xref: /openbmc/linux/drivers/dma/ste_dma40.c (revision afb46f79)
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 	bool callback_active;
1645 	dma_async_tx_callback callback;
1646 	void *callback_param;
1647 
1648 	spin_lock_irqsave(&d40c->lock, flags);
1649 
1650 	/* Get first entry from the done list */
1651 	d40d = d40_first_done(d40c);
1652 	if (d40d == NULL) {
1653 		/* Check if we have reached here for cyclic job */
1654 		d40d = d40_first_active_get(d40c);
1655 		if (d40d == NULL || !d40d->cyclic)
1656 			goto err;
1657 	}
1658 
1659 	if (!d40d->cyclic)
1660 		dma_cookie_complete(&d40d->txd);
1661 
1662 	/*
1663 	 * If terminating a channel pending_tx is set to zero.
1664 	 * This prevents any finished active jobs to return to the client.
1665 	 */
1666 	if (d40c->pending_tx == 0) {
1667 		spin_unlock_irqrestore(&d40c->lock, flags);
1668 		return;
1669 	}
1670 
1671 	/* Callback to client */
1672 	callback_active = !!(d40d->txd.flags & DMA_PREP_INTERRUPT);
1673 	callback = d40d->txd.callback;
1674 	callback_param = d40d->txd.callback_param;
1675 
1676 	if (!d40d->cyclic) {
1677 		if (async_tx_test_ack(&d40d->txd)) {
1678 			d40_desc_remove(d40d);
1679 			d40_desc_free(d40c, d40d);
1680 		} else if (!d40d->is_in_client_list) {
1681 			d40_desc_remove(d40d);
1682 			d40_lcla_free_all(d40c, d40d);
1683 			list_add_tail(&d40d->node, &d40c->client);
1684 			d40d->is_in_client_list = true;
1685 		}
1686 	}
1687 
1688 	d40c->pending_tx--;
1689 
1690 	if (d40c->pending_tx)
1691 		tasklet_schedule(&d40c->tasklet);
1692 
1693 	spin_unlock_irqrestore(&d40c->lock, flags);
1694 
1695 	if (callback_active && callback)
1696 		callback(callback_param);
1697 
1698 	return;
1699 
1700 err:
1701 	/* Rescue manouver if receiving double interrupts */
1702 	if (d40c->pending_tx > 0)
1703 		d40c->pending_tx--;
1704 	spin_unlock_irqrestore(&d40c->lock, flags);
1705 }
1706 
1707 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1708 {
1709 	int i;
1710 	u32 idx;
1711 	u32 row;
1712 	long chan = -1;
1713 	struct d40_chan *d40c;
1714 	unsigned long flags;
1715 	struct d40_base *base = data;
1716 	u32 regs[base->gen_dmac.il_size];
1717 	struct d40_interrupt_lookup *il = base->gen_dmac.il;
1718 	u32 il_size = base->gen_dmac.il_size;
1719 
1720 	spin_lock_irqsave(&base->interrupt_lock, flags);
1721 
1722 	/* Read interrupt status of both logical and physical channels */
1723 	for (i = 0; i < il_size; i++)
1724 		regs[i] = readl(base->virtbase + il[i].src);
1725 
1726 	for (;;) {
1727 
1728 		chan = find_next_bit((unsigned long *)regs,
1729 				     BITS_PER_LONG * il_size, chan + 1);
1730 
1731 		/* No more set bits found? */
1732 		if (chan == BITS_PER_LONG * il_size)
1733 			break;
1734 
1735 		row = chan / BITS_PER_LONG;
1736 		idx = chan & (BITS_PER_LONG - 1);
1737 
1738 		if (il[row].offset == D40_PHY_CHAN)
1739 			d40c = base->lookup_phy_chans[idx];
1740 		else
1741 			d40c = base->lookup_log_chans[il[row].offset + idx];
1742 
1743 		if (!d40c) {
1744 			/*
1745 			 * No error because this can happen if something else
1746 			 * in the system is using the channel.
1747 			 */
1748 			continue;
1749 		}
1750 
1751 		/* ACK interrupt */
1752 		writel(BIT(idx), base->virtbase + il[row].clr);
1753 
1754 		spin_lock(&d40c->lock);
1755 
1756 		if (!il[row].is_error)
1757 			dma_tc_handle(d40c);
1758 		else
1759 			d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1760 				chan, il[row].offset, idx);
1761 
1762 		spin_unlock(&d40c->lock);
1763 	}
1764 
1765 	spin_unlock_irqrestore(&base->interrupt_lock, flags);
1766 
1767 	return IRQ_HANDLED;
1768 }
1769 
1770 static int d40_validate_conf(struct d40_chan *d40c,
1771 			     struct stedma40_chan_cfg *conf)
1772 {
1773 	int res = 0;
1774 	bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1775 
1776 	if (!conf->dir) {
1777 		chan_err(d40c, "Invalid direction.\n");
1778 		res = -EINVAL;
1779 	}
1780 
1781 	if ((is_log && conf->dev_type > d40c->base->num_log_chans)  ||
1782 	    (!is_log && conf->dev_type > d40c->base->num_phy_chans) ||
1783 	    (conf->dev_type < 0)) {
1784 		chan_err(d40c, "Invalid device type (%d)\n", conf->dev_type);
1785 		res = -EINVAL;
1786 	}
1787 
1788 	if (conf->dir == DMA_DEV_TO_DEV) {
1789 		/*
1790 		 * DMAC HW supports it. Will be added to this driver,
1791 		 * in case any dma client requires it.
