xref: /openbmc/linux/drivers/dma/ste_dma40.c (revision 12eb4683)
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  * @initialized: true if the dma has been initialized
560  * @gen_dmac: the struct for generic registers values to represent u8500/8540
561  * DMA controller
562  */
563 struct d40_base {
564 	spinlock_t			 interrupt_lock;
565 	spinlock_t			 execmd_lock;
566 	struct device			 *dev;
567 	void __iomem			 *virtbase;
568 	u8				  rev:4;
569 	struct clk			 *clk;
570 	phys_addr_t			  phy_start;
571 	resource_size_t			  phy_size;
572 	int				  irq;
573 	int				  num_memcpy_chans;
574 	int				  num_phy_chans;
575 	int				  num_log_chans;
576 	struct device_dma_parameters	  dma_parms;
577 	struct dma_device		  dma_both;
578 	struct dma_device		  dma_slave;
579 	struct dma_device		  dma_memcpy;
580 	struct d40_chan			 *phy_chans;
581 	struct d40_chan			 *log_chans;
582 	struct d40_chan			**lookup_log_chans;
583 	struct d40_chan			**lookup_phy_chans;
584 	struct stedma40_platform_data	 *plat_data;
585 	struct regulator		 *lcpa_regulator;
586 	/* Physical half channels */
587 	struct d40_phy_res		 *phy_res;
588 	struct d40_lcla_pool		  lcla_pool;
589 	void				 *lcpa_base;
590 	dma_addr_t			  phy_lcpa;
591 	resource_size_t			  lcpa_size;
592 	struct kmem_cache		 *desc_slab;
593 	u32				  reg_val_backup[BACKUP_REGS_SZ];
594 	u32				  reg_val_backup_v4[BACKUP_REGS_SZ_MAX];
595 	u32				 *reg_val_backup_chan;
596 	u16				  gcc_pwr_off_mask;
597 	bool				  initialized;
598 	struct d40_gen_dmac		  gen_dmac;
599 };
600 
601 static struct device *chan2dev(struct d40_chan *d40c)
602 {
603 	return &d40c->chan.dev->device;
604 }
605 
606 static bool chan_is_physical(struct d40_chan *chan)
607 {
608 	return chan->log_num == D40_PHY_CHAN;
609 }
610 
611 static bool chan_is_logical(struct d40_chan *chan)
612 {
613 	return !chan_is_physical(chan);
614 }
615 
616 static void __iomem *chan_base(struct d40_chan *chan)
617 {
618 	return chan->base->virtbase + D40_DREG_PCBASE +
619 	       chan->phy_chan->num * D40_DREG_PCDELTA;
620 }
621 
622 #define d40_err(dev, format, arg...)		\
623 	dev_err(dev, "[%s] " format, __func__, ## arg)
624 
625 #define chan_err(d40c, format, arg...)		\
626 	d40_err(chan2dev(d40c), format, ## arg)
627 
628 static int d40_pool_lli_alloc(struct d40_chan *d40c, struct d40_desc *d40d,
629 			      int lli_len)
630 {
631 	bool is_log = chan_is_logical(d40c);
632 	u32 align;
633 	void *base;
634 
635 	if (is_log)
636 		align = sizeof(struct d40_log_lli);
637 	else
638 		align = sizeof(struct d40_phy_lli);
639 
640 	if (lli_len == 1) {
641 		base = d40d->lli_pool.pre_alloc_lli;
642 		d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
643 		d40d->lli_pool.base = NULL;
644 	} else {
645 		d40d->lli_pool.size = lli_len * 2 * align;
646 
647 		base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
648 		d40d->lli_pool.base = base;
649 
650 		if (d40d->lli_pool.base == NULL)
651 			return -ENOMEM;
652 	}
653 
654 	if (is_log) {
655 		d40d->lli_log.src = PTR_ALIGN(base, align);
656 		d40d->lli_log.dst = d40d->lli_log.src + lli_len;
657 
658 		d40d->lli_pool.dma_addr = 0;
659 	} else {
660 		d40d->lli_phy.src = PTR_ALIGN(base, align);
661 		d40d->lli_phy.dst = d40d->lli_phy.src + lli_len;
662 
663 		d40d->lli_pool.dma_addr = dma_map_single(d40c->base->dev,
664 							 d40d->lli_phy.src,
665 							 d40d->lli_pool.size,
666 							 DMA_TO_DEVICE);
667 
668 		if (dma_mapping_error(d40c->base->dev,
669 				      d40d->lli_pool.dma_addr)) {
670 			kfree(d40d->lli_pool.base);
671 			d40d->lli_pool.base = NULL;
672 			d40d->lli_pool.dma_addr = 0;
673 			return -ENOMEM;
674 		}
675 	}
676 
677 	return 0;
678 }
679 
680 static void d40_pool_lli_free(struct d40_chan *d40c, struct d40_desc *d40d)
681 {
682 	if (d40d->lli_pool.dma_addr)
683 		dma_unmap_single(d40c->base->dev, d40d->lli_pool.dma_addr,
684 				 d40d->lli_pool.size, DMA_TO_DEVICE);
685 
686 	kfree(d40d->lli_pool.base);
687 	d40d->lli_pool.base = NULL;
688 	d40d->lli_pool.size = 0;
689 	d40d->lli_log.src = NULL;
690 	d40d->lli_log.dst = NULL;
691 	d40d->lli_phy.src = NULL;
692 	d40d->lli_phy.dst = NULL;
693 }
694 
695 static int d40_lcla_alloc_one(struct d40_chan *d40c,
696 			      struct d40_desc *d40d)
697 {
698 	unsigned long flags;
699 	int i;
700 	int ret = -EINVAL;
701 
702 	spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
703 
704 	/*
705 	 * Allocate both src and dst at the same time, therefore the half
706 	 * start on 1 since 0 can't be used since zero is used as end marker.
707 	 */
708 	for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
709 		int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
710 
711 		if (!d40c->base->lcla_pool.alloc_map[idx]) {
712 			d40c->base->lcla_pool.alloc_map[idx] = d40d;
713 			d40d->lcla_alloc++;
714 			ret = i;
715 			break;
716 		}
717 	}
718 
719 	spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
720 
721 	return ret;
722 }
723 
724 static int d40_lcla_free_all(struct d40_chan *d40c,
725 			     struct d40_desc *d40d)
726 {
727 	unsigned long flags;
728 	int i;
729 	int ret = -EINVAL;
730 
731 	if (chan_is_physical(d40c))
732 		return 0;
733 
734 	spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
735 
736 	for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
737 		int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
738 
739 		if (d40c->base->lcla_pool.alloc_map[idx] == d40d) {
740 			d40c->base->lcla_pool.alloc_map[idx] = NULL;
741 			d40d->lcla_alloc--;
742 			if (d40d->lcla_alloc == 0) {
743 				ret = 0;
744 				break;
745 			}
746 		}
747 	}
748 
749 	spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
750 
751 	return ret;
752 
753 }
754 
755 static void d40_desc_remove(struct d40_desc *d40d)
756 {
757 	list_del(&d40d->node);
758 }
759 
760 static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
761 {
762 	struct d40_desc *desc = NULL;
763 
764 	if (!list_empty(&d40c->client)) {
765 		struct d40_desc *d;
766 		struct d40_desc *_d;
767 
768 		list_for_each_entry_safe(d, _d, &d40c->client, node) {
769 			if (async_tx_test_ack(&d->txd)) {
770 				d40_desc_remove(d);
771 				desc = d;
772 				memset(desc, 0, sizeof(*desc));
773 				break;
774 			}
775 		}
776 	}
777 
778 	if (!desc)
779 		desc = kmem_cache_zalloc(d40c->base->desc_slab, GFP_NOWAIT);
780 
781 	if (desc)
782 		INIT_LIST_HEAD(&desc->node);
783 
784 	return desc;
785 }
786 
787 static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
788 {
789 
790 	d40_pool_lli_free(d40c, d40d);
791 	d40_lcla_free_all(d40c, d40d);
792 	kmem_cache_free(d40c->base->desc_slab, d40d);
793 }
794 
795 static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
796 {
797 	list_add_tail(&desc->node, &d40c->active);
798 }
799 
800 static void d40_phy_lli_load(struct d40_chan *chan, struct d40_desc *desc)
801 {
802 	struct d40_phy_lli *lli_dst = desc->lli_phy.dst;
803 	struct d40_phy_lli *lli_src = desc->lli_phy.src;
804 	void __iomem *base = chan_base(chan);
805 
806 	writel(lli_src->reg_cfg, base + D40_CHAN_REG_SSCFG);
807 	writel(lli_src->reg_elt, base + D40_CHAN_REG_SSELT);
808 	writel(lli_src->reg_ptr, base + D40_CHAN_REG_SSPTR);
809 	writel(lli_src->reg_lnk, base + D40_CHAN_REG_SSLNK);
810 
811 	writel(lli_dst->reg_cfg, base + D40_CHAN_REG_SDCFG);
812 	writel(lli_dst->reg_elt, base + D40_CHAN_REG_SDELT);
813 	writel(lli_dst->reg_ptr, base + D40_CHAN_REG_SDPTR);
814 	writel(lli_dst->reg_lnk, base + D40_CHAN_REG_SDLNK);
815 }
816 
817 static void d40_desc_done(struct d40_chan *d40c, struct d40_desc *desc)
818 {
819 	list_add_tail(&desc->node, &d40c->done);
820 }
821 
822 static void d40_log_lli_to_lcxa(struct d40_chan *chan, struct d40_desc *desc)
823 {
824 	struct d40_lcla_pool *pool = &chan->base->lcla_pool;
825 	struct d40_log_lli_bidir *lli = &desc->lli_log;
826 	int lli_current = desc->lli_current;
827 	int lli_len = desc->lli_len;
828 	bool cyclic = desc->cyclic;
829 	int curr_lcla = -EINVAL;
830 	int first_lcla = 0;
831 	bool use_esram_lcla = chan->base->plat_data->use_esram_lcla;
832 	bool linkback;
833 
834 	/*
835 	 * We may have partially running cyclic transfers, in case we did't get
836 	 * enough LCLA entries.
837 	 */
838 	linkback = cyclic && lli_current == 0;
839 
840 	/*
841 	 * For linkback, we need one LCLA even with only one link, because we
842 	 * can't link back to the one in LCPA space
843 	 */
844 	if (linkback || (lli_len - lli_current > 1)) {
845 		/*
846 		 * If the channel is expected to use only soft_lli don't
847 		 * allocate a lcla. This is to avoid a HW issue that exists
848 		 * in some controller during a peripheral to memory transfer
849 		 * that uses linked lists.
850 		 */
851 		if (!(chan->phy_chan->use_soft_lli &&
852 			chan->dma_cfg.dir == DMA_DEV_TO_MEM))
853 			curr_lcla = d40_lcla_alloc_one(chan, desc);
854 
855 		first_lcla = curr_lcla;
856 	}
857 
858 	/*
859 	 * For linkback, we normally load the LCPA in the loop since we need to
860 	 * link it to the second LCLA and not the first.  However, if we
861 	 * couldn't even get a first LCLA, then we have to run in LCPA and
862 	 * reload manually.
863 	 */
864 	if (!linkback || curr_lcla == -EINVAL) {
865 		unsigned int flags = 0;
866 
867 		if (curr_lcla == -EINVAL)
868 			flags |= LLI_TERM_INT;
869 
870 		d40_log_lli_lcpa_write(chan->lcpa,
871 				       &lli->dst[lli_current],
872 				       &lli->src[lli_current],
873 				       curr_lcla,
874 				       flags);
875 		lli_current++;
876 	}
877 
878 	if (curr_lcla < 0)
879 		goto out;
880 
881 	for (; lli_current < lli_len; lli_current++) {
882 		unsigned int lcla_offset = chan->phy_chan->num * 1024 +
883 					   8 * curr_lcla * 2;
884 		struct d40_log_lli *lcla = pool->base + lcla_offset;
885 		unsigned int flags = 0;
886 		int next_lcla;
887 
888 		if (lli_current + 1 < lli_len)
889 			next_lcla = d40_lcla_alloc_one(chan, desc);
890 		else
891 			next_lcla = linkback ? first_lcla : -EINVAL;
892 
893 		if (cyclic || next_lcla == -EINVAL)
894 			flags |= LLI_TERM_INT;
895 
896 		if (linkback && curr_lcla == first_lcla) {
897 			/* First link goes in both LCPA and LCLA */
898 			d40_log_lli_lcpa_write(chan->lcpa,
899 					       &lli->dst[lli_current],
900 					       &lli->src[lli_current],
901 					       next_lcla, flags);
902 		}
903 
904 		/*
905 		 * One unused LCLA in the cyclic case if the very first
906 		 * next_lcla fails...
