dev,
3621 				"Failed to enable lcpa_regulator\n");
3622 			regulator_put(base->lcpa_regulator);
3623 			base->lcpa_regulator = NULL;
3624 			goto destroy_cache;
3625 		}
3626 	}
3627 
3628 	writel_relaxed(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3629 
3630 	pm_runtime_irq_safe(base->dev);
3631 	pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3632 	pm_runtime_use_autosuspend(base->dev);
3633 	pm_runtime_mark_last_busy(base->dev);
3634 	pm_runtime_set_active(base->dev);
3635 	pm_runtime_enable(base->dev);
3636 
3637 	ret = d40_dmaengine_init(base, num_reserved_chans);
3638 	if (ret)
3639 		goto destroy_cache;
3640 
3641 	base->dev->dma_parms = &base->dma_parms;
3642 	ret = dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
3643 	if (ret) {
3644 		d40_err(&pdev->dev, "Failed to set dma max seg size\n");
3645 		goto destroy_cache;
3646 	}
3647 
3648 	d40_hw_init(base);
3649 
3650 	if (np) {
3651 		ret = of_dma_controller_register(np, d40_xlate, NULL);
3652 		if (ret)
3653 			dev_err(&pdev->dev,
3654 				"could not register of_dma_controller\n");
3655 	}
3656 
3657 	dev_info(base->dev, "initialized\n");
3658 	return 0;
3659  destroy_cache:
3660 	kmem_cache_destroy(base->desc_slab);
3661 	if (base->virtbase)
3662 		iounmap(base->virtbase);
3663 
3664 	if (base->lcla_pool.base && base->plat_data->use_esram_lcla) {
3665 		iounmap(base->lcla_pool.base);
3666 		base->lcla_pool.base = NULL;
3667 	}
3668 
3669 	if (base->lcla_pool.dma_addr)
3670 		dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3671 				 SZ_1K * base->num_phy_chans,
3672 				 DMA_TO_DEVICE);
3673 
3674 	if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3675 		free_pages((unsigned long)base->lcla_pool.base,
3676 			   base->lcla_pool.pages);
3677 
3678 	kfree(base->lcla_pool.base_unaligned);
3679 
3680 	if (base->phy_lcpa)
3681 		release_mem_region(base->phy_lcpa,
3682 				   base->lcpa_size);
3683 	if (base->phy_start)
3684 		release_mem_region(base->phy_start,
3685 				   base->phy_size);
3686 	if (base->clk) {
3687 		clk_disable_unprepare(base->clk);
3688 		clk_put(base->clk);
3689 	}
3690 
3691 	if (base->lcpa_regulator) {
3692 		regulator_disable(base->lcpa_regulator);
3693 		regulator_put(base->lcpa_regulator);
3694 	}
3695 
3696 	kfree(base->lcla_pool.alloc_map);
3697 	kfree(base->lookup_log_chans);
3698 	kfree(base->lookup_phy_chans);
3699 	kfree(base->phy_res);
3700 	kfree(base);
3701  report_failure:
3702 	d40_err(&pdev->dev, "probe failed\n");
3703 	return ret;
3704 }
3705 
3706 static const struct of_device_id d40_match[] = {
3707         { .compatible = "stericsson,dma40", },
3708         {}
3709 };
3710 
3711 static struct platform_driver d40_driver = {
3712 	.driver = {
3713 		.name  = D40_NAME,
3714 		.pm = &dma40_pm_ops,
3715 		.of_match_table = d40_match,
3716 	},
3717 };
3718 
3719 static int __init stedma40_init(void)
3720 {
3721 	return platform_driver_probe(&d40_driver, d40_probe);
3722 }
3723 subsys_initcall(stedma40_init);
3724