1792 		 */
1793 		chan_err(d40c, "periph to periph not supported\n");
1794 		res = -EINVAL;
1795 	}
1796 
1797 	if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1798 	    conf->src_info.data_width !=
1799 	    d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1800 	    conf->dst_info.data_width) {
1801 		/*
1802 		 * The DMAC hardware only supports
1803 		 * src (burst x width) == dst (burst x width)
1804 		 */
1805 
1806 		chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1807 		res = -EINVAL;
1808 	}
1809 
1810 	return res;
1811 }
1812 
1813 static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1814 			       bool is_src, int log_event_line, bool is_log,
1815 			       bool *first_user)
1816 {
1817 	unsigned long flags;
1818 	spin_lock_irqsave(&phy->lock, flags);
1819 
1820 	*first_user = ((phy->allocated_src | phy->allocated_dst)
1821 			== D40_ALLOC_FREE);
1822 
1823 	if (!is_log) {
1824 		/* Physical interrupts are masked per physical full channel */
1825 		if (phy->allocated_src == D40_ALLOC_FREE &&
1826 		    phy->allocated_dst == D40_ALLOC_FREE) {
1827 			phy->allocated_dst = D40_ALLOC_PHY;
1828 			phy->allocated_src = D40_ALLOC_PHY;
1829 			goto found;
1830 		} else
1831 			goto not_found;
1832 	}
1833 
1834 	/* Logical channel */
1835 	if (is_src) {
1836 		if (phy->allocated_src == D40_ALLOC_PHY)
1837 			goto not_found;
1838 
1839 		if (phy->allocated_src == D40_ALLOC_FREE)
1840 			phy->allocated_src = D40_ALLOC_LOG_FREE;
1841 
1842 		if (!(phy->allocated_src & BIT(log_event_line))) {
1843 			phy->allocated_src |= BIT(log_event_line);
1844 			goto found;
1845 		} else
1846 			goto not_found;
1847 	} else {
1848 		if (phy->allocated_dst == D40_ALLOC_PHY)
1849 			goto not_found;
1850 
1851 		if (phy->allocated_dst == D40_ALLOC_FREE)
1852 			phy->allocated_dst = D40_ALLOC_LOG_FREE;
1853 
1854 		if (!(phy->allocated_dst & BIT(log_event_line))) {
1855 			phy->allocated_dst |= BIT(log_event_line);
1856 			goto found;
1857 		} else
1858 			goto not_found;
1859 	}
1860 
1861 not_found:
1862 	spin_unlock_irqrestore(&phy->lock, flags);
1863 	return false;
1864 found:
1865 	spin_unlock_irqrestore(&phy->lock, flags);
1866 	return true;
1867 }
1868 
1869 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1870 			       int log_event_line)
1871 {
1872 	unsigned long flags;
1873 	bool is_free = false;
1874 
1875 	spin_lock_irqsave(&phy->lock, flags);
1876 	if (!log_event_line) {
1877 		phy->allocated_dst = D40_ALLOC_FREE;
1878 		phy->allocated_src = D40_ALLOC_FREE;
1879 		is_free = true;
1880 		goto out;
1881 	}
1882 
1883 	/* Logical channel */
1884 	if (is_src) {
1885 		phy->allocated_src &= ~BIT(log_event_line);
1886 		if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1887 			phy->allocated_src = D40_ALLOC_FREE;
1888 	} else {
1889 		phy->allocated_dst &= ~BIT(log_event_line);
1890 		if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1891 			phy->allocated_dst = D40_ALLOC_FREE;
1892 	}
1893 
1894 	is_free = ((phy->allocated_src | phy->allocated_dst) ==
1895 		   D40_ALLOC_FREE);
1896 
1897 out:
1898 	spin_unlock_irqrestore(&phy->lock, flags);
1899 
1900 	return is_free;
1901 }
1902 
1903 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1904 {
1905 	int dev_type = d40c->dma_cfg.dev_type;
1906 	int event_group;
1907 	int event_line;
1908 	struct d40_phy_res *phys;
1909 	int i;
1910 	int j;
1911 	int log_num;
1912 	int num_phy_chans;
1913 	bool is_src;
1914 	bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1915 
1916 	phys = d40c->base->phy_res;
1917 	num_phy_chans = d40c->base->num_phy_chans;
1918 
1919 	if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
1920 		log_num = 2 * dev_type;
1921 		is_src = true;
1922 	} else if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
1923 		   d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1924 		/* dst event lines are used for logical memcpy */
1925 		log_num = 2 * dev_type + 1;
1926 		is_src = false;
1927 	} else
1928 		return -EINVAL;
1929 
1930 	event_group = D40_TYPE_TO_GROUP(dev_type);
1931 	event_line = D40_TYPE_TO_EVENT(dev_type);
1932 
1933 	if (!is_log) {
1934 		if (d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1935 			/* Find physical half channel */
1936 			if (d40c->dma_cfg.use_fixed_channel) {
1937 				i = d40c->dma_cfg.phy_channel;
1938 				if (d40_alloc_mask_set(&phys[i], is_src,
1939 						       0, is_log,
1940 						       first_phy_user))
1941 					goto found_phy;
1942 			} else {
1943 				for (i = 0; i < num_phy_chans; i++) {
1944 					if (d40_alloc_mask_set(&phys[i], is_src,
1945 						       0, is_log,
1946 						       first_phy_user))
1947 						goto found_phy;
1948 				}
1949 			}
1950 		} else
1951 			for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1952 				int phy_num = j  + event_group * 2;
1953 				for (i = phy_num; i < phy_num + 2; i++) {
1954 					if (d40_alloc_mask_set(&phys[i],
1955 							       is_src,
1956 							       0,
1957 							       is_log,
1958 							       first_phy_user))
1959 						goto found_phy;
1960 				}
1961 			}
1962 		return -EINVAL;
1963 found_phy:
1964 		d40c->phy_chan = &phys[i];
1965 		d40c->log_num = D40_PHY_CHAN;
1966 		goto out;
1967 	}
1968 	if (dev_type == -1)
1969 		return -EINVAL;
1970 
1971 	/* Find logical channel */
1972 	for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1973 		int phy_num = j + event_group * 2;
1974 
1975 		if (d40c->dma_cfg.use_fixed_channel) {
1976 			i = d40c->dma_cfg.phy_channel;
1977 
1978 			if ((i != phy_num) && (i != phy_num + 1)) {
1979 				dev_err(chan2dev(d40c),
1980 					"invalid fixed phy channel %d\n", i);
1981 				return -EINVAL;
1982 			}
1983 
1984 			if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1985 					       is_log, first_phy_user))
1986 				goto found_log;
1987 
1988 			dev_err(chan2dev(d40c),
1989 				"could not allocate fixed phy channel %d\n", i);
1990 			return -EINVAL;
1991 		}
1992 
1993 		/*
1994 		 * Spread logical channels across all available physical rather
1995 		 * than pack every logical channel at the first available phy
1996 		 * channels.