907 		 */
908 		d40_log_lli_lcla_write(lcla,
909 				       &lli->dst[lli_current],
910 				       &lli->src[lli_current],
911 				       next_lcla, flags);
912 
913 		/*
914 		 * Cache maintenance is not needed if lcla is
915 		 * mapped in esram
916 		 */
917 		if (!use_esram_lcla) {
918 			dma_sync_single_range_for_device(chan->base->dev,
919 						pool->dma_addr, lcla_offset,
920 						2 * sizeof(struct d40_log_lli),
921 						DMA_TO_DEVICE);
922 		}
923 		curr_lcla = next_lcla;
924 
925 		if (curr_lcla == -EINVAL || curr_lcla == first_lcla) {
926 			lli_current++;
927 			break;
928 		}
929 	}
930 
931 out:
932 	desc->lli_current = lli_current;
933 }
934 
935 static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
936 {
937 	if (chan_is_physical(d40c)) {
938 		d40_phy_lli_load(d40c, d40d);
939 		d40d->lli_current = d40d->lli_len;
940 	} else
941 		d40_log_lli_to_lcxa(d40c, d40d);
942 }
943 
944 static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
945 {
946 	struct d40_desc *d;
947 
948 	if (list_empty(&d40c->active))
949 		return NULL;
950 
951 	d = list_first_entry(&d40c->active,
952 			     struct d40_desc,
953 			     node);
954 	return d;
955 }
956 
957 /* remove desc from current queue and add it to the pending_queue */
958 static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
959 {
960 	d40_desc_remove(desc);
961 	desc->is_in_client_list = false;
962 	list_add_tail(&desc->node, &d40c->pending_queue);
963 }
964 
965 static struct d40_desc *d40_first_pending(struct d40_chan *d40c)
966 {
967 	struct d40_desc *d;
968 
969 	if (list_empty(&d40c->pending_queue))
970 		return NULL;
971 
972 	d = list_first_entry(&d40c->pending_queue,
973 			     struct d40_desc,
974 			     node);
975 	return d;
976 }
977 
978 static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
979 {
980 	struct d40_desc *d;
981 
982 	if (list_empty(&d40c->queue))
983 		return NULL;
984 
985 	d = list_first_entry(&d40c->queue,
986 			     struct d40_desc,
987 			     node);
988 	return d;
989 }
990 
991 static struct d40_desc *d40_first_done(struct d40_chan *d40c)
992 {
993 	if (list_empty(&d40c->done))
994 		return NULL;
995 
996 	return list_first_entry(&d40c->done, struct d40_desc, node);
997 }
998 
999 static int d40_psize_2_burst_size(bool is_log, int psize)
1000 {
1001 	if (is_log) {
1002 		if (psize == STEDMA40_PSIZE_LOG_1)
1003 			return 1;
1004 	} else {
1005 		if (psize == STEDMA40_PSIZE_PHY_1)
1006 			return 1;
1007 	}
1008 
1009 	return 2 << psize;
1010 }
1011 
1012 /*
1013  * The dma only supports transmitting packages up to
1014  * STEDMA40_MAX_SEG_SIZE * data_width, where data_width is stored in Bytes.
1015  *
1016  * Calculate the total number of dma elements required to send the entire sg list.
1017  */
1018 static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2)
1019 {
1020 	int dmalen;
1021 	u32 max_w = max(data_width1, data_width2);
1022 	u32 min_w = min(data_width1, data_width2);
1023 	u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE * min_w, max_w);
1024 
1025 	if (seg_max > STEDMA40_MAX_SEG_SIZE)
1026 		seg_max -= max_w;
1027 
1028 	if (!IS_ALIGNED(size, max_w))
1029 		return -EINVAL;
1030 
1031 	if (size <= seg_max)
1032 		dmalen = 1;
1033 	else {
1034 		dmalen = size / seg_max;
1035 		if (dmalen * seg_max < size)
1036 			dmalen++;
1037 	}
1038 	return dmalen;
1039 }
1040 
1041 static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len,
1042 			   u32 data_width1, u32 data_width2)
1043 {
1044 	struct scatterlist *sg;
1045 	int i;
1046 	int len = 0;
1047 	int ret;
1048 
1049 	for_each_sg(sgl, sg, sg_len, i) {
1050 		ret = d40_size_2_dmalen(sg_dma_len(sg),
1051 					data_width1, data_width2);
1052 		if (ret < 0)
1053 			return ret;
1054 		len += ret;
1055 	}
1056 	return len;
1057 }
1058 
1059 
1060 #ifdef CONFIG_PM
1061 static void dma40_backup(void __iomem *baseaddr, u32 *backup,
1062 			 u32 *regaddr, int num, bool save)
1063 {
1064 	int i;
1065 
1066 	for (i = 0; i < num; i++) {
1067 		void __iomem *addr = baseaddr + regaddr[i];
1068 
1069 		if (save)
1070 			backup[i] = readl_relaxed(addr);
1071 		else
1072 			writel_relaxed(backup[i], addr);
1073 	}
1074 }
1075 
1076 static void d40_save_restore_registers(struct d40_base *base, bool save)
1077 {
1078 	int i;
1079 
1080 	/* Save/Restore channel specific registers */
1081 	for (i = 0; i < base->num_phy_chans; i++) {
1082 		void __iomem *addr;
1083 		int idx;
1084 
1085 		if (base->phy_res[i].reserved)
1086 			continue;
1087 
1088 		addr = base->virtbase + D40_DREG_PCBASE + i * D40_DREG_PCDELTA;
1089 		idx = i * ARRAY_SIZE(d40_backup_regs_chan);
1090 
1091 		dma40_backup(addr, &base->reg_val_backup_chan[idx],
1092 			     d40_backup_regs_chan,
1093 			     ARRAY_SIZE(d40_backup_regs_chan),
1094 			     save);
1095 	}
1096 
1097 	/* Save/Restore global registers */
1098 	dma40_backup(base->virtbase, base->reg_val_backup,
1099 		     d40_backup_regs, ARRAY_SIZE(d40_backup_regs),
1100 		     save);
1101 
1102 	/* Save/Restore registers only existing on dma40 v3 and later */
1103 	if (base->gen_dmac.backup)
1104 		dma40_backup(base->virtbase, base->reg_val_backup_v4,
1105 			     base->gen_dmac.backup,
1106 			base->gen_dmac.backup_size,
1107 			save);
1108 }
1109 #else
1110 static void d40_save_restore_registers(struct d40_base *base, bool save)
1111 {
1112 }
1113 #endif
1114 
1115 static int __d40_execute_command_phy(struct d40_chan *d40c,
1116 				     enum d40_command command)
1117 {
1118 	u32 status;
1119 	int i;
1120 	void __iomem *active_reg;
1121 	int ret = 0;
1122 	unsigned long flags;
1123 	u32 wmask;
1124 
1125 	if (command == D40_DMA_STOP) {
1126 		ret = __d40_execute_command_phy(d40c, D40_DMA_SUSPEND_REQ);
1127 		if (ret)
1128 			return ret;
1129 	}
1130 
1131 	spin_lock_irqsave(&d40c->base->execmd_lock, flags);
1132 
1133 	if (d40c->phy_chan->num % 2 == 0)
1134 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1135 	else
1136 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1137 
1138 	if (command == D40_DMA_SUSPEND_REQ) {
1139 		status = (readl(active_reg) &
1140 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1141 			D40_CHAN_POS(d40c->phy_chan->num);
1142 
1143 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1144 			goto done;
1145 	}
1146 
1147 	wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
1148 	writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
1149 	       active_reg);
1150 
1151 	if (command == D40_DMA_SUSPEND_REQ) {
1152 
1153 		for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
1154 			status = (readl(active_reg) &
1155 				  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1156 				D40_CHAN_POS(d40c->phy_chan->num);
1157 
1158 			cpu_relax();
1159 			/*
1160 			 * Reduce the number of bus accesses while
1161 			 * waiting for the DMA to suspend.
1162 			 */
1163 			udelay(3);
1164 
1165 			if (status == D40_DMA_STOP ||
1166 			    status == D40_DMA_SUSPENDED)
1167 				break;
1168 		}
1169 
1170 		if (i == D40_SUSPEND_MAX_IT) {
1171 			chan_err(d40c,
1172 				"unable to suspend the chl %d (log: %d) status %x\n",
1173 				d40c->phy_chan->num, d40c->log_num,
1174 				status);
1175 			dump_stack();
1176 			ret = -EBUSY;
1177 		}
1178 
1179 	}
1180 done:
1181 	spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
1182 	return ret;
1183 }
1184 
1185 static void d40_term_all(struct d40_chan *d40c)
1186 {
1187 	struct d40_desc *d40d;
1188 	struct d40_desc *_d;
1189 
1190 	/* Release completed descriptors */
1191 	while ((d40d = d40_first_done(d40c))) {
1192 		d40_desc_remove(d40d);
1193 		d40_desc_free(d40c, d40d);
1194 	}
1195 
1196 	/* Release active descriptors */
1197 	while ((d40d = d40_first_active_get(d40c))) {
1198 		d40_desc_remove(d40d);
1199 		d40_desc_free(d40c, d40d);
1200 	}
1201 
1202 	/* Release queued descriptors waiting for transfer */
1203 	while ((d40d = d40_first_queued(d40c))) {
1204 		d40_desc_remove(d40d);
1205 		d40_desc_free(d40c, d40d);
1206 	}
1207 
1208 	/* Release pending descriptors */
1209 	while ((d40d = d40_first_pending(d40c))) {
1210 		d40_desc_remove(d40d);
1211 		d40_desc_free(d40c, d40d);
1212 	}
1213 
1214 	/* Release client owned descriptors */
1215 	if (!list_empty(&d40c->client))
1216 		list_for_each_entry_safe(d40d, _d, &d40c->client, node) {
1217 			d40_desc_remove(d40d);
1218 			d40_desc_free(d40c, d40d);
1219 		}
1220 
1221 	/* Release descriptors in prepare queue */
1222 	if (!list_empty(&d40c->prepare_queue))
1223 		list_for_each_entry_safe(d40d, _d,
1224 					 &d40c->prepare_queue, node) {
1225 			d40_desc_remove(d40d);
1226 			d40_desc_free(d40c, d40d);
1227 		}
1228 
1229 	d40c->pending_tx = 0;
1230 }
1231 
1232 static void __d40_config_set_event(struct d40_chan *d40c,
1233 				   enum d40_events event_type, u32 event,
1234 				   int reg)
1235 {
1236 	void __iomem *addr = chan_base(d40c) + reg;
1237 	int tries;
1238 	u32 status;
1239 
1240 	switch (event_type) {
1241 
1242 	case D40_DEACTIVATE_EVENTLINE:
1243 
1244 		writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
1245 		       | ~D40_EVENTLINE_MASK(event), addr);
1246 		break;
1247 
1248 	case D40_SUSPEND_REQ_EVENTLINE:
1249 		status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1250 			  D40_EVENTLINE_POS(event);
1251 
1252 		if (status == D40_DEACTIVATE_EVENTLINE ||
1253 		    status == D40_SUSPEND_REQ_EVENTLINE)
1254 			break;
1255 
1256 		writel((D40_SUSPEND_REQ_EVENTLINE << D40_EVENTLINE_POS(event))
1257 		       | ~D40_EVENTLINE_MASK(event), addr);
1258 
1259 		for (tries = 0 ; tries < D40_SUSPEND_MAX_IT; tries++) {
1260 
1261 			status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1262 				  D40_EVENTLINE_POS(event);
1263 
1264 			cpu_relax();
1265 			/*
1266 			 * Reduce the number of bus accesses while
1267 			 * waiting for the DMA to suspend.
1268 			 */
1269 			udelay(3);
1270 
1271 			if (status == D40_DEACTIVATE_EVENTLINE)
1272 				break;
1273 		}
1274 
1275 		if (tries == D40_SUSPEND_MAX_IT) {
1276 			chan_err(d40c,
1277 				"unable to stop the event_line chl %d (log: %d)"
1278 				"status %x\n", d40c->phy_chan->num,
1279 				 d40c->log_num, status);
1280 		}
1281 		break;
1282 
1283 	case D40_ACTIVATE_EVENTLINE:
1284 	/*
1285 	 * The hardware sometimes doesn't register the enable when src and dst
1286 	 * event lines are active on the same logical channel.  Retry to ensure
1287 	 * it does.  Usually only one retry is sufficient.
1288 	 */
1289 		tries = 100;
1290 		while (--tries) {
1291 			writel((D40_ACTIVATE_EVENTLINE <<
1292 				D40_EVENTLINE_POS(event)) |
1293 				~D40_EVENTLINE_MASK(event), addr);
1294 
1295 			if (readl(addr) & D40_EVENTLINE_MASK(event))
1296 				break;
1297 		}
1298 
1299 		if (tries != 99)
1300 			dev_dbg(chan2dev(d40c),
1301 				"[%s] workaround enable S%cLNK (%d tries)\n",
1302 				__func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D',
1303 				100 - tries);
1304 
1305 		WARN_ON(!tries);
1306 		break;
1307 
1308 	case D40_ROUND_EVENTLINE:
1309 		BUG();
1310 		break;
1311 
1312 	}
1313 }
1314 
1315 static void d40_config_set_event(struct d40_chan *d40c,
1316 				 enum d40_events event_type)
1317 {
1318 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
1319 
1320 	/* Enable event line connected to device (or memcpy) */
1321 	if ((d40c->dma_cfg.dir == DMA_DEV_TO_MEM) ||
1322 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
1323 		__d40_config_set_event(d40c, event_type, event,
1324 				       D40_CHAN_REG_SSLNK);
1325 
1326 	if (d40c->dma_cfg.dir !=  DMA_DEV_TO_MEM)
1327 		__d40_config_set_event(d40c, event_type, event,
1328 				       D40_CHAN_REG_SDLNK);
1329 }
1330 
1331 static u32 d40_chan_has_events(struct d40_chan *d40c)
1332 {
1333 	void __iomem *chanbase = chan_base(d40c);
1334 	u32 val;
1335 
1336 	val = readl(chanbase + D40_CHAN_REG_SSLNK);
1337 	val |= readl(chanbase + D40_CHAN_REG_SDLNK);
1338 
1339 	return val;
1340 }
1341 
1342 static int
1343 __d40_execute_command_log(struct d40_chan *d40c, enum d40_command command)
1344 {
1345 	unsigned long flags;
1346 	int ret = 0;
1347 	u32 active_status;
1348 	void __iomem *active_reg;
1349 
1350 	if (d40c->phy_chan->num % 2 == 0)
1351 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1352 	else
1353 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1354 
1355 
1356 	spin_lock_irqsave(&d40c->phy_chan->lock, flags);
1357 
1358 	switch (command) {
1359 	case D40_DMA_STOP:
1360 	case D40_DMA_SUSPEND_REQ:
1361 
1362 		active_status = (readl(active_reg) &
1363 				 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1364 				 D40_CHAN_POS(d40c->phy_chan->num);
1365 
1366 		if (active_status == D40_DMA_RUN)
1367 			d40_config_set_event(d40c, D40_SUSPEND_REQ_EVENTLINE);
1368 		else
1369 			d40_config_set_event(d40c, D40_DEACTIVATE_EVENTLINE);
1370 
1371 		if (!d40_chan_has_events(d40c) && (command == D40_DMA_STOP))
1372 			ret = __d40_execute_command_phy(d40c, command);
1373 
1374 		break;
1375 
1376 	case D40_DMA_RUN:
1377 
1378 		d40_config_set_event(d40c, D40_ACTIVATE_EVENTLINE);
1379 		ret = __d40_execute_command_phy(d40c, command);
1380 		break;
1381 
1382 	case D40_DMA_SUSPENDED:
1383 		BUG();
1384 		break;
1385 	}
1386 
1387 	spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
1388 	return ret;
1389 }
1390 
1391 static int d40_channel_execute_command(struct d40_chan *d40c,
1392 				       enum d40_command command)
1393 {
1394 	if (chan_is_logical(d40c))
1395 		return __d40_execute_command_log(d40c, command);
1396 	else
1397 		return __d40_execute_command_phy(d40c, command);
1398 }
1399 
1400 static u32 d40_get_prmo(struct d40_chan *d40c)
1401 {
1402 	static const unsigned int phy_map[] = {
1403 		[STEDMA40_PCHAN_BASIC_MODE]
1404 			= D40_DREG_PRMO_PCHAN_BASIC,
1405 		[STEDMA40_PCHAN_MODULO_MODE]
1406 			= D40_DREG_PRMO_PCHAN_MODULO,
1407 		[STEDMA40_PCHAN_DOUBLE_DST_MODE]
1408 			= D40_DREG_PRMO_PCHAN_DOUBLE_DST,
1409 	};
1410 	static const unsigned int log_map[] = {
1411 		[STEDMA40_LCHAN_SRC_PHY_DST_LOG]
1412 			= D40_DREG_PRMO_LCHAN_SRC_PHY_DST_LOG,
1413 		[STEDMA40_LCHAN_SRC_LOG_DST_PHY]
1414 			= D40_DREG_PRMO_LCHAN_SRC_LOG_DST_PHY,
1415 		[STEDMA40_LCHAN_SRC_LOG_DST_LOG]
1416 			= D40_DREG_PRMO_LCHAN_SRC_LOG_DST_LOG,
1417 	};
1418 
1419 	if (chan_is_physical(d40c))
1420 		return phy_map[d40c->dma_cfg.mode_opt];
1421 	else
1422 		return log_map[d40c->dma_cfg.mode_opt];
1423 }
1424 
1425 static void d40_config_write(struct d40_chan *d40c)
1426 {
1427 	u32 addr_base;
1428 	u32 var;
1429 
1430 	/* Odd addresses are even addresses + 4 */
1431 	addr_base = (d40c->phy_chan->num % 2) * 4;
1432 	/* Setup channel mode to logical or physical */
1433 	var = ((u32)(chan_is_logical(d40c)) + 1) <<
1434 		D40_CHAN_POS(d40c->phy_chan->num);
1435 	writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
1436 
1437 	/* Setup operational mode option register */
1438 	var = d40_get_prmo(d40c) << D40_CHAN_POS(d40c->phy_chan->num);
1439 
1440 	writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
1441 
1442 	if (chan_is_logical(d40c)) {
1443 		int lidx = (d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS)
1444 			   & D40_SREG_ELEM_LOG_LIDX_MASK;
1445 		void __iomem *chanbase = chan_base(d40c);
1446 
1447 		/* Set default config for CFG reg */
1448 		writel(d40c->src_def_cfg, chanbase + D40_CHAN_REG_SSCFG);
1449 		writel(d40c->dst_def_cfg, chanbase + D40_CHAN_REG_SDCFG);
1450 
1451 		/* Set LIDX for lcla */
1452 		writel(lidx, chanbase + D40_CHAN_REG_SSELT);
1453 		writel(lidx, chanbase + D40_CHAN_REG_SDELT);
1454 
1455 		/* Clear LNK which will be used by d40_chan_has_events() */
1456 		writel(0, chanbase + D40_CHAN_REG_SSLNK);
1457 		writel(0, chanbase + D40_CHAN_REG_SDLNK);
1458 	}
1459 }
1460 
1461 static u32 d40_residue(struct d40_chan *d40c)
1462 {
1463 	u32 num_elt;
1464 
1465 	if (chan_is_logical(d40c))
1466 		num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
1467 			>> D40_MEM_LCSP2_ECNT_POS;
1468 	else {
1469 		u32 val = readl(chan_base(d40c) + D40_CHAN_REG_SDELT);
1470 		num_elt = (val & D40_SREG_ELEM_PHY_ECNT_MASK)
1471 			  >> D40_SREG_ELEM_PHY_ECNT_POS;
1472 	}
1473 
1474 	return num_elt * d40c->dma_cfg.dst_info.data_width;
1475 }
1476 
1477 static bool d40_tx_is_linked(struct d40_chan *d40c)
1478 {
1479 	bool is_link;
1480 
1481 	if (chan_is_logical(d40c))
1482 		is_link = readl(&d40c->lcpa->lcsp3) &  D40_MEM_LCSP3_DLOS_MASK;
1483 	else
1484 		is_link = readl(chan_base(d40c) + D40_CHAN_REG_SDLNK)
1485 			  & D40_SREG_LNK_PHYS_LNK_MASK;
1486 
1487 	return is_link;
1488 }
1489 
1490 static int d40_pause(struct d40_chan *d40c)
1491 {
1492 	int res = 0;
1493 	unsigned long flags;
1494 
1495 	if (!d40c->busy)
1496 		return 0;
1497 
1498 	pm_runtime_get_sync(d40c->base->dev);
1499 	spin_lock_irqsave(&d40c->lock, flags);
1500 
1501 	res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1502 
1503 	pm_runtime_mark_last_busy(d40c->base->dev);
1504 	pm_runtime_put_autosuspend(d40c->base->dev);
1505 	spin_unlock_irqrestore(&d40c->lock, flags);
1506 	return res;
1507 }
1508 
1509 static int d40_resume(struct d40_chan *d40c)
1510 {
1511 	int res = 0;
1512 	unsigned long flags;
1513 
1514 	if (!d40c->busy)
1515 		return 0;
1516 
1517 	spin_lock_irqsave(&d40c->lock, flags);
1518 	pm_runtime_get_sync(d40c->base->dev);
1519 
1520 	/* If bytes left to transfer or linked tx resume job */
1521 	if (d40_residue(d40c) || d40_tx_is_linked(d40c))
1522 		res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1523 
1524 	pm_runtime_mark_last_busy(d40c->base->dev);
1525 	pm_runtime_put_autosuspend(d40c->base->dev);
1526 	spin_unlock_irqrestore(&d40c->lock, flags);
1527 	return res;
1528 }
1529 
1530 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
1531 {
1532 	struct d40_chan *d40c = container_of(tx->chan,
1533 					     struct d40_chan,
1534 					     chan);
1535 	struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
1536 	unsigned long flags;
1537 	dma_cookie_t cookie;
1538 
1539 	spin_lock_irqsave(&d40c->lock, flags);
1540 	cookie = dma_cookie_assign(tx);
1541 	d40_desc_queue(d40c, d40d);
1542 	spin_unlock_irqrestore(&d40c->lock, flags);
1543 
1544 	return cookie;
1545 }
1546 
1547 static int d40_start(struct d40_chan *d40c)
1548 {
1549 	return d40_channel_execute_command(d40c, D40_DMA_RUN);
1550 }
1551 
1552 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
1553 {
1554 	struct d40_desc *d40d;
1555 	int err;
1556 
1557 	/* Start queued jobs, if any */
1558 	d40d = d40_first_queued(d40c);
1559 
1560 	if (d40d != NULL) {
1561 		if (!d40c->busy) {
1562 			d40c->busy = true;
1563 			pm_runtime_get_sync(d40c->base->dev);
1564 		}
1565 
1566 		/* Remove from queue */
1567 		d40_desc_remove(d40d);
1568 
1569 		/* Add to active queue */
1570 		d40_desc_submit(d40c, d40d);
1571 
1572 		/* Initiate DMA job */
1573 		d40_desc_load(d40c, d40d);
1574 
1575 		/* Start dma job */
1576 		err = d40_start(d40c);
1577 
1578 		if (err)
1579 			return NULL;
1580 	}
1581 
1582 	return d40d;
1583 }
1584 
1585 /* called from interrupt context */
1586 static void dma_tc_handle(struct d40_chan *d40c)
1587 {
1588 	struct d40_desc *d40d;
1589 
1590 	/* Get first active entry from list */
1591 	d40d = d40_first_active_get(d40c);
1592 
1593 	if (d40d == NULL)
1594 		return;
1595 
1596 	if (d40d->cyclic) {
1597 		/*
1598 		 * If this was a paritially loaded list, we need to reloaded
1599 		 * it, and only when the list is completed.  We need to check
1600 		 * for done because the interrupt will hit for every link, and
1601 		 * not just the last one.
1602 		 */
1603 		if (d40d->lli_current < d40d->lli_len
1604 		    && !d40_tx_is_linked(d40c)
1605 		    && !d40_residue(d40c)) {
1606 			d40_lcla_free_all(d40c, d40d);
1607 			d40_desc_load(d40c, d40d);
1608 			(void) d40_start(d40c);
1609 
1610 			if (d40d->lli_current == d40d->lli_len)
1611 				d40d->lli_current = 0;
1612 		}
1613 	} else {
1614 		d40_lcla_free_all(d40c, d40d);
1615 
1616 		if (d40d->lli_current < d40d->lli_len) {
1617 			d40_desc_load(d40c, d40d);
1618 			/* Start dma job */
1619 			(void) d40_start(d40c);
1620 			return;
1621 		}
1622 
1623 		if (d40_queue_start(d40c) == NULL) {
1624 			d40c->busy = false;
1625 
1626 			pm_runtime_mark_last_busy(d40c->base->dev);
1627 			pm_runtime_put_autosuspend(d40c->base->dev);
1628 		}
1629 
1630 		d40_desc_remove(d40d);
1631 		d40_desc_done(d40c, d40d);
1632 	}
1633 
1634 	d40c->pending_tx++;
1635 	tasklet_schedule(&d40c->tasklet);
1636 
1637 }
1638 
1639 static void dma_tasklet(unsigned long data)
1640 {
1641 	struct d40_chan *d40c = (struct d40_chan *) data;
1642 	struct d40_desc *d40d;
1643 	unsigned long flags;
1644 	dma_async_tx_callback callback;
1645 	void *callback_param;
1646 
1647 	spin_lock_irqsave(&d40c->lock, flags);
1648 
1649 	/* Get first entry from the done list */
1650 	d40d = d40_first_done(d40c);
1651 	if (d40d == NULL) {
1652 		/* Check if we have reached here for cyclic job */
1653 		d40d = d40_first_active_get(d40c);
1654 		if (d40d == NULL || !d40d->cyclic)
1655 			goto err;
1656 	}
1657 
1658 	if (!d40d->cyclic)
1659 		dma_cookie_complete(&d40d->txd);
1660 
1661 	/*
1662 	 * If terminating a channel pending_tx is set to zero.
1663 	 * This prevents any finished active jobs to return to the client.
1664 	 */
1665 	if (d40c->pending_tx == 0) {
1666 		spin_unlock_irqrestore(&d40c->lock, flags);
1667 		return;
1668 	}
1669 
1670 	/* Callback to client */
1671 	callback = d40d->txd.callback;
1672 	callback_param = d40d->txd.callback_param;
1673 
1674 	if (!d40d->cyclic) {
1675 		if (async_tx_test_ack(&d40d->txd)) {
1676 			d40_desc_remove(d40d);
1677 			d40_desc_free(d40c, d40d);
1678 		} else if (!d40d->is_in_client_list) {
1679 			d40_desc_remove(d40d);
1680 			d40_lcla_free_all(d40c, d40d);
1681 			list_add_tail(&d40d->node, &d40c->client);
1682 			d40d->is_in_client_list = true;
1683 		}
1684 	}
1685 
1686 	d40c->pending_tx--;
1687 
1688 	if (d40c->pending_tx)
1689 		tasklet_schedule(&d40c->tasklet);
1690 
1691 	spin_unlock_irqrestore(&d40c->lock, flags);
1692 
1693 	if (callback && (d40d->txd.flags & DMA_PREP_INTERRUPT))
1694 		callback(callback_param);
1695 
1696 	return;
1697 
1698 err:
1699 	/* Rescue manouver if receiving double interrupts */
1700 	if (d40c->pending_tx > 0)
1701 		d40c->pending_tx--;
1702 	spin_unlock_irqrestore(&d40c->lock, flags);
1703 }
1704 
1705 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1706 {
1707 	int i;
1708 	u32 idx;
1709 	u32 row;
1710 	long chan = -1;
1711 	struct d40_chan *d40c;
1712 	unsigned long flags;
1713 	struct d40_base *base = data;
1714 	u32 regs[base->gen_dmac.il_size];
1715 	struct d40_interrupt_lookup *il = base->gen_dmac.il;
1716 	u32 il_size = base->gen_dmac.il_size;
1717 
1718 	spin_lock_irqsave(&base->interrupt_lock, flags);
1719 
1720 	/* Read interrupt status of both logical and physical channels */
1721 	for (i = 0; i < il_size; i++)
1722 		regs[i] = readl(base->virtbase + il[i].src);
1723 
1724 	for (;;) {
1725 
1726 		chan = find_next_bit((unsigned long *)regs,
1727 				     BITS_PER_LONG * il_size, chan + 1);
1728 
1729 		/* No more set bits found? */
1730 		if (chan == BITS_PER_LONG * il_size)
1731 			break;
1732 
1733 		row = chan / BITS_PER_LONG;
1734 		idx = chan & (BITS_PER_LONG - 1);
1735 
1736 		if (il[row].offset == D40_PHY_CHAN)
1737 			d40c = base->lookup_phy_chans[idx];
1738 		else
1739 			d40c = base->lookup_log_chans[il[row].offset + idx];
1740 
1741 		if (!d40c) {
1742 			/*
1743 			 * No error because this can happen if something else
1744 			 * in the system is using the channel.