1997 		 */
1998 		if (is_src) {
1999 			for (i = phy_num; i < phy_num + 2; i++) {
2000 				if (d40_alloc_mask_set(&phys[i], is_src,
2001 						       event_line, is_log,
2002 						       first_phy_user))
2003 					goto found_log;
2004 			}
2005 		} else {
2006 			for (i = phy_num + 1; i >= phy_num; i--) {
2007 				if (d40_alloc_mask_set(&phys[i], is_src,
2008 						       event_line, is_log,
2009 						       first_phy_user))
2010 					goto found_log;
2011 			}
2012 		}
2013 	}
2014 	return -EINVAL;
2015 
2016 found_log:
2017 	d40c->phy_chan = &phys[i];
2018 	d40c->log_num = log_num;
2019 out:
2020 
2021 	if (is_log)
2022 		d40c->base->lookup_log_chans[d40c->log_num] = d40c;
2023 	else
2024 		d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
2025 
2026 	return 0;
2027 
2028 }
2029 
2030 static int d40_config_memcpy(struct d40_chan *d40c)
2031 {
2032 	dma_cap_mask_t cap = d40c->chan.device->cap_mask;
2033 
2034 	if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
2035 		d40c->dma_cfg = dma40_memcpy_conf_log;
2036 		d40c->dma_cfg.dev_type = dma40_memcpy_channels[d40c->chan.chan_id];
2037 
2038 		d40_log_cfg(&d40c->dma_cfg,
2039 			    &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2040 
2041 	} else if (dma_has_cap(DMA_MEMCPY, cap) &&
2042 		   dma_has_cap(DMA_SLAVE, cap)) {
2043 		d40c->dma_cfg = dma40_memcpy_conf_phy;
2044 
2045 		/* Generate interrrupt at end of transfer or relink. */
2046 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_TIM_POS);
2047 
2048 		/* Generate interrupt on error. */
2049 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
2050 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
2051 
2052 	} else {
2053 		chan_err(d40c, "No memcpy\n");
2054 		return -EINVAL;
2055 	}
2056 
2057 	return 0;
2058 }
2059 
2060 static int d40_free_dma(struct d40_chan *d40c)
2061 {
2062 
2063 	int res = 0;
2064 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2065 	struct d40_phy_res *phy = d40c->phy_chan;
2066 	bool is_src;
2067 
2068 	/* Terminate all queued and active transfers */
2069 	d40_term_all(d40c);
2070 
2071 	if (phy == NULL) {
2072 		chan_err(d40c, "phy == null\n");
2073 		return -EINVAL;
2074 	}
2075 
2076 	if (phy->allocated_src == D40_ALLOC_FREE &&
2077 	    phy->allocated_dst == D40_ALLOC_FREE) {
2078 		chan_err(d40c, "channel already free\n");
2079 		return -EINVAL;
2080 	}
2081 
2082 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2083 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM)
2084 		is_src = false;
2085 	else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2086 		is_src = true;
2087 	else {
2088 		chan_err(d40c, "Unknown direction\n");
2089 		return -EINVAL;
2090 	}
2091 
2092 	pm_runtime_get_sync(d40c->base->dev);
2093 	res = d40_channel_execute_command(d40c, D40_DMA_STOP);
2094 	if (res) {
2095 		chan_err(d40c, "stop failed\n");
2096 		goto out;
2097 	}
2098 
2099 	d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
2100 
2101 	if (chan_is_logical(d40c))
2102 		d40c->base->lookup_log_chans[d40c->log_num] = NULL;
2103 	else
2104 		d40c->base->lookup_phy_chans[phy->num] = NULL;
2105 
2106 	if (d40c->busy) {
2107 		pm_runtime_mark_last_busy(d40c->base->dev);
2108 		pm_runtime_put_autosuspend(d40c->base->dev);
2109 	}
2110 
2111 	d40c->busy = false;
2112 	d40c->phy_chan = NULL;
2113 	d40c->configured = false;
2114 out:
2115 
2116 	pm_runtime_mark_last_busy(d40c->base->dev);
2117 	pm_runtime_put_autosuspend(d40c->base->dev);
2118 	return res;
2119 }
2120 
2121 static bool d40_is_paused(struct d40_chan *d40c)
2122 {
2123 	void __iomem *chanbase = chan_base(d40c);
2124 	bool is_paused = false;
2125 	unsigned long flags;
2126 	void __iomem *active_reg;
2127 	u32 status;
2128 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2129 
2130 	spin_lock_irqsave(&d40c->lock, flags);
2131 
2132 	if (chan_is_physical(d40c)) {
2133 		if (d40c->phy_chan->num % 2 == 0)
2134 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
2135 		else
2136 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
2137 
2138 		status = (readl(active_reg) &
2139 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
2140 			D40_CHAN_POS(d40c->phy_chan->num);
2141 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
2142 			is_paused = true;
2143 
2144 		goto _exit;
2145 	}
2146 
2147 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2148 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
2149 		status = readl(chanbase + D40_CHAN_REG_SDLNK);
2150 	} else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
2151 		status = readl(chanbase + D40_CHAN_REG_SSLNK);
2152 	} else {
2153 		chan_err(d40c, "Unknown direction\n");
2154 		goto _exit;
2155 	}
2156 