1745 			 */
1746 			continue;
1747 		}
1748 
1749 		/* ACK interrupt */
1750 		writel(BIT(idx), base->virtbase + il[row].clr);
1751 
1752 		spin_lock(&d40c->lock);
1753 
1754 		if (!il[row].is_error)
1755 			dma_tc_handle(d40c);
1756 		else
1757 			d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1758 				chan, il[row].offset, idx);
1759 
1760 		spin_unlock(&d40c->lock);
1761 	}
1762 
1763 	spin_unlock_irqrestore(&base->interrupt_lock, flags);
1764 
1765 	return IRQ_HANDLED;
1766 }
1767 
1768 static int d40_validate_conf(struct d40_chan *d40c,
1769 			     struct stedma40_chan_cfg *conf)
1770 {
1771 	int res = 0;
1772 	bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1773 
1774 	if (!conf->dir) {
1775 		chan_err(d40c, "Invalid direction.\n");
1776 		res = -EINVAL;
1777 	}
1778 
1779 	if ((is_log && conf->dev_type > d40c->base->num_log_chans)  ||
1780 	    (!is_log && conf->dev_type > d40c->base->num_phy_chans) ||
1781 	    (conf->dev_type < 0)) {
1782 		chan_err(d40c, "Invalid device type (%d)\n", conf->dev_type);
1783 		res = -EINVAL;
1784 	}
1785 
1786 	if (conf->dir == DMA_DEV_TO_DEV) {
1787 		/*
1788 		 * DMAC HW supports it. Will be added to this driver,
1789 		 * in case any dma client requires it.
1790 		 */
1791 		chan_err(d40c, "periph to periph not supported\n");
1792 		res = -EINVAL;
1793 	}
1794 
1795 	if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1796 	    conf->src_info.data_width !=
1797 	    d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1798 	    conf->dst_info.data_width) {
1799 		/*
1800 		 * The DMAC hardware only supports
1801 		 * src (burst x width) == dst (burst x width)
1802 		 */
1803 
1804 		chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1805 		res = -EINVAL;
1806 	}
1807 
1808 	return res;
1809 }
1810 
1811 static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1812 			       bool is_src, int log_event_line, bool is_log,
1813 			       bool *first_user)
1814 {
1815 	unsigned long flags;
1816 	spin_lock_irqsave(&phy->lock, flags);
1817 
1818 	*first_user = ((phy->allocated_src | phy->allocated_dst)
1819 			== D40_ALLOC_FREE);
1820 
1821 	if (!is_log) {
1822 		/* Physical interrupts are masked per physical full channel */
1823 		if (phy->allocated_src == D40_ALLOC_FREE &&
1824 		    phy->allocated_dst == D40_ALLOC_FREE) {
1825 			phy->allocated_dst = D40_ALLOC_PHY;
1826 			phy->allocated_src = D40_ALLOC_PHY;
1827 			goto found;
1828 		} else
1829 			goto not_found;
1830 	}
1831 
1832 	/* Logical channel */
1833 	if (is_src) {
1834 		if (phy->allocated_src == D40_ALLOC_PHY)
1835 			goto not_found;
1836 
1837 		if (phy->allocated_src == D40_ALLOC_FREE)
1838 			phy->allocated_src = D40_ALLOC_LOG_FREE;
1839 
1840 		if (!(phy->allocated_src & BIT(log_event_line))) {
1841 			phy->allocated_src |= BIT(log_event_line);
1842 			goto found;
1843 		} else
1844 			goto not_found;
1845 	} else {
1846 		if (phy->allocated_dst == D40_ALLOC_PHY)
1847 			goto not_found;
1848 
1849 		if (phy->allocated_dst == D40_ALLOC_FREE)
1850 			phy->allocated_dst = D40_ALLOC_LOG_FREE;
1851 
1852 		if (!(phy->allocated_dst & BIT(log_event_line))) {
1853 			phy->allocated_dst |= BIT(log_event_line);
1854 			goto found;
1855 		} else
1856 			goto not_found;
1857 	}
1858 
1859 not_found:
1860 	spin_unlock_irqrestore(&phy->lock, flags);
1861 	return false;
1862 found:
1863 	spin_unlock_irqrestore(&phy->lock, flags);
1864 	return true;
1865 }
1866 
1867 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1868 			       int log_event_line)
1869 {
1870 	unsigned long flags;
1871 	bool is_free = false;
1872 
1873 	spin_lock_irqsave(&phy->lock, flags);
1874 	if (!log_event_line) {
1875 		phy->allocated_dst = D40_ALLOC_FREE;
1876 		phy->allocated_src = D40_ALLOC_FREE;
1877 		is_free = true;
1878 		goto out;
1879 	}
1880 
1881 	/* Logical channel */
1882 	if (is_src) {
1883 		phy->allocated_src &= ~BIT(log_event_line);
1884 		if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1885 			phy->allocated_src = D40_ALLOC_FREE;
1886 	} else {
1887 		phy->allocated_dst &= ~BIT(log_event_line);
1888 		if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1889 			phy->allocated_dst = D40_ALLOC_FREE;
1890 	}
1891 
1892 	is_free = ((phy->allocated_src | phy->allocated_dst) ==
1893 		   D40_ALLOC_FREE);
1894 
1895 out:
1896 	spin_unlock_irqrestore(&phy->lock, flags);
1897 
1898 	return is_free;
1899 }
1900 
1901 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1902 {
1903 	int dev_type = d40c->dma_cfg.dev_type;
1904 	int event_group;
1905 	int event_line;
1906 	struct d40_phy_res *phys;
1907 	int i;
1908 	int j;
1909 	int log_num;
1910 	int num_phy_chans;
1911 	bool is_src;
1912 	bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1913 
1914 	phys = d40c->base->phy_res;
1915 	num_phy_chans = d40c->base->num_phy_chans;
1916 
1917 	if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
1918 		log_num = 2 * dev_type;
1919 		is_src = true;
1920 	} else if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
1921 		   d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1922 		/* dst event lines are used for logical memcpy */
1923 		log_num = 2 * dev_type + 1;
1924 		is_src = false;
1925 	} else
1926 		return -EINVAL;
1927 
1928 	event_group = D40_TYPE_TO_GROUP(dev_type);
1929 	event_line = D40_TYPE_TO_EVENT(dev_type);
1930 
1931 	if (!is_log) {
1932 		if (d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1933 			/* Find physical half channel */
1934 			if (d40c->dma_cfg.use_fixed_channel) {
1935 				i = d40c->dma_cfg.phy_channel;
1936 				if (d40_alloc_mask_set(&phys[i], is_src,
1937 						       0, is_log,
1938 						       first_phy_user))
1939 					goto found_phy;
1940 			} else {
1941 				for (i = 0; i < num_phy_chans; i++) {
1942 					if (d40_alloc_mask_set(&phys[i], is_src,
1943 						       0, is_log,
1944 						       first_phy_user))
1945 						goto found_phy;
1946 				}
1947 			}
1948 		} else
1949 			for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1950 				int phy_num = j  + event_group * 2;
1951 				for (i = phy_num; i < phy_num + 2; i++) {
1952 					if (d40_alloc_mask_set(&phys[i],
1953 							       is_src,
1954 							       0,
1955 							       is_log,
1956 							       first_phy_user))
1957 						goto found_phy;
1958 				}
1959 			}
1960 		return -EINVAL;
1961 found_phy:
1962 		d40c->phy_chan = &phys[i];
1963 		d40c->log_num = D40_PHY_CHAN;
1964 		goto out;
1965 	}
1966 	if (dev_type == -1)
1967 		return -EINVAL;
1968 
1969 	/* Find logical channel */
1970 	for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1971 		int phy_num = j + event_group * 2;
1972 
1973 		if (d40c->dma_cfg.use_fixed_channel) {
1974 			i = d40c->dma_cfg.phy_channel;
1975 
1976 			if ((i != phy_num) && (i != phy_num + 1)) {
1977 				dev_err(chan2dev(d40c),
1978 					"invalid fixed phy channel %d\n", i);
1979 				return -EINVAL;
1980 			}
1981 
1982 			if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1983 					       is_log, first_phy_user))
1984 				goto found_log;
1985 
1986 			dev_err(chan2dev(d40c),
1987 				"could not allocate fixed phy channel %d\n", i);
1988 			return -EINVAL;
1989 		}
1990 
1991 		/*
1992 		 * Spread logical channels across all available physical rather
1993 		 * than pack every logical channel at the first available phy
1994 		 * channels.
1995 		 */
1996 		if (is_src) {
1997 			for (i = phy_num; i < phy_num + 2; i++) {
1998 				if (d40_alloc_mask_set(&phys[i], is_src,
1999 						       event_line, is_log,
2000 						       first_phy_user))
2001 					goto found_log;
2002 			}
2003 		} else {
2004 			for (i = phy_num + 1; i >= phy_num; i--) {
2005 				if (d40_alloc_mask_set(&phys[i], is_src,
2006 						       event_line, is_log,
2007 						       first_phy_user))
2008 					goto found_log;
2009 			}
2010 		}
2011 	}
2012 	return -EINVAL;
2013 
2014 found_log:
2015 	d40c->phy_chan = &phys[i];
2016 	d40c->log_num = log_num;
2017 out:
2018 
2019 	if (is_log)
2020 		d40c->base->lookup_log_chans[d40c->log_num] = d40c;
2021 	else
2022 		d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
2023 
2024 	return 0;
2025 
2026 }
2027 
2028 static int d40_config_memcpy(struct d40_chan *d40c)
2029 {
2030 	dma_cap_mask_t cap = d40c->chan.device->cap_mask;
2031 
2032 	if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
2033 		d40c->dma_cfg = dma40_memcpy_conf_log;
2034 		d40c->dma_cfg.dev_type = dma40_memcpy_channels[d40c->chan.chan_id];
2035 
2036 		d40_log_cfg(&d40c->dma_cfg,
2037 			    &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2038 
2039 	} else if (dma_has_cap(DMA_MEMCPY, cap) &&
2040 		   dma_has_cap(DMA_SLAVE, cap)) {
2041 		d40c->dma_cfg = dma40_memcpy_conf_phy;
2042 
2043 		/* Generate interrrupt at end of transfer or relink. */
2044 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_TIM_POS);
2045 
2046 		/* Generate interrupt on error. */
2047 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
2048 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
2049 
2050 	} else {
2051 		chan_err(d40c, "No memcpy\n");
2052 		return -EINVAL;
2053 	}
2054 
2055 	return 0;
2056 }
2057 
2058 static int d40_free_dma(struct d40_chan *d40c)
2059 {
2060 
2061 	int res = 0;
2062 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2063 	struct d40_phy_res *phy = d40c->phy_chan;
2064 	bool is_src;
2065 
2066 	/* Terminate all queued and active transfers */
2067 	d40_term_all(d40c);
2068 
2069 	if (phy == NULL) {
2070 		chan_err(d40c, "phy == null\n");
2071 		return -EINVAL;
2072 	}
2073 
2074 	if (phy->allocated_src == D40_ALLOC_FREE &&
2075 	    phy->allocated_dst == D40_ALLOC_FREE) {
2076 		chan_err(d40c, "channel already free\n");
2077 		return -EINVAL;
2078 	}
2079 
2080 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2081 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM)
2082 		is_src = false;
2083 	else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2084 		is_src = true;
2085 	else {
2086 		chan_err(d40c, "Unknown direction\n");
2087 		return -EINVAL;
2088 	}
2089 
2090 	pm_runtime_get_sync(d40c->base->dev);
2091 	res = d40_channel_execute_command(d40c, D40_DMA_STOP);
2092 	if (res) {
2093 		chan_err(d40c, "stop failed\n");
2094 		goto out;
2095 	}
2096 
2097 	d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
2098 
2099 	if (chan_is_logical(d40c))
2100 		d40c->base->lookup_log_chans[d40c->log_num] = NULL;
2101 	else
2102 		d40c->base->lookup_phy_chans[phy->num] = NULL;
2103 
2104 	if (d40c->busy) {
2105 		pm_runtime_mark_last_busy(d40c->base->dev);
2106 		pm_runtime_put_autosuspend(d40c->base->dev);
2107 	}
2108 
2109 	d40c->busy = false;
2110 	d40c->phy_chan = NULL;
2111 	d40c->configured = false;
2112 out:
2113 
2114 	pm_runtime_mark_last_busy(d40c->base->dev);
2115 	pm_runtime_put_autosuspend(d40c->base->dev);
2116 	return res;
2117 }
2118 
2119 static bool d40_is_paused(struct d40_chan *d40c)
2120 {
2121 	void __iomem *chanbase = chan_base(d40c);
2122 	bool is_paused = false;
2123 	unsigned long flags;
2124 	void __iomem *active_reg;
2125 	u32 status;
2126 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2127 
2128 	spin_lock_irqsave(&d40c->lock, flags);
2129 
2130 	if (chan_is_physical(d40c)) {
2131 		if (d40c->phy_chan->num % 2 == 0)
2132 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
2133 		else
2134 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
2135 
2136 		status = (readl(active_reg) &
2137 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
2138 			D40_CHAN_POS(d40c->phy_chan->num);
2139 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
2140 			is_paused = true;
2141 
2142 		goto _exit;
2143 	}
2144 
2145 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2146 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
2147 		status = readl(chanbase + D40_CHAN_REG_SDLNK);
2148 	} else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
2149 		status = readl(chanbase + D40_CHAN_REG_SSLNK);
2150 	} else {
2151 		chan_err(d40c, "Unknown direction\n");
2152 		goto _exit;
2153 	}
2154 
2155 	status = (status & D40_EVENTLINE_MASK(event)) >>
2156 		D40_EVENTLINE_POS(event);
2157 
2158 	if (status != D40_DMA_RUN)
2159 		is_paused = true;
2160 _exit:
2161 	spin_unlock_irqrestore(&d40c->lock, flags);
2162 	return is_paused;
2163 
2164 }
2165 
2166 static u32 stedma40_residue(struct dma_chan *chan)
2167 {
2168 	struct d40_chan *d40c =
2169 		container_of(chan, struct d40_chan, chan);
2170 	u32 bytes_left;
2171 	unsigned long flags;
2172 
2173 	spin_lock_irqsave(&d40c->lock, flags);
2174 	bytes_left = d40_residue(d40c);
2175 	spin_unlock_irqrestore(&d40c->lock, flags);
2176 
2177 	return bytes_left;
2178 }
2179 
2180 static int
2181 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
2182 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2183 		unsigned int sg_len, dma_addr_t src_dev_addr,
2184 		dma_addr_t dst_dev_addr)
2185 {
2186 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2187 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2188 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2189 	int ret;
2190 
2191 	ret = d40_log_sg_to_lli(sg_src, sg_len,
2192 				src_dev_addr,
2193 				desc->lli_log.src,
2194 				chan->log_def.lcsp1,
2195 				src_info->data_width,
2196 				dst_info->data_width);
2197 
2198 	ret = d40_log_sg_to_lli(sg_dst, sg_len,
2199 				dst_dev_addr,
2200 				desc->lli_log.dst,
2201 				chan->log_def.lcsp3,
2202 				dst_info->data_width,
2203 				src_info->data_width);
2204 
2205 	return ret < 0 ? ret : 0;
2206 }
2207 
2208 static int
2209 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2210 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2211 		unsigned int sg_len, dma_addr_t src_dev_addr,
2212 		dma_addr_t dst_dev_addr)
2213 {
2214 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2215 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2216 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2217 	unsigned long flags = 0;
2218 	int ret;
2219 
2220 	if (desc->cyclic)
2221 		flags |= LLI_CYCLIC | LLI_TERM_INT;
2222 
2223 	ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2224 				desc->lli_phy.src,
2225 				virt_to_phys(desc->lli_phy.src),
2226 				chan->src_def_cfg,
2227 				src_info, dst_info, flags);
2228 
2229 	ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2230 				desc->lli_phy.dst,
2231 				virt_to_phys(desc->lli_phy.dst),
2232 				chan->dst_def_cfg,
2233 				dst_info, src_info, flags);
2234 
2235 	dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2236 				   desc->lli_pool.size, DMA_TO_DEVICE);
2237 
2238 	return ret < 0 ? ret : 0;
2239 }
2240 
2241 static struct d40_desc *
2242 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2243 	      unsigned int sg_len, unsigned long dma_flags)
2244 {
2245 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2246 	struct d40_desc *desc;
2247 	int ret;
2248 
2249 	desc = d40_desc_get(chan);
2250 	if (!desc)
2251 		return NULL;
2252 
2253 	desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2254 					cfg->dst_info.data_width);
2255 	if (desc->lli_len < 0) {
2256 		chan_err(chan, "Unaligned size\n");
2257 		goto err;
2258 	}
2259 
2260 	ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2261 	if (ret < 0) {
2262 		chan_err(chan, "Could not allocate lli\n");
2263 		goto err;
2264 	}
2265 
2266 	desc->lli_current = 0;
2267 	desc->txd.flags = dma_flags;
2268 	desc->txd.tx_submit = d40_tx_submit;
2269 
2270 	dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2271 
2272 	return desc;
2273 
2274 err:
2275 	d40_desc_free(chan, desc);
2276 	return NULL;
2277 }
2278 
2279 static struct dma_async_tx_descriptor *
2280 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2281 	    struct scatterlist *sg_dst, unsigned int sg_len,
2282 	    enum dma_transfer_direction direction, unsigned long dma_flags)
2283 {
2284 	struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2285 	dma_addr_t src_dev_addr = 0;
2286 	dma_addr_t dst_dev_addr = 0;
2287 	struct d40_desc *desc;
2288 	unsigned long flags;
2289 	int ret;
2290 
2291 	if (!chan->phy_chan) {
2292 		chan_err(chan, "Cannot prepare unallocated channel\n");
2293 		return NULL;
2294 	}
2295 
2296 	spin_lock_irqsave(&chan->lock, flags);
2297 
2298 	desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2299 	if (desc == NULL)
2300 		goto err;
2301 
2302 	if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2303 		desc->cyclic = true;
2304 
2305 	if (direction == DMA_DEV_TO_MEM)
2306 		src_dev_addr = chan->runtime_addr;
2307 	else if (direction == DMA_MEM_TO_DEV)
2308 		dst_dev_addr = chan->runtime_addr;
2309 
2310 	if (chan_is_logical(chan))
2311 		ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2312 				      sg_len, src_dev_addr, dst_dev_addr);
2313 	else
2314 		ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2315 				      sg_len, src_dev_addr, dst_dev_addr);
2316 
2317 	if (ret) {
2318 		chan_err(chan, "Failed to prepare %s sg job: %d\n",
2319 			 chan_is_logical(chan) ? "log" : "phy", ret);
2320 		goto err;
2321 	}
2322 
2323 	/*
2324 	 * add descriptor to the prepare queue in order to be able
2325 	 * to free them later in terminate_all
2326 	 */
2327 	list_add_tail(&desc->node, &chan->prepare_queue);
2328 
2329 	spin_unlock_irqrestore(&chan->lock, flags);
2330 
2331 	return &desc->txd;
2332 
2333 err:
2334 	if (desc)
2335 		d40_desc_free(chan, desc);
2336 	spin_unlock_irqrestore(&chan->lock, flags);
2337 	return NULL;
2338 }
2339 
2340 bool stedma40_filter(struct dma_chan *chan, void *data)
2341 {
2342 	struct stedma40_chan_cfg *info = data;
2343 	struct d40_chan *d40c =
2344 		container_of(chan, struct d40_chan, chan);
2345 	int err;
2346 
2347 	if (data) {
2348 		err = d40_validate_conf(d40c, info);
2349 		if (!err)
2350 			d40c->dma_cfg = *info;
2351 	} else
2352 		err = d40_config_memcpy(d40c);
2353 
2354 	if (!err)
2355 		d40c->configured = true;
2356 
2357 	return err == 0;
2358 }
2359 EXPORT_SYMBOL(stedma40_filter);
2360 
2361 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2362 {
2363 	bool realtime = d40c->dma_cfg.realtime;
2364 	bool highprio = d40c->dma_cfg.high_priority;
2365 	u32 rtreg;
2366 	u32 event = D40_TYPE_TO_EVENT(dev_type);
2367 	u32 group = D40_TYPE_TO_GROUP(dev_type);
2368 	u32 bit = BIT(event);
2369 	u32 prioreg;
2370 	struct d40_gen_dmac *dmac = &d40c->base->gen_dmac;
2371 
2372 	rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear;
2373 	/*
2374 	 * Due to a hardware bug, in some cases a logical channel triggered by
2375 	 * a high priority destination event line can generate extra packet
2376 	 * transactions.
2377 	 *
2378 	 * The workaround is to not set the high priority level for the
2379 	 * destination event lines that trigger logical channels.
2380 	 */
2381 	if (!src && chan_is_logical(d40c))
2382 		highprio = false;
2383 
2384 	prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear;
2385 
2386 	/* Destination event lines are stored in the upper halfword */
2387 	if (!src)
2388 		bit <<= 16;
2389 
2390 	writel(bit, d40c->base->virtbase + prioreg + group * 4);
2391 	writel(bit, d40c->base->virtbase + rtreg + group * 4);
2392 }
2393 
2394 static void d40_set_prio_realtime(struct d40_chan *d40c)
2395 {
2396 	if (d40c->base->rev < 3)
2397 		return;
2398 
2399 	if ((d40c->dma_cfg.dir ==  DMA_DEV_TO_MEM) ||
2400 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2401 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, true);
2402 
2403 	if ((d40c->dma_cfg.dir ==  DMA_MEM_TO_DEV) ||
2404 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2405 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, false);
2406 }
2407 
2408 #define D40_DT_FLAGS_MODE(flags)       ((flags >> 0) & 0x1)
2409 #define D40_DT_FLAGS_DIR(flags)        ((flags >> 1) & 0x1)
2410 #define D40_DT_FLAGS_BIG_ENDIAN(flags) ((flags >> 2) & 0x1)
2411 #define D40_DT_FLAGS_FIXED_CHAN(flags) ((flags >> 3) & 0x1)
2412 
2413 static struct dma_chan *d40_xlate(struct of_phandle_args *dma_spec,
2414 				  struct of_dma *ofdma)
2415 {
2416 	struct stedma40_chan_cfg cfg;
2417 	dma_cap_mask_t cap;
2418 	u32 flags;
2419 
2420 	memset(&cfg, 0, sizeof(struct stedma40_chan_cfg));
2421 
2422 	dma_cap_zero(cap);
2423 	dma_cap_set(DMA_SLAVE, cap);
2424 
2425 	cfg.dev_type = dma_spec->args[0];
2426 	flags = dma_spec->args[2];
2427 
2428 	switch (D40_DT_FLAGS_MODE(flags)) {
2429 	case 0: cfg.mode = STEDMA40_MODE_LOGICAL; break;
2430 	case 1: cfg.mode = STEDMA40_MODE_PHYSICAL; break;
2431 	}
2432 
2433 	switch (D40_DT_FLAGS_DIR(flags)) {
2434 	case 0:
2435 		cfg.dir = DMA_MEM_TO_DEV;
2436 		cfg.dst_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2437 		break;
2438 	case 1:
2439 		cfg.dir = DMA_DEV_TO_MEM;
2440 		cfg.src_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2441 		break;
2442 	}
2443 
2444 	if (D40_DT_FLAGS_FIXED_CHAN(flags)) {
2445 		cfg.phy_channel = dma_spec->args[1];
2446 		cfg.use_fixed_channel = true;
2447 	}
2448 
2449 	return dma_request_channel(cap, stedma40_filter, &cfg);
2450 }
2451 
2452 /* DMA ENGINE functions */
2453 static int d40_alloc_chan_resources(struct dma_chan *chan)
2454 {
2455 	int err;
2456 	unsigned long flags;
2457 	struct d40_chan *d40c =
2458 		container_of(chan, struct d40_chan, chan);
2459 	bool is_free_phy;
2460 	spin_lock_irqsave(&d40c->lock, flags);
2461 
2462 	dma_cookie_init(chan);
2463 
2464 	/* If no dma configuration is set use default configuration (memcpy) */
2465 	if (!d40c->configured) {
2466 		err = d40_config_memcpy(d40c);
2467 		if (err) {
2468 			chan_err(d40c, "Failed to configure memcpy channel\n");
2469 			goto fail;
2470 		}
2471 	}
2472 
2473 	err = d40_allocate_channel(d40c, &is_free_phy);
2474 	if (err) {
2475 		chan_err(d40c, "Failed to allocate channel\n");
2476 		d40c->configured = false;
2477 		goto fail;
2478 	}
2479 
2480 	pm_runtime_get_sync(d40c->base->dev);
2481 
2482 	d40_set_prio_realtime(d40c);
2483 
2484 	if (chan_is_logical(d40c)) {
2485 		if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2486 			d40c->lcpa = d40c->base->lcpa_base +
2487 				d40c->dma_cfg.dev_type * D40_LCPA_CHAN_SIZE;
2488 		else
2489 			d40c->lcpa = d40c->base->lcpa_base +
2490 				d40c->dma_cfg.dev_type *
2491 				D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2492 
2493 		/* Unmask the Global Interrupt Mask. */
2494 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2495 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2496 	}
2497 
2498 	dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2499 		 chan_is_logical(d40c) ? "logical" : "physical",
2500 		 d40c->phy_chan->num,
2501 		 d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2502 
2503 
2504 	/*
2505 	 * Only write channel configuration to the DMA if the physical
2506 	 * resource is free. In case of multiple logical channels
2507 	 * on the same physical resource, only the first write is necessary.