2157 	status = (status & D40_EVENTLINE_MASK(event)) >>
2158 		D40_EVENTLINE_POS(event);
2159 
2160 	if (status != D40_DMA_RUN)
2161 		is_paused = true;
2162 _exit:
2163 	spin_unlock_irqrestore(&d40c->lock, flags);
2164 	return is_paused;
2165 
2166 }
2167 
2168 static u32 stedma40_residue(struct dma_chan *chan)
2169 {
2170 	struct d40_chan *d40c =
2171 		container_of(chan, struct d40_chan, chan);
2172 	u32 bytes_left;
2173 	unsigned long flags;
2174 
2175 	spin_lock_irqsave(&d40c->lock, flags);
2176 	bytes_left = d40_residue(d40c);
2177 	spin_unlock_irqrestore(&d40c->lock, flags);
2178 
2179 	return bytes_left;
2180 }
2181 
2182 static int
2183 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
2184 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2185 		unsigned int sg_len, dma_addr_t src_dev_addr,
2186 		dma_addr_t dst_dev_addr)
2187 {
2188 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2189 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2190 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2191 	int ret;
2192 
2193 	ret = d40_log_sg_to_lli(sg_src, sg_len,
2194 				src_dev_addr,
2195 				desc->lli_log.src,
2196 				chan->log_def.lcsp1,
2197 				src_info->data_width,
2198 				dst_info->data_width);
2199 
2200 	ret = d40_log_sg_to_lli(sg_dst, sg_len,
2201 				dst_dev_addr,
2202 				desc->lli_log.dst,
2203 				chan->log_def.lcsp3,
2204 				dst_info->data_width,
2205 				src_info->data_width);
2206 
2207 	return ret < 0 ? ret : 0;
2208 }
2209 
2210 static int
2211 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2212 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2213 		unsigned int sg_len, dma_addr_t src_dev_addr,
2214 		dma_addr_t dst_dev_addr)
2215 {
2216 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2217 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2218 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2219 	unsigned long flags = 0;
2220 	int ret;
2221 
2222 	if (desc->cyclic)
2223 		flags |= LLI_CYCLIC | LLI_TERM_INT;
2224 
2225 	ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2226 				desc->lli_phy.src,
2227 				virt_to_phys(desc->lli_phy.src),
2228 				chan->src_def_cfg,
2229 				src_info, dst_info, flags);
2230 
2231 	ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2232 				desc->lli_phy.dst,
2233 				virt_to_phys(desc->lli_phy.dst),
2234 				chan->dst_def_cfg,
2235 				dst_info, src_info, flags);
2236 
2237 	dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2238 				   desc->lli_pool.size, DMA_TO_DEVICE);
2239 
2240 	return ret < 0 ? ret : 0;
2241 }
2242 
2243 static struct d40_desc *
2244 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2245 	      unsigned int sg_len, unsigned long dma_flags)
2246 {
2247 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2248 	struct d40_desc *desc;
2249 	int ret;
2250 
2251 	desc = d40_desc_get(chan);
2252 	if (!desc)
2253 		return NULL;
2254 
2255 	desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2256 					cfg->dst_info.data_width);
2257 	if (desc->lli_len < 0) {
2258 		chan_err(chan, "Unaligned size\n");
2259 		goto err;
2260 	}
2261 
2262 	ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2263 	if (ret < 0) {
2264 		chan_err(chan, "Could not allocate lli\n");
2265 		goto err;
2266 	}
2267 
2268 	desc->lli_current = 0;
2269 	desc->txd.flags = dma_flags;
2270 	desc->txd.tx_submit = d40_tx_submit;
2271 
2272 	dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2273 
2274 	return desc;
2275 
2276 err:
2277 	d40_desc_free(chan, desc);
2278 	return NULL;
2279 }
2280 
2281 static struct dma_async_tx_descriptor *
2282 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2283 	    struct scatterlist *sg_dst, unsigned int sg_len,
2284 	    enum dma_transfer_direction direction, unsigned long dma_flags)
2285 {
2286 	struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2287 	dma_addr_t src_dev_addr = 0;
2288 	dma_addr_t dst_dev_addr = 0;
2289 	struct d40_desc *desc;
2290 	unsigned long flags;
2291 	int ret;
2292 
2293 	if (!chan->phy_chan) {
2294 		chan_err(chan, "Cannot prepare unallocated channel\n");
2295 		return NULL;
2296 	}
2297 
2298 	spin_lock_irqsave(&chan->lock, flags);
2299 
2300 	desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2301 	if (desc == NULL)
2302 		goto err;
2303 
2304 	if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2305 		desc->cyclic = true;
2306 
2307 	if (direction == DMA_DEV_TO_MEM)
2308 		src_dev_addr = chan->runtime_addr;
2309 	else if (direction == DMA_MEM_TO_DEV)
2310 		dst_dev_addr = chan->runtime_addr;
2311 
2312 	if (chan_is_logical(chan))
2313 		ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2314 				      sg_len, src_dev_addr, dst_dev_addr);