2508 	 */
2509 	if (is_free_phy)
2510 		d40_config_write(d40c);
2511 fail:
2512 	pm_runtime_mark_last_busy(d40c->base->dev);
2513 	pm_runtime_put_autosuspend(d40c->base->dev);
2514 	spin_unlock_irqrestore(&d40c->lock, flags);
2515 	return err;
2516 }
2517 
2518 static void d40_free_chan_resources(struct dma_chan *chan)
2519 {
2520 	struct d40_chan *d40c =
2521 		container_of(chan, struct d40_chan, chan);
2522 	int err;
2523 	unsigned long flags;
2524 
2525 	if (d40c->phy_chan == NULL) {
2526 		chan_err(d40c, "Cannot free unallocated channel\n");
2527 		return;
2528 	}
2529 
2530 	spin_lock_irqsave(&d40c->lock, flags);
2531 
2532 	err = d40_free_dma(d40c);
2533 
2534 	if (err)
2535 		chan_err(d40c, "Failed to free channel\n");
2536 	spin_unlock_irqrestore(&d40c->lock, flags);
2537 }
2538 
2539 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2540 						       dma_addr_t dst,
2541 						       dma_addr_t src,
2542 						       size_t size,
2543 						       unsigned long dma_flags)
2544 {
2545 	struct scatterlist dst_sg;
2546 	struct scatterlist src_sg;
2547 
2548 	sg_init_table(&dst_sg, 1);
2549 	sg_init_table(&src_sg, 1);
2550 
2551 	sg_dma_address(&dst_sg) = dst;
2552 	sg_dma_address(&src_sg) = src;
2553 
2554 	sg_dma_len(&dst_sg) = size;
2555 	sg_dma_len(&src_sg) = size;
2556 
2557 	return d40_prep_sg(chan, &src_sg, &dst_sg, 1, DMA_NONE, dma_flags);
2558 }
2559 
2560 static struct dma_async_tx_descriptor *
2561 d40_prep_memcpy_sg(struct dma_chan *chan,
2562 		   struct scatterlist *dst_sg, unsigned int dst_nents,
2563 		   struct scatterlist *src_sg, unsigned int src_nents,
2564 		   unsigned long dma_flags)
2565 {
2566 	if (dst_nents != src_nents)
2567 		return NULL;
2568 
2569 	return d40_prep_sg(chan, src_sg, dst_sg, src_nents, DMA_NONE, dma_flags);
2570 }
2571 
2572 static struct dma_async_tx_descriptor *
2573 d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2574 		  unsigned int sg_len, enum dma_transfer_direction direction,
2575 		  unsigned long dma_flags, void *context)
2576 {
2577 	if (!is_slave_direction(direction))
2578 		return NULL;
2579 
2580 	return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2581 }
2582 
2583 static struct dma_async_tx_descriptor *
2584 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2585 		     size_t buf_len, size_t period_len,
2586 		     enum dma_transfer_direction direction, unsigned long flags,
2587 		     void *context)
2588 {
2589 	unsigned int periods = buf_len / period_len;
2590 	struct dma_async_tx_descriptor *txd;
2591 	struct scatterlist *sg;
2592 	int i;
2593 
2594 	sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2595 	if (!sg)
2596 		return NULL;
2597 
2598 	for (i = 0; i < periods; i++) {
2599 		sg_dma_address(&sg[i]) = dma_addr;
2600 		sg_dma_len(&sg[i]) = period_len;
2601 		dma_addr += period_len;
2602 	}
2603 
2604 	sg[periods].offset = 0;
2605 	sg_dma_len(&sg[periods]) = 0;
2606 	sg[periods].page_link =
2607 		((unsigned long)sg | 0x01) & ~0x02;
2608 
2609 	txd = d40_prep_sg(chan, sg, sg, periods, direction,
2610 			  DMA_PREP_INTERRUPT);
2611 
2612 	kfree(sg);
2613 
2614 	return txd;
2615 }
2616 
2617 static enum dma_status d40_tx_status(struct dma_chan *chan,
2618 				     dma_cookie_t cookie,
2619 				     struct dma_tx_state *txstate)
2620 {
2621 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2622 	enum dma_status ret;
2623 
2624 	if (d40c->phy_chan == NULL) {
2625 		chan_err(d40c, "Cannot read status of unallocated channel\n");
2626 		return -EINVAL;
2627 	}
2628 
2629 	ret = dma_cookie_status(chan, cookie, txstate);
2630 	if (ret != DMA_COMPLETE)
2631 		dma_set_residue(txstate, stedma40_residue(chan));
2632 
2633 	if (d40_is_paused(d40c))
2634 		ret = DMA_PAUSED;
2635 
2636 	return ret;
2637 }
2638 
2639 static void d40_issue_pending(struct dma_chan *chan)
2640 {
2641 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2642 	unsigned long flags;
2643 
2644 	if (d40c->phy_chan == NULL) {
2645 		chan_err(d40c, "Channel is not allocated!\n");
2646 		return;
2647 	}
2648 
2649 	spin_lock_irqsave(&d40c->lock, flags);
2650 
2651 	list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2652 
2653 	/* Busy means that queued jobs are already being processed */
2654 	if (!d40c->busy)
2655 		(void) d40_queue_start(d40c);
2656 
2657 	spin_unlock_irqrestore(&d40c->lock, flags);
2658 }
2659 
2660 static void d40_terminate_all(struct dma_chan *chan)
2661 {
2662 	unsigned long flags;
2663 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2664 	int ret;
2665 
2666 	spin_lock_irqsave(&d40c->lock, flags);
2667 
2668 	pm_runtime_get_sync(d40c->base->dev);
2669 	ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2670 	if (ret)
2671 		chan_err(d40c, "Failed to stop channel\n");
2672 
2673 	d40_term_all(d40c);
2674 	pm_runtime_mark_last_busy(d40c->base->dev);
2675 	pm_runtime_put_autosuspend(d40c->base->dev);
2676 	if (d40c->busy) {
2677 		pm_runtime_mark_last_busy(d40c->base->dev);
2678 		pm_runtime_put_autosuspend(d40c->base->dev);
2679 	}
2680 	d40c->busy = false;
2681 
2682 	spin_unlock_irqrestore(&d40c->lock, flags);
2683 }
2684 
2685 static int
2686 dma40_config_to_halfchannel(struct d40_chan *d40c,
2687 			    struct stedma40_half_channel_info *info,
2688 			    u32 maxburst)
2689 {
2690 	int psize;
2691 
2692 	if (chan_is_logical(d40c)) {
2693 		if (maxburst >= 16)
2694 			psize = STEDMA40_PSIZE_LOG_16;
2695 		else if (maxburst >= 8)
2696 			psize = STEDMA40_PSIZE_LOG_8;
2697 		else if (maxburst >= 4)
2698 			psize = STEDMA40_PSIZE_LOG_4;
2699 		else
2700 			psize = STEDMA40_PSIZE_LOG_1;
2701 	} else {
2702 		if (maxburst >= 16)
2703 			psize = STEDMA40_PSIZE_PHY_16;
2704 		else if (maxburst >= 8)
2705 			psize = STEDMA40_PSIZE_PHY_8;
2706 		else if (maxburst >= 4)
2707 			psize = STEDMA40_PSIZE_PHY_4;
2708 		else
2709 			psize = STEDMA40_PSIZE_PHY_1;
2710 	}
2711 
2712 	info->psize = psize;
2713 	info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2714 
2715 	return 0;
2716 }
2717 
2718 /* Runtime reconfiguration extension */
2719 static int d40_set_runtime_config(struct dma_chan *chan,
2720 				  struct dma_slave_config *config)
2721 {
2722 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2723 	struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2724 	enum dma_slave_buswidth src_addr_width, dst_addr_width;
2725 	dma_addr_t config_addr;
2726 	u32 src_maxburst, dst_maxburst;
2727 	int ret;
2728 
2729 	src_addr_width = config->src_addr_width;
2730 	src_maxburst = config->src_maxburst;
2731 	dst_addr_width = config->dst_addr_width;
2732 	dst_maxburst = config->dst_maxburst;
2733 
2734 	if (config->direction == DMA_DEV_TO_MEM) {
2735 		config_addr = config->src_addr;
2736 
2737 		if (cfg->dir != DMA_DEV_TO_MEM)
2738 			dev_dbg(d40c->base->dev,
2739 				"channel was not configured for peripheral "
2740 				"to memory transfer (%d) overriding\n",
2741 				cfg->dir);
2742 		cfg->dir = DMA_DEV_TO_MEM;
2743 
2744 		/* Configure the memory side */
2745 		if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2746 			dst_addr_width = src_addr_width;
2747 		if (dst_maxburst == 0)
2748 			dst_maxburst = src_maxburst;
2749 
2750 	} else if (config->direction == DMA_MEM_TO_DEV) {
2751 		config_addr = config->dst_addr;
2752 
2753 		if (cfg->dir != DMA_MEM_TO_DEV)
2754 			dev_dbg(d40c->base->dev,
2755 				"channel was not configured for memory "
2756 				"to peripheral transfer (%d) overriding\n",
2757 				cfg->dir);
2758 		cfg->dir = DMA_MEM_TO_DEV;
2759 
2760 		/* Configure the memory side */
2761 		if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2762 			src_addr_width = dst_addr_width;
2763 		if (src_maxburst == 0)
2764 			src_maxburst = dst_maxburst;
2765 	} else {
2766 		dev_err(d40c->base->dev,
2767 			"unrecognized channel direction %d\n",
2768 			config->direction);
2769 		return -EINVAL;
2770 	}
2771 
2772 	if (config_addr <= 0) {
2773 		dev_err(d40c->base->dev, "no address supplied\n");
2774 		return -EINVAL;
2775 	}
2776 
2777 	if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2778 		dev_err(d40c->base->dev,
2779 			"src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2780 			src_maxburst,
2781 			src_addr_width,
2782 			dst_maxburst,
2783 			dst_addr_width);
2784 		return -EINVAL;
2785 	}
2786 
2787 	if (src_maxburst > 16) {
2788 		src_maxburst = 16;
2789 		dst_maxburst = src_maxburst * src_addr_width / dst_addr_width;
2790 	} else if (dst_maxburst > 16) {
2791 		dst_maxburst = 16;
2792 		src_maxburst = dst_maxburst * dst_addr_width / src_addr_width;
2793 	}
2794 
2795 	/* Only valid widths are; 1, 2, 4 and 8. */
2796 	if (src_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2797 	    src_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2798 	    dst_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2799 	    dst_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2800 	    !is_power_of_2(src_addr_width) ||
2801 	    !is_power_of_2(dst_addr_width))
2802 		return -EINVAL;
2803 
2804 	cfg->src_info.data_width = src_addr_width;
2805 	cfg->dst_info.data_width = dst_addr_width;
2806 
2807 	ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2808 					  src_maxburst);
2809 	if (ret)
2810 		return ret;
2811 
2812 	ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2813 					  dst_maxburst);
2814 	if (ret)
2815 		return ret;
2816 
2817 	/* Fill in register values */
2818 	if (chan_is_logical(d40c))
2819 		d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2820 	else
2821 		d40_phy_cfg(cfg, &d40c->src_def_cfg, &d40c->dst_def_cfg);
2822 
2823 	/* These settings will take precedence later */
2824 	d40c->runtime_addr = config_addr;
2825 	d40c->runtime_direction = config->direction;
2826 	dev_dbg(d40c->base->dev,
2827 		"configured channel %s for %s, data width %d/%d, "
2828 		"maxburst %d/%d elements, LE, no flow control\n",
2829 		dma_chan_name(chan),
2830 		(config->direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2831 		src_addr_width, dst_addr_width,
2832 		src_maxburst, dst_maxburst);
2833 
2834 	return 0;
2835 }
2836 
2837 static int d40_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
2838 		       unsigned long arg)
2839 {
2840 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2841 
2842 	if (d40c->phy_chan == NULL) {
2843 		chan_err(d40c, "Channel is not allocated!\n");
2844 		return -EINVAL;
2845 	}
2846 
2847 	switch (cmd) {
2848 	case DMA_TERMINATE_ALL:
2849 		d40_terminate_all(chan);
2850 		return 0;
2851 	case DMA_PAUSE:
2852 		return d40_pause(d40c);
2853 	case DMA_RESUME:
2854 		return d40_resume(d40c);
2855 	case DMA_SLAVE_CONFIG:
2856 		return d40_set_runtime_config(chan,
2857 			(struct dma_slave_config *) arg);
2858 	default:
2859 		break;
2860 	}
2861 
2862 	/* Other commands are unimplemented */
2863 	return -ENXIO;
2864 }
2865 
2866 /* Initialization functions */
2867 
2868 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2869 				 struct d40_chan *chans, int offset,
2870 				 int num_chans)
2871 {
2872 	int i = 0;
2873 	struct d40_chan *d40c;
2874 
2875 	INIT_LIST_HEAD(&dma->channels);
2876 
2877 	for (i = offset; i < offset + num_chans; i++) {
2878 		d40c = &chans[i];
2879 		d40c->base = base;
2880 		d40c->chan.device = dma;
2881 
2882 		spin_lock_init(&d40c->lock);
2883 
2884 		d40c->log_num = D40_PHY_CHAN;
2885 
2886 		INIT_LIST_HEAD(&d40c->done);
2887 		INIT_LIST_HEAD(&d40c->active);
2888 		INIT_LIST_HEAD(&d40c->queue);
2889 		INIT_LIST_HEAD(&d40c->pending_queue);
2890 		INIT_LIST_HEAD(&d40c->client);
2891 		INIT_LIST_HEAD(&d40c->prepare_queue);
2892 
2893 		tasklet_init(&d40c->tasklet, dma_tasklet,
2894 			     (unsigned long) d40c);
2895 
2896 		list_add_tail(&d40c->chan.device_node,
2897 			      &dma->channels);
2898 	}
2899 }
2900 
2901 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2902 {
2903 	if (dma_has_cap(DMA_SLAVE, dev->cap_mask))
2904 		dev->device_prep_slave_sg = d40_prep_slave_sg;
2905 
2906 	if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2907 		dev->device_prep_dma_memcpy = d40_prep_memcpy;
2908 
2909 		/*
2910 		 * This controller can only access address at even
2911 		 * 32bit boundaries, i.e. 2^2
2912 		 */
2913 		dev->copy_align = 2;
2914 	}
2915 
2916 	if (dma_has_cap(DMA_SG, dev->cap_mask))
2917 		dev->device_prep_dma_sg = d40_prep_memcpy_sg;
2918 
2919 	if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2920 		dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2921 
2922 	dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2923 	dev->device_free_chan_resources = d40_free_chan_resources;
2924 	dev->device_issue_pending = d40_issue_pending;
2925 	dev->device_tx_status = d40_tx_status;
2926 	dev->device_control = d40_control;
2927 	dev->dev = base->dev;
2928 }
2929 
2930 static int __init d40_dmaengine_init(struct d40_base *base,
2931 				     int num_reserved_chans)
2932 {
2933 	int err ;
2934 
2935 	d40_chan_init(base, &base->dma_slave, base->log_chans,
2936 		      0, base->num_log_chans);
2937 
2938 	dma_cap_zero(base->dma_slave.cap_mask);
2939 	dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2940 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2941 
2942 	d40_ops_init(base, &base->dma_slave);
2943 
2944 	err = dma_async_device_register(&base->dma_slave);
2945 
2946 	if (err) {
2947 		d40_err(base->dev, "Failed to register slave channels\n");
2948 		goto failure1;
2949 	}
2950 
2951 	d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2952 		      base->num_log_chans, base->num_memcpy_chans);
2953 
2954 	dma_cap_zero(base->dma_memcpy.cap_mask);
2955 	dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2956 	dma_cap_set(DMA_SG, base->dma_memcpy.cap_mask);
2957 
2958 	d40_ops_init(base, &base->dma_memcpy);
2959 
2960 	err = dma_async_device_register(&base->dma_memcpy);
2961 
2962 	if (err) {
2963 		d40_err(base->dev,
2964 			"Failed to regsiter memcpy only channels\n");
2965 		goto failure2;
2966 	}
2967 
2968 	d40_chan_init(base, &base->dma_both, base->phy_chans,
2969 		      0, num_reserved_chans);
2970 
2971 	dma_cap_zero(base->dma_both.cap_mask);
2972 	dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2973 	dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2974 	dma_cap_set(DMA_SG, base->dma_both.cap_mask);
2975 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2976 
2977 	d40_ops_init(base, &base->dma_both);
2978 	err = dma_async_device_register(&base->dma_both);
2979 
2980 	if (err) {
2981 		d40_err(base->dev,
2982 			"Failed to register logical and physical capable channels\n");
2983 		goto failure3;
2984 	}
2985 	return 0;
2986 failure3:
2987 	dma_async_device_unregister(&base->dma_memcpy);
2988 failure2:
2989 	dma_async_device_unregister(&base->dma_slave);
2990 failure1:
2991 	return err;
2992 }
2993 
2994 /* Suspend resume functionality */
2995 #ifdef CONFIG_PM
2996 static int dma40_pm_suspend(struct device *dev)
2997 {
2998 	struct platform_device *pdev = to_platform_device(dev);
2999 	struct d40_base *base = platform_get_drvdata(pdev);
3000 	int ret = 0;
3001 
3002 	if (base->lcpa_regulator)
3003 		ret = regulator_disable(base->lcpa_regulator);
3004 	return ret;
3005 }
3006 
3007 static int dma40_runtime_suspend(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, true);
3013 
3014 	/* Don't disable/enable clocks for v1 due to HW bugs */
3015 	if (base->rev != 1)
3016 		writel_relaxed(base->gcc_pwr_off_mask,
3017 			       base->virtbase + D40_DREG_GCC);
3018 
3019 	return 0;
3020 }
3021 
3022 static int dma40_runtime_resume(struct device *dev)
3023 {
3024 	struct platform_device *pdev = to_platform_device(dev);
3025 	struct d40_base *base = platform_get_drvdata(pdev);
3026 
3027 	if (base->initialized)
3028 		d40_save_restore_registers(base, false);
3029 
3030 	writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
3031 		       base->virtbase + D40_DREG_GCC);
3032 	return 0;
3033 }
3034 
3035 static int dma40_resume(struct device *dev)
3036 {
3037 	struct platform_device *pdev = to_platform_device(dev);
3038 	struct d40_base *base = platform_get_drvdata(pdev);
3039 	int ret = 0;
3040 
3041 	if (base->lcpa_regulator)
3042 		ret = regulator_enable(base->lcpa_regulator);
3043 
3044 	return ret;
3045 }
3046 
3047 static const struct dev_pm_ops dma40_pm_ops = {
3048 	.suspend		= dma40_pm_suspend,
3049 	.runtime_suspend	= dma40_runtime_suspend,
3050 	.runtime_resume		= dma40_runtime_resume,
3051 	.resume			= dma40_resume,
3052 };
3053 #define DMA40_PM_OPS	(&dma40_pm_ops)
3054 #else
3055 #define DMA40_PM_OPS	NULL
3056 #endif
3057 
3058 /* Initialization functions. */
3059 
3060 static int __init d40_phy_res_init(struct d40_base *base)
3061 {
3062 	int i;
3063 	int num_phy_chans_avail = 0;
3064 	u32 val[2];
3065 	int odd_even_bit = -2;
3066 	int gcc = D40_DREG_GCC_ENA;
3067 
3068 	val[0] = readl(base->virtbase + D40_DREG_PRSME);
3069 	val[1] = readl(base->virtbase + D40_DREG_PRSMO);
3070 
3071 	for (i = 0; i < base->num_phy_chans; i++) {
3072 		base->phy_res[i].num = i;
3073 		odd_even_bit += 2 * ((i % 2) == 0);
3074 		if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
3075 			/* Mark security only channels as occupied */
3076 			base->phy_res[i].allocated_src = D40_ALLOC_PHY;
3077 			base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
3078 			base->phy_res[i].reserved = true;
3079 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3080 						       D40_DREG_GCC_SRC);
3081 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3082 						       D40_DREG_GCC_DST);
3083 
3084 
3085 		} else {
3086 			base->phy_res[i].allocated_src = D40_ALLOC_FREE;
3087 			base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
3088 			base->phy_res[i].reserved = false;
3089 			num_phy_chans_avail++;
3090 		}
3091 		spin_lock_init(&base->phy_res[i].lock);
3092 	}
3093 
3094 	/* Mark disabled channels as occupied */
3095 	for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
3096 		int chan = base->plat_data->disabled_channels[i];
3097 
3098 		base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
3099 		base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
3100 		base->phy_res[chan].reserved = true;
3101 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3102 					       D40_DREG_GCC_SRC);
3103 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3104 					       D40_DREG_GCC_DST);
3105 		num_phy_chans_avail--;
3106 	}
3107 
3108 	/* Mark soft_lli channels */
3109 	for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) {
3110 		int chan = base->plat_data->soft_lli_chans[i];
3111 
3112 		base->phy_res[chan].use_soft_lli = true;
3113 	}
3114 
3115 	dev_info(base->dev, "%d of %d physical DMA channels available\n",
3116 		 num_phy_chans_avail, base->num_phy_chans);
3117 
3118 	/* Verify settings extended vs standard */
3119 	val[0] = readl(base->virtbase + D40_DREG_PRTYP);
3120 
3121 	for (i = 0; i < base->num_phy_chans; i++) {
3122 
3123 		if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
3124 		    (val[0] & 0x3) != 1)
3125 			dev_info(base->dev,
3126 				 "[%s] INFO: channel %d is misconfigured (%d)\n",
3127 				 __func__, i, val[0] & 0x3);
3128 
3129 		val[0] = val[0] >> 2;
3130 	}
3131 
3132 	/*
3133 	 * To keep things simple, Enable all clocks initially.
3134 	 * The clocks will get managed later post channel allocation.
3135 	 * The clocks for the event lines on which reserved channels exists
3136 	 * are not managed here.
3137 	 */
3138 	writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3139 	base->gcc_pwr_off_mask = gcc;
3140 
3141 	return num_phy_chans_avail;
3142 }
3143 
3144 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
3145 {
3146 	struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3147 	struct clk *clk = NULL;
3148 	void __iomem *virtbase = NULL;
3149 	struct resource *res = NULL;
3150 	struct d40_base *base = NULL;
3151 	int num_log_chans = 0;
3152 	int num_phy_chans;
3153 	int num_memcpy_chans;
3154 	int clk_ret = -EINVAL;
3155 	int i;
3156 	u32 pid;
3157 	u32 cid;
3158 	u8 rev;
3159 
3160 	clk = clk_get(&pdev->dev, NULL);
3161 	if (IS_ERR(clk)) {
3162 		d40_err(&pdev->dev, "No matching clock found\n");
3163 		goto failure;
3164 	}
3165 
3166 	clk_ret = clk_prepare_enable(clk);
3167 	if (clk_ret) {
3168 		d40_err(&pdev->dev, "Failed to prepare/enable clock\n");
3169 		goto failure;
3170 	}
3171 
3172 	/* Get IO for DMAC base address */
3173 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
3174 	if (!res)
3175 		goto failure;
3176 
3177 	if (request_mem_region(res->start, resource_size(res),
3178 			       D40_NAME " I/O base") == NULL)
3179 		goto failure;
3180 
3181 	virtbase = ioremap(res->start, resource_size(res));
3182 	if (!virtbase)
3183 		goto failure;
3184 
3185 	/* This is just a regular AMBA PrimeCell ID actually */
3186 	for (pid = 0, i = 0; i < 4; i++)
3187 		pid |= (readl(virtbase + resource_size(res) - 0x20 + 4 * i)
3188 			& 255) << (i * 8);
3189 	for (cid = 0, i = 0; i < 4; i++)
3190 		cid |= (readl(virtbase + resource_size(res) - 0x10 + 4 * i)
3191 			& 255) << (i * 8);
3192 
3193 	if (cid != AMBA_CID) {
3194 		d40_err(&pdev->dev, "Unknown hardware! No PrimeCell ID\n");
3195 		goto failure;
3196 	}
3197 	if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
3198 		d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
3199 			AMBA_MANF_BITS(pid),
3200 			AMBA_VENDOR_ST);
3201 		goto failure;
3202 	}
3203 	/*
3204 	 * HW revision:
3205 	 * DB8500ed has revision 0
3206 	 * ? has revision 1
3207 	 * DB8500v1 has revision 2
3208 	 * DB8500v2 has revision 3
3209 	 * AP9540v1 has revision 4
3210 	 * DB8540v1 has revision 4
3211 	 */
3212 	rev = AMBA_REV_BITS(pid);
3213 	if (rev < 2) {
3214 		d40_err(&pdev->dev, "hardware revision: %d is not supported", rev);
3215 		goto failure;
3216 	}
3217 
3218 	/* The number of physical channels on this HW */
3219 	if (plat_data->num_of_phy_chans)
3220 		num_phy_chans = plat_data->num_of_phy_chans;
3221 	else
3222 		num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
3223 
3224 	/* The number of channels used for memcpy */
3225 	if (plat_data->num_of_memcpy_chans)
3226 		num_memcpy_chans = plat_data->num_of_memcpy_chans;
3227 	else
3228 		num_memcpy_chans = ARRAY_SIZE(dma40_memcpy_channels);
3229 
3230 	num_log_chans = num_phy_chans * D40_MAX_LOG_CHAN_PER_PHY;
3231 
3232 	dev_info(&pdev->dev,
3233 		 "hardware rev: %d @ %pa with %d physical and %d logical channels\n",
3234 		 rev, &res->start, num_phy_chans, num_log_chans);
3235 
3236 	base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
3237 		       (num_phy_chans + num_log_chans + num_memcpy_chans) *
3238 		       sizeof(struct d40_chan), GFP_KERNEL);
3239 
3240 	if (base == NULL) {
3241 		d40_err(&pdev->dev, "Out of memory\n");
3242 		goto failure;
3243 	}
3244 
3245 	base->rev = rev;
3246 	base->clk = clk;
3247 	base->num_memcpy_chans = num_memcpy_chans;
3248 	base->num_phy_chans = num_phy_chans;
3249 	base->num_log_chans = num_log_chans;
3250 	base->phy_start = res->start;
3251 	base->phy_size = resource_size(res);
3252 	base->virtbase = virtbase;
3253 	base->plat_data = plat_data;
3254 	base->dev = &pdev->dev;
3255 	base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3256 	base->log_chans = &base->phy_chans[num_phy_chans];
3257 
3258 	if (base->plat_data->num_of_phy_chans == 14) {
3259 		base->gen_dmac.backup = d40_backup_regs_v4b;
3260 		base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
3261 		base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
3262 		base->gen_dmac.interrupt_clear = D40_DREG_CPCICR;
3263 		base->gen_dmac.realtime_en = D40_DREG_CRSEG1;
3264 		base->gen_dmac.realtime_clear = D40_DREG_CRCEG1;
3265 		base->gen_dmac.high_prio_en = D40_DREG_CPSEG1;
3266 		base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1;
3267 		base->gen_dmac.il = il_v4b;
3268 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4b);
3269 		base->gen_dmac.init_reg = dma_init_reg_v4b;
3270 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
3271 	} else {
3272 		if (base->rev >= 3) {
3273 			base->gen_dmac.backup = d40_backup_regs_v4a;
3274 			base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
3275 		}
3276 		base->gen_dmac.interrupt_en = D40_DREG_PCMIS;
3277 		base->gen_dmac.interrupt_clear = D40_DREG_PCICR;
3278 		base->gen_dmac.realtime_en = D40_DREG_RSEG1;
3279 		base->gen_dmac.realtime_clear = D40_DREG_RCEG1;
3280 		base->gen_dmac.high_prio_en = D40_DREG_PSEG1;
3281 		base->gen_dmac.high_prio_clear = D40_DREG_PCEG1;
3282 		base->gen_dmac.il = il_v4a;
3283 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4a);
3284 		base->gen_dmac.init_reg = dma_init_reg_v4a;
3285 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a);
3286 	}
3287 
3288 	base->phy_res = kzalloc(num_phy_chans * sizeof(struct d40_phy_res),
3289 				GFP_KERNEL);
3290 	if (!base->phy_res)
3291 		goto failure;
3292 
3293 	base->lookup_phy_chans = kzalloc(num_phy_chans *
3294 					 sizeof(struct d40_chan *),
3295 					 GFP_KERNEL);
3296 	if (!base->lookup_phy_chans)
3297 		goto failure;
3298 
3299 	base->lookup_log_chans = kzalloc(num_log_chans *
3300 					 sizeof(struct d40_chan *),
3301 					 GFP_KERNEL);
3302 	if (!base->lookup_log_chans)
3303 		goto failure;
3304 
3305 	base->reg_val_backup_chan = kmalloc(base->num_phy_chans *
3306 					    sizeof(d40_backup_regs_chan),
3307 					    GFP_KERNEL);
3308 	if (!base->reg_val_backup_chan)
3309 		goto failure;
3310 
3311 	base->lcla_pool.alloc_map =
3312 		kzalloc(num_phy_chans * sizeof(struct d40_desc *)
3313 			* D40_LCLA_LINK_PER_EVENT_GRP, GFP_KERNEL);
3314 	if (!base->lcla_pool.alloc_map)
3315 		goto failure;
3316 
3317 	base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3318 					    0, SLAB_HWCACHE_ALIGN,
3319 					    NULL);
3320 	if (base->desc_slab == NULL)
3321 		goto failure;
3322 
3323 	return base;
3324 
3325 failure:
3326 	if (!clk_ret)
3327 		clk_disable_unprepare(clk);
3328 	if (!IS_ERR(clk))
3329 		clk_put(clk);
3330 	if (virtbase)
3331 		iounmap(virtbase);
3332 	if (res)
3333 		release_mem_region(res->start,
3334 				   resource_size(res));
3335 	if (virtbase)
3336 		iounmap(virtbase);
3337 
3338 	if (base) {
3339 		kfree(base->lcla_pool.alloc_map);
3340 		kfree(base->reg_val_backup_chan);
3341 		kfree(base->lookup_log_chans);
3342 		kfree(base->lookup_phy_chans);
3343 		kfree(base->phy_res);
3344 		kfree(base);
3345 	}
3346 
3347 	return NULL;
3348 }
3349 
3350 static void __init d40_hw_init(struct d40_base *base)
3351 {
3352 
3353 	int i;
3354 	u32 prmseo[2] = {0, 0};
3355 	u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3356 	u32 pcmis = 0;
3357 	u32 pcicr = 0;
3358 	struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg;
3359 	u32 reg_size = base->gen_dmac.init_reg_size;
3360 
3361 	for (i = 0; i < reg_size; i++)
3362 		writel(dma_init_reg[i].val,
3363 		       base->virtbase + dma_init_reg[i].reg);
3364 
3365 	/* Configure all our dma channels to default settings */
3366 	for (i = 0; i < base->num_phy_chans; i++) {
3367 
3368 		activeo[i % 2] = activeo[i % 2] << 2;
3369 
3370 		if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3371 		    == D40_ALLOC_PHY) {
3372 			activeo[i % 2] |= 3;
3373 			continue;
3374 		}
3375 
3376 		/* Enable interrupt # */
3377 		pcmis = (pcmis << 1) | 1;
3378 
3379 		/* Clear interrupt # */
3380 		pcicr = (pcicr << 1) | 1;
3381 
3382 		/* Set channel to physical mode */
3383 		prmseo[i % 2] = prmseo[i % 2] << 2;
3384 		prmseo[i % 2] |= 1;
3385 
3386 	}
3387 
3388 	writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3389 	writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3390 	writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3391 	writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3392 
3393 	/* Write which interrupt to enable */
3394 	writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en);
3395 
3396 	/* Write which interrupt to clear */
3397 	writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear);
3398 
3399 	/* These are __initdata and cannot be accessed after init */
3400 	base->gen_dmac.init_reg = NULL;
3401 	base->gen_dmac.init_reg_size = 0;
3402 }
3403 
3404 static int __init d40_lcla_allocate(struct d40_base *base)
3405 {
3406 	struct d40_lcla_pool *pool = &base->lcla_pool;
3407 	unsigned long *page_list;
3408 	int i, j;
3409 	int ret = 0;
3410 
3411 	/*
3412 	 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3413 	 * To full fill this hardware requirement without wasting 256 kb
3414 	 * we allocate pages until we get an aligned one.