2315 	else
2316 		ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2317 				      sg_len, src_dev_addr, dst_dev_addr);
2318 
2319 	if (ret) {
2320 		chan_err(chan, "Failed to prepare %s sg job: %d\n",
2321 			 chan_is_logical(chan) ? "log" : "phy", ret);
2322 		goto err;
2323 	}
2324 
2325 	/*
2326 	 * add descriptor to the prepare queue in order to be able
2327 	 * to free them later in terminate_all
2328 	 */
2329 	list_add_tail(&desc->node, &chan->prepare_queue);
2330 
2331 	spin_unlock_irqrestore(&chan->lock, flags);
2332 
2333 	return &desc->txd;
2334 
2335 err:
2336 	if (desc)
2337 		d40_desc_free(chan, desc);
2338 	spin_unlock_irqrestore(&chan->lock, flags);
2339 	return NULL;
2340 }
2341 
2342 bool stedma40_filter(struct dma_chan *chan, void *data)
2343 {
2344 	struct stedma40_chan_cfg *info = data;
2345 	struct d40_chan *d40c =
2346 		container_of(chan, struct d40_chan, chan);
2347 	int err;
2348 
2349 	if (data) {
2350 		err = d40_validate_conf(d40c, info);
2351 		if (!err)
2352 			d40c->dma_cfg = *info;
2353 	} else
2354 		err = d40_config_memcpy(d40c);
2355 
2356 	if (!err)
2357 		d40c->configured = true;
2358 
2359 	return err == 0;
2360 }
2361 EXPORT_SYMBOL(stedma40_filter);
2362 
2363 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2364 {
2365 	bool realtime = d40c->dma_cfg.realtime;
2366 	bool highprio = d40c->dma_cfg.high_priority;
2367 	u32 rtreg;
2368 	u32 event = D40_TYPE_TO_EVENT(dev_type);
2369 	u32 group = D40_TYPE_TO_GROUP(dev_type);
2370 	u32 bit = BIT(event);
2371 	u32 prioreg;
2372 	struct d40_gen_dmac *dmac = &d40c->base->gen_dmac;
2373 
2374 	rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear;
2375 	/*
2376 	 * Due to a hardware bug, in some cases a logical channel triggered by
2377 	 * a high priority destination event line can generate extra packet
2378 	 * transactions.
2379 	 *
2380 	 * The workaround is to not set the high priority level for the
2381 	 * destination event lines that trigger logical channels.
2382 	 */
2383 	if (!src && chan_is_logical(d40c))
2384 		highprio = false;
2385 
2386 	prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear;
2387 
2388 	/* Destination event lines are stored in the upper halfword */
2389 	if (!src)
2390 		bit <<= 16;
2391 
2392 	writel(bit, d40c->base->virtbase + prioreg + group * 4);
2393 	writel(bit, d40c->base->virtbase + rtreg + group * 4);
2394 }
2395 
2396 static void d40_set_prio_realtime(struct d40_chan *d40c)
2397 {
2398 	if (d40c->base->rev < 3)
2399 		return;
2400 
2401 	if ((d40c->dma_cfg.dir ==  DMA_DEV_TO_MEM) ||
2402 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2403 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, true);
2404 
2405 	if ((d40c->dma_cfg.dir ==  DMA_MEM_TO_DEV) ||
2406 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2407 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, false);
2408 }
2409 
2410 #define D40_DT_FLAGS_MODE(flags)       ((flags >> 0) & 0x1)
2411 #define D40_DT_FLAGS_DIR(flags)        ((flags >> 1) & 0x1)
2412 #define D40_DT_FLAGS_BIG_ENDIAN(flags) ((flags >> 2) & 0x1)
2413 #define D40_DT_FLAGS_FIXED_CHAN(flags) ((flags >> 3) & 0x1)
2414 #define D40_DT_FLAGS_HIGH_PRIO(flags)  ((flags >> 4) & 0x1)
2415 
2416 static struct dma_chan *d40_xlate(struct of_phandle_args *dma_spec,
2417 				  struct of_dma *ofdma)
2418 {
2419 	struct stedma40_chan_cfg cfg;
2420 	dma_cap_mask_t cap;
2421 	u32 flags;
2422 
2423 	memset(&cfg, 0, sizeof(struct stedma40_chan_cfg));
2424 
2425 	dma_cap_zero(cap);
2426 	dma_cap_set(DMA_SLAVE, cap);
2427 
2428 	cfg.dev_type = dma_spec->args[0];
2429 	flags = dma_spec->args[2];
2430 
2431 	switch (D40_DT_FLAGS_MODE(flags)) {
2432 	case 0: cfg.mode = STEDMA40_MODE_LOGICAL; break;
2433 	case 1: cfg.mode = STEDMA40_MODE_PHYSICAL; break;
2434 	}
2435 
2436 	switch (D40_DT_FLAGS_DIR(flags)) {
2437 	case 0:
2438 		cfg.dir = DMA_MEM_TO_DEV;
2439 		cfg.dst_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2440 		break;
2441 	case 1:
2442 		cfg.dir = DMA_DEV_TO_MEM;
2443 		cfg.src_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2444 		break;
2445 	}
2446 
2447 	if (D40_DT_FLAGS_FIXED_CHAN(flags)) {
2448 		cfg.phy_channel = dma_spec->args[1];
2449 		cfg.use_fixed_channel = true;
2450 	}
2451 
2452 	if (D40_DT_FLAGS_HIGH_PRIO(flags))
2453 		cfg.high_priority = true;
2454 
2455 	return dma_request_channel(cap, stedma40_filter, &cfg);
2456 }
2457 
2458 /* DMA ENGINE functions */
2459 static int d40_alloc_chan_resources(struct dma_chan *chan)
2460 {
2461 	int err;
2462 	unsigned long flags;
2463 	struct d40_chan *d40c =
2464 		container_of(chan, struct d40_chan, chan);
2465 	bool is_free_phy;
2466 	spin_lock_irqsave(&d40c->lock, flags);
2467 
2468 	dma_cookie_init(chan);
2469 
2470 	/* If no dma configuration is set use default configuration (memcpy) */
2471 	if (!d40c->configured) {
2472 		err = d40_config_memcpy(d40c);
2473 		if (err) {
2474 			chan_err(d40c, "Failed to configure memcpy channel\n");
2475 			goto fail;
2476 		}
2477 	}
2478 
2479 	err = d40_allocate_channel(d40c, &is_free_phy);
2480 	if (err) {
2481 		chan_err(d40c, "Failed to allocate channel\n");
2482 		d40c->configured = false;
2483 		goto fail;
2484 	}
2485 
2486 	pm_runtime_get_sync(d40c->base->dev);
2487 
2488 	d40_set_prio_realtime(d40c);
2489 
2490 	if (chan_is_logical(d40c)) {
2491 		if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2492 			d40c->lcpa = d40c->base->lcpa_base +
2493 				d40c->dma_cfg.dev_type * D40_LCPA_CHAN_SIZE;
2494 		else
2495 			d40c->lcpa = d40c->base->lcpa_base +
2496 				d40c->dma_cfg.dev_type *
2497 				D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2498 
2499 		/* Unmask the Global Interrupt Mask. */
2500 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2501 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2502 	}
2503 
2504 	dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2505 		 chan_is_logical(d40c) ? "logical" : "physical",
2506 		 d40c->phy_chan->num,
2507 		 d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2508 
2509 
2510 	/*
2511 	 * Only write channel configuration to the DMA if the physical
2512 	 * resource is free. In case of multiple logical channels
2513 	 * on the same physical resource, only the first write is necessary.
2514 	 */
2515 	if (is_free_phy)
2516 		d40_config_write(d40c);
2517 fail:
2518 	pm_runtime_mark_last_busy(d40c->base->dev);
2519 	pm_runtime_put_autosuspend(d40c->base->dev);
2520 	spin_unlock_irqrestore(&d40c->lock, flags);
2521 	return err;
2522 }
2523 
2524 static void d40_free_chan_resources(struct dma_chan *chan)
2525 {
2526 	struct d40_chan *d40c =
2527 		container_of(chan, struct d40_chan, chan);
2528 	int err;
2529 	unsigned long flags;
2530 
2531 	if (d40c->phy_chan == NULL) {
2532 		chan_err(d40c, "Cannot free unallocated channel\n");
2533 		return;
2534 	}
2535 
2536 	spin_lock_irqsave(&d40c->lock, flags);
2537 
2538 	err = d40_free_dma(d40c);
2539 
2540 	if (err)
2541 		chan_err(d40c, "Failed to free channel\n");
2542 	spin_unlock_irqrestore(&d40c->lock, flags);
2543 }
2544 
2545 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2546 						       dma_addr_t dst,
2547 						       dma_addr_t src,
2548 						       size_t size,
2549 						       unsigned long dma_flags)
2550 {
2551 	struct scatterlist dst_sg;
2552 	struct scatterlist src_sg;
2553 
2554 	sg_init_table(&dst_sg, 1);
2555 	sg_init_table(&src_sg, 1);
2556 
2557 	sg_dma_address(&dst_sg) = dst;
2558 	sg_dma_address(&src_sg) = src;
2559 
2560 	sg_dma_len(&dst_sg) = size;
2561 	sg_dma_len(&src_sg) = size;
2562 
2563 	return d40_prep_sg(chan, &src_sg, &dst_sg, 1, DMA_NONE, dma_flags);
2564 }
2565 
2566 static struct dma_async_tx_descriptor *
2567 d40_prep_memcpy_sg(struct dma_chan *chan,
2568 		   struct scatterlist *dst_sg, unsigned int dst_nents,
2569 		   struct scatterlist *src_sg, unsigned int src_nents,
2570 		   unsigned long dma_flags)
2571 {
2572 	if (dst_nents != src_nents)
2573 		return NULL;
2574 
2575 	return d40_prep_sg(chan, src_sg, dst_sg, src_nents, DMA_NONE, dma_flags);
2576 }
2577 
2578 static struct dma_async_tx_descriptor *
2579 d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2580 		  unsigned int sg_len, enum dma_transfer_direction direction,
2581 		  unsigned long dma_flags, void *context)
2582 {
2583 	if (!is_slave_direction(direction))
2584 		return NULL;
2585 
2586 	return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2587 }
2588 
2589 static struct dma_async_tx_descriptor *
2590 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2591 		     size_t buf_len, size_t period_len,
2592 		     enum dma_transfer_direction direction, unsigned long flags,
2593 		     void *context)
2594 {
2595 	unsigned int periods = buf_len / period_len;
2596 	struct dma_async_tx_descriptor *txd;
2597 	struct scatterlist *sg;
2598 	int i;
2599 
2600 	sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2601 	if (!sg)
2602 		return NULL;
2603 
2604 	for (i = 0; i < periods; i++) {
2605 		sg_dma_address(&sg[i]) = dma_addr;
2606 		sg_dma_len(&sg[i]) = period_len;
2607 		dma_addr += period_len;
2608 	}
2609 
2610 	sg[periods].offset = 0;
2611 	sg_dma_len(&sg[periods]) = 0;
2612 	sg[periods].page_link =
2613 		((unsigned long)sg | 0x01) & ~0x02;
2614 
2615 	txd = d40_prep_sg(chan, sg, sg, periods, direction,
2616 			  DMA_PREP_INTERRUPT);
2617 
2618 	kfree(sg);
2619 