3415 	 */
3416 	page_list = kmalloc(sizeof(unsigned long) * MAX_LCLA_ALLOC_ATTEMPTS,
3417 			    GFP_KERNEL);
3418 
3419 	if (!page_list) {
3420 		ret = -ENOMEM;
3421 		goto failure;
3422 	}
3423 
3424 	/* Calculating how many pages that are required */
3425 	base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3426 
3427 	for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3428 		page_list[i] = __get_free_pages(GFP_KERNEL,
3429 						base->lcla_pool.pages);
3430 		if (!page_list[i]) {
3431 
3432 			d40_err(base->dev, "Failed to allocate %d pages.\n",
3433 				base->lcla_pool.pages);
3434 
3435 			for (j = 0; j < i; j++)
3436 				free_pages(page_list[j], base->lcla_pool.pages);
3437 			goto failure;
3438 		}
3439 
3440 		if ((virt_to_phys((void *)page_list[i]) &
3441 		     (LCLA_ALIGNMENT - 1)) == 0)
3442 			break;
3443 	}
3444 
3445 	for (j = 0; j < i; j++)
3446 		free_pages(page_list[j], base->lcla_pool.pages);
3447 
3448 	if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3449 		base->lcla_pool.base = (void *)page_list[i];
3450 	} else {
3451 		/*
3452 		 * After many attempts and no succees with finding the correct
3453 		 * alignment, try with allocating a big buffer.
3454 		 */
3455 		dev_warn(base->dev,
3456 			 "[%s] Failed to get %d pages @ 18 bit align.\n",
3457 			 __func__, base->lcla_pool.pages);
3458 		base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3459 							 base->num_phy_chans +
3460 							 LCLA_ALIGNMENT,
3461 							 GFP_KERNEL);
3462 		if (!base->lcla_pool.base_unaligned) {
3463 			ret = -ENOMEM;
3464 			goto failure;
3465 		}
3466 
3467 		base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3468 						 LCLA_ALIGNMENT);
3469 	}
3470 
3471 	pool->dma_addr = dma_map_single(base->dev, pool->base,
3472 					SZ_1K * base->num_phy_chans,
3473 					DMA_TO_DEVICE);
3474 	if (dma_mapping_error(base->dev, pool->dma_addr)) {
3475 		pool->dma_addr = 0;
3476 		ret = -ENOMEM;
3477 		goto failure;
3478 	}
3479 
3480 	writel(virt_to_phys(base->lcla_pool.base),
3481 	       base->virtbase + D40_DREG_LCLA);
3482 failure:
3483 	kfree(page_list);
3484 	return ret;
3485 }
3486 
3487 static int __init d40_of_probe(struct platform_device *pdev,
3488 			       struct device_node *np)
3489 {
3490 	struct stedma40_platform_data *pdata;
3491 	int num_phy = 0, num_memcpy = 0, num_disabled = 0;
3492 	const __be32 *list;
3493 
3494 	pdata = devm_kzalloc(&pdev->dev,
3495 			     sizeof(struct stedma40_platform_data),
3496 			     GFP_KERNEL);
3497 	if (!pdata)
3498 		return -ENOMEM;
3499 
3500 	/* If absent this value will be obtained from h/w. */
3501 	of_property_read_u32(np, "dma-channels", &num_phy);
3502 	if (num_phy > 0)
3503 		pdata->num_of_phy_chans = num_phy;
3504 
3505 	list = of_get_property(np, "memcpy-channels", &num_memcpy);
3506 	num_memcpy /= sizeof(*list);
3507 
3508 	if (num_memcpy > D40_MEMCPY_MAX_CHANS || num_memcpy <= 0) {
3509 		d40_err(&pdev->dev,
3510 			"Invalid number of memcpy channels specified (%d)\n",
3511 			num_memcpy);
3512 		return -EINVAL;
3513 	}
3514 	pdata->num_of_memcpy_chans = num_memcpy;
3515 
3516 	of_property_read_u32_array(np, "memcpy-channels",
3517 				   dma40_memcpy_channels,
3518 				   num_memcpy);
3519 
3520 	list = of_get_property(np, "disabled-channels", &num_disabled);
3521 	num_disabled /= sizeof(*list);
3522 
3523 	if (num_disabled >= STEDMA40_MAX_PHYS || num_disabled < 0) {
3524 		d40_err(&pdev->dev,
3525 			"Invalid number of disabled channels specified (%d)\n",
3526 			num_disabled);
3527 		return -EINVAL;
3528 	}
3529 
3530 	of_property_read_u32_array(np, "disabled-channels",
3531 				   pdata->disabled_channels,
3532 				   num_disabled);
3533 	pdata->disabled_channels[num_disabled] = -1;
3534 
3535 	pdev->dev.platform_data = pdata;
3536 
3537 	return 0;
3538 }
3539 
3540 static int __init d40_probe(struct platform_device *pdev)
3541 {
3542 	struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3543 	struct device_node *np = pdev->dev.of_node;
3544 	int ret = -ENOENT;
3545 	struct d40_base *base = NULL;
3546 	struct resource *res = NULL;
3547 	int num_reserved_chans;
3548 	u32 val;
3549 
3550 	if (!plat_data) {
3551 		if (np) {
3552 			if(d40_of_probe(pdev, np)) {
3553 				ret = -ENOMEM;
3554 				goto failure;
3555 			}
3556 		} else {
3557 			d40_err(&pdev->dev, "No pdata or Device Tree provided\n");
3558 			goto failure;
3559 		}
3560 	}
3561 
3562 	base = d40_hw_detect_init(pdev);
3563 	if (!base)
3564 		goto failure;
3565 
3566 	num_reserved_chans = d40_phy_res_init(base);
3567 
3568 	platform_set_drvdata(pdev, base);
3569 
3570 	spin_lock_init(&base->interrupt_lock);
3571 	spin_lock_init(&base->execmd_lock);
3572 
3573 	/* Get IO for logical channel parameter address */
3574 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
3575 	if (!res) {
3576 		ret = -ENOENT;
3577 		d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
3578 		goto failure;
3579 	}
3580 	base->lcpa_size = resource_size(res);
3581 	base->phy_lcpa = res->start;
3582 
3583 	if (request_mem_region(res->start, resource_size(res),
3584 			       D40_NAME " I/O lcpa") == NULL) {
3585 		ret = -EBUSY;
3586 		d40_err(&pdev->dev, "Failed to request LCPA region %pR\n", res);
3587 		goto failure;
3588 	}
3589 
3590 	/* We make use of ESRAM memory for this. */
3591 	val = readl(base->virtbase + D40_DREG_LCPA);
3592 	if (res->start != val && val != 0) {
3593 		dev_warn(&pdev->dev,
3594 			 "[%s] Mismatch LCPA dma 0x%x, def %pa\n",
3595 			 __func__, val, &res->start);
3596 	} else
3597 		writel(res->start, base->virtbase + D40_DREG_LCPA);
3598 
3599 	base->lcpa_base = ioremap(res->start, resource_size(res));
3600 	if (!base->lcpa_base) {
3601 		ret = -ENOMEM;
3602 		d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
3603 		goto failure;
3604 	}
3605 	/* If lcla has to be located in ESRAM we don't need to allocate */
3606 	if (base->plat_data->use_esram_lcla) {
3607 		res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3608 							"lcla_esram");
3609 		if (!res) {
3610 			ret = -ENOENT;
3611 			d40_err(&pdev->dev,
3612 				"No \"lcla_esram\" memory resource\n");
3613 			goto failure;
3614 		}
3615 		base->lcla_pool.base = ioremap(res->start,
3616 						resource_size(res));
3617 		if (!base->lcla_pool.base) {
3618 			ret = -ENOMEM;
3619 			d40_err(&pdev->dev, "Failed to ioremap LCLA region\n");
3620 			goto failure;
3621 		}
3622 		writel(res->start, base->virtbase + D40_DREG_LCLA);
3623 
3624 	} else {
3625 		ret = d40_lcla_allocate(base);
3626 		if (ret) {
3627 			d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
3628 			goto failure;
3629 		}
3630 	}
3631 
3632 	spin_lock_init(&base->lcla_pool.lock);
3633 
3634 	base->irq = platform_get_irq(pdev, 0);
3635 
3636 	ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3637 	if (ret) {
3638 		d40_err(&pdev->dev, "No IRQ defined\n");
3639 		goto failure;
3640 	}
3641 
3642 	pm_runtime_irq_safe(base->dev);
3643 	pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3644 	pm_runtime_use_autosuspend(base->dev);
3645 	pm_runtime_enable(base->dev);
3646 	pm_runtime_resume(base->dev);
3647 
3648 	if (base->plat_data->use_esram_lcla) {
3649 
3650 		base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3651 		if (IS_ERR(base->lcpa_regulator)) {
3652 			d40_err(&pdev->dev, "Failed to get lcpa_regulator\n");
3653 			ret = PTR_ERR(base->lcpa_regulator);
3654 			base->lcpa_regulator = NULL;
3655 			goto failure;
3656 		}
3657 
3658 		ret = regulator_enable(base->lcpa_regulator);
3659 		if (ret) {
3660 			d40_err(&pdev->dev,
3661 				"Failed to enable lcpa_regulator\n");
3662 			regulator_put(base->lcpa_regulator);
3663 			base->lcpa_regulator = NULL;
3664 			goto failure;
3665 		}
3666 	}
3667 
3668 	base->initialized = true;
3669 	ret = d40_dmaengine_init(base, num_reserved_chans);
3670 	if (ret)
3671 		goto failure;
3672 
3673 	base->dev->dma_parms = &base->dma_parms;
3674 	ret = dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
3675 	if (ret) {
3676 		d40_err(&pdev->dev, "Failed to set dma max seg size\n");
3677 		goto failure;
3678 	}
3679 
3680 	d40_hw_init(base);
3681 
3682 	if (np) {
3683 		ret = of_dma_controller_register(np, d40_xlate, NULL);
3684 		if (ret)
3685 			dev_err(&pdev->dev,
3686 				"could not register of_dma_controller\n");
3687 	}
3688 
3689 	dev_info(base->dev, "initialized\n");
3690 	return 0;
3691 
3692 failure:
3693 	if (base) {
3694 		if (base->desc_slab)
3695 			kmem_cache_destroy(base->desc_slab);
3696 		if (base->virtbase)
3697 			iounmap(base->virtbase);
3698 
3699 		if (base->lcla_pool.base && base->plat_data->use_esram_lcla) {
3700 			iounmap(base->lcla_pool.base);
3701 			base->lcla_pool.base = NULL;
3702 		}
3703 
3704 		if (base->lcla_pool.dma_addr)
3705 			dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3706 					 SZ_1K * base->num_phy_chans,
3707 					 DMA_TO_DEVICE);
3708 
3709 		if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3710 			free_pages((unsigned long)base->lcla_pool.base,
3711 				   base->lcla_pool.pages);
3712 
3713 		kfree(base->lcla_pool.base_unaligned);
3714 
3715 		if (base->phy_lcpa)
3716 			release_mem_region(base->phy_lcpa,
3717 					   base->lcpa_size);
3718 		if (base->phy_start)
3719 			release_mem_region(base->phy_start,
3720 					   base->phy_size);
3721 		if (base->clk) {
3722 			clk_disable_unprepare(base->clk);
3723 			clk_put(base->clk);
3724 		}
3725 
3726 		if (base->lcpa_regulator) {
3727 			regulator_disable(base->lcpa_regulator);
3728 			regulator_put(base->lcpa_regulator);
3729 		}
3730 
3731 		kfree(base->lcla_pool.alloc_map);
3732 		kfree(base->lookup_log_chans);
3733 		kfree(base->lookup_phy_chans);
3734 		kfree(base->phy_res);
3735 		kfree(base);
3736 	}
3737 
3738 	d40_err(&pdev->dev, "probe failed\n");
3739 	return ret;
3740 }
3741 
3742 static const struct of_device_id d40_match[] = {
3743         { .compatible = "stericsson,dma40", },
3744         {}
3745 };
3746 
3747 static struct platform_driver d40_driver = {
3748 	.driver = {
3749 		.owner = THIS_MODULE,
3750 		.name  = D40_NAME,
3751 		.pm = DMA40_PM_OPS,
3752 		.of_match_table = d40_match,
3753 	},
3754 };
3755 
3756 static int __init stedma40_init(void)
3757 {
3758 	return platform_driver_probe(&d40_driver, d40_probe);
3759 }
3760 subsys_initcall(stedma40_init);
3761