2620 	return txd;
2621 }
2622 
2623 static enum dma_status d40_tx_status(struct dma_chan *chan,
2624 				     dma_cookie_t cookie,
2625 				     struct dma_tx_state *txstate)
2626 {
2627 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2628 	enum dma_status ret;
2629 
2630 	if (d40c->phy_chan == NULL) {
2631 		chan_err(d40c, "Cannot read status of unallocated channel\n");
2632 		return -EINVAL;
2633 	}
2634 
2635 	ret = dma_cookie_status(chan, cookie, txstate);
2636 	if (ret != DMA_COMPLETE)
2637 		dma_set_residue(txstate, stedma40_residue(chan));
2638 
2639 	if (d40_is_paused(d40c))
2640 		ret = DMA_PAUSED;
2641 
2642 	return ret;
2643 }
2644 
2645 static void d40_issue_pending(struct dma_chan *chan)
2646 {
2647 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2648 	unsigned long flags;
2649 
2650 	if (d40c->phy_chan == NULL) {
2651 		chan_err(d40c, "Channel is not allocated!\n");
2652 		return;
2653 	}
2654 
2655 	spin_lock_irqsave(&d40c->lock, flags);
2656 
2657 	list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2658 
2659 	/* Busy means that queued jobs are already being processed */
2660 	if (!d40c->busy)
2661 		(void) d40_queue_start(d40c);
2662 
2663 	spin_unlock_irqrestore(&d40c->lock, flags);
2664 }
2665 
2666 static void d40_terminate_all(struct dma_chan *chan)
2667 {
2668 	unsigned long flags;
2669 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2670 	int ret;
2671 
2672 	spin_lock_irqsave(&d40c->lock, flags);
2673 
2674 	pm_runtime_get_sync(d40c->base->dev);
2675 	ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2676 	if (ret)
2677 		chan_err(d40c, "Failed to stop channel\n");
2678 
2679 	d40_term_all(d40c);
2680 	pm_runtime_mark_last_busy(d40c->base->dev);
2681 	pm_runtime_put_autosuspend(d40c->base->dev);
2682 	if (d40c->busy) {
2683 		pm_runtime_mark_last_busy(d40c->base->dev);
2684 		pm_runtime_put_autosuspend(d40c->base->dev);
2685 	}
2686 	d40c->busy = false;
2687 
2688 	spin_unlock_irqrestore(&d40c->lock, flags);
2689 }
2690 
2691 static int
2692 dma40_config_to_halfchannel(struct d40_chan *d40c,
2693 			    struct stedma40_half_channel_info *info,
2694 			    u32 maxburst)
2695 {
2696 	int psize;
2697 
2698 	if (chan_is_logical(d40c)) {
2699 		if (maxburst >= 16)
2700 			psize = STEDMA40_PSIZE_LOG_16;
2701 		else if (maxburst >= 8)
2702 			psize = STEDMA40_PSIZE_LOG_8;
2703 		else if (maxburst >= 4)
2704 			psize = STEDMA40_PSIZE_LOG_4;
2705 		else
2706 			psize = STEDMA40_PSIZE_LOG_1;
2707 	} else {
2708 		if (maxburst >= 16)
2709 			psize = STEDMA40_PSIZE_PHY_16;
2710 		else if (maxburst >= 8)
2711 			psize = STEDMA40_PSIZE_PHY_8;
2712 		else if (maxburst >= 4)
2713 			psize = STEDMA40_PSIZE_PHY_4;
2714 		else
2715 			psize = STEDMA40_PSIZE_PHY_1;
2716 	}
2717 
2718 	info->psize = psize;
2719 	info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2720 
2721 	return 0;
2722 }
2723 
2724 /* Runtime reconfiguration extension */
2725 static int d40_set_runtime_config(struct dma_chan *chan,
2726 				  struct dma_slave_config *config)
2727 {
2728 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2729 	struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2730 	enum dma_slave_buswidth src_addr_width, dst_addr_width;
2731 	dma_addr_t config_addr;
2732 	u32 src_maxburst, dst_maxburst;
2733 	int ret;
2734 
2735 	src_addr_width = config->src_addr_width;
2736 	src_maxburst = config->src_maxburst;
2737 	dst_addr_width = config->dst_addr_width;
2738 	dst_maxburst = config->dst_maxburst;
2739 
2740 	if (config->direction == DMA_DEV_TO_MEM) {
2741 		config_addr = config->src_addr;
2742 
2743 		if (cfg->dir != DMA_DEV_TO_MEM)
2744 			dev_dbg(d40c->base->dev,
2745 				"channel was not configured for peripheral "
2746 				"to memory transfer (%d) overriding\n",
2747 				cfg->dir);
2748 		cfg->dir = DMA_DEV_TO_MEM;
2749 
2750 		/* Configure the memory side */
2751 		if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2752 			dst_addr_width = src_addr_width;
2753 		if (dst_maxburst == 0)
2754 			dst_maxburst = src_maxburst;
2755 
2756 	} else if (config->direction == DMA_MEM_TO_DEV) {
2757 		config_addr = config->dst_addr;
2758 
2759 		if (cfg->dir != DMA_MEM_TO_DEV)
2760 			dev_dbg(d40c->base->dev,
2761 				"channel was not configured for memory "
2762 				"to peripheral transfer (%d) overriding\n",
2763 				cfg->dir);
2764 		cfg->dir = DMA_MEM_TO_DEV;
2765 
2766 		/* Configure the memory side */
2767 		if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2768 			src_addr_width = dst_addr_width;
2769 		if (src_maxburst == 0)
2770 			src_maxburst = dst_maxburst;
2771 	} else {
2772 		dev_err(d40c->base->dev,
2773 			"unrecognized channel direction %d\n",
2774 			config->direction);
2775 		return -EINVAL;
2776 	}
2777 
2778 	if (config_addr <= 0) {
2779 		dev_err(d40c->base->dev, "no address supplied\n");
2780 		return -EINVAL;
2781 	}
2782 
2783 	if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2784 		dev_err(d40c->base->dev,
2785 			"src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2786 			src_maxburst,
2787 			src_addr_width,
2788 			dst_maxburst,
2789 			dst_addr_width);
2790 		return -EINVAL;
2791 	}
2792 
2793 	if (src_maxburst > 16) {
2794 		src_maxburst = 16;
2795 		dst_maxburst = src_maxburst * src_addr_width / dst_addr_width;
2796 	} else if (dst_maxburst > 16) {
2797 		dst_maxburst = 16;
2798 		src_maxburst = dst_maxburst * dst_addr_width / src_addr_width;
2799 	}
2800 
2801 	/* Only valid widths are; 1, 2, 4 and 8. */
2802 	if (src_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2803 	    src_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2804 	    dst_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2805 	    dst_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2806 	    !is_power_of_2(src_addr_width) ||
2807 	    !is_power_of_2(dst_addr_width))
2808 		return -EINVAL;
2809 
2810 	cfg->src_info.data_width = src_addr_width;
2811 	cfg->dst_info.data_width = dst_addr_width;
2812 
2813 	ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2814 					  src_maxburst);
2815 	if (ret)
2816 		return ret;
2817 
2818 	ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2819 					  dst_maxburst);
2820 	if (ret)
2821 		return ret;
2822 
2823 	/* Fill in register values */
2824 	if (chan_is_logical(d40c))
2825 		d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2826 	else
2827 		d40_phy_cfg(cfg, &d40c->src_def_cfg, &d40c->dst_def_cfg);
2828 
2829 	/* These settings will take precedence later */
2830 	d40c->runtime_addr = config_addr;
2831 	d40c->runtime_direction = config->direction;
2832 	dev_dbg(d40c->base->dev,
2833 		"configured channel %s for %s, data width %d/%d, "
2834 		"maxburst %d/%d elements, LE, no flow control\n",
2835 		dma_chan_name(chan),
2836 		(config->direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2837 		src_addr_width, dst_addr_width,
2838 		src_maxburst, dst_maxburst);
2839 
2840 	return 0;
2841 }
2842 
2843 static int d40_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
2844 		       unsigned long arg)
2845 {
2846 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2847 
2848 	if (d40c->phy_chan == NULL) {
2849 		chan_err(d40c, "Channel is not allocated!\n");
2850 		return -EINVAL;
2851 	}
2852 
2853 	switch (cmd) {
2854 	case DMA_TERMINATE_ALL:
2855 		d40_terminate_all(chan);
2856 		return 0;
2857 	case DMA_PAUSE:
2858 		return d40_pause(d40c);
2859 	case DMA_RESUME:
2860 		return d40_resume(d40c);
2861 	case DMA_SLAVE_CONFIG:
2862 		return d40_set_runtime_config(chan,
2863 			(struct dma_slave_config *) arg);
2864 	default:
2865 		break;
2866 	}
2867 
2868 	/* Other commands are unimplemented */
2869 	return -ENXIO;
2870 }
2871 
2872 /* Initialization functions */
2873 
2874 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2875 				 struct d40_chan *chans, int offset,
2876 				 int num_chans)
2877 {
2878 	int i = 0;
2879 	struct d40_chan *d40c;
2880 
2881 	INIT_LIST_HEAD(&dma->channels);
2882 
2883 	for (i = offset; i < offset + num_chans; i++) {
2884 		d40c = &chans[i];
2885 		d40c->base = base;
2886 		d40c->chan.device = dma;
2887 
2888 		spin_lock_init(&d40c->lock);
2889 
2890 		d40c->log_num = D40_PHY_CHAN;
2891 
2892 		INIT_LIST_HEAD(&d40c->done);
2893 		INIT_LIST_HEAD(&d40c->active);
2894 		INIT_LIST_HEAD(&d40c->queue);
2895 		INIT_LIST_HEAD(&d40c->pending_queue);
2896 		INIT_LIST_HEAD(&d40c->client);
2897 		INIT_LIST_HEAD(&d40c->prepare_queue);
2898 
2899 		tasklet_init(&d40c->tasklet, dma_tasklet,
2900 			     (unsigned long) d40c);
2901 
2902 		list_add_tail(&d40c->chan.device_node,
2903 			      &dma->channels);
2904 	}
2905 }
2906 
2907 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2908 {
2909 	if (dma_has_cap(DMA_SLAVE, dev->cap_mask))
2910 		dev->device_prep_slave_sg = d40_prep_slave_sg;
2911 
2912 	if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2913 		dev->device_prep_dma_memcpy = d40_prep_memcpy;
2914 
2915 		/*
2916 		 * This controller can only access address at even
2917 		 * 32bit boundaries, i.e. 2^2
2918 		 */
2919 		dev->copy_align = 2;
2920 	}
2921 
2922 	if (dma_has_cap(DMA_SG, dev->cap_mask))
2923 		dev->device_prep_dma_sg = d40_prep_memcpy_sg;
2924 
2925 	if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2926 		dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2927 
2928 	dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2929 	dev->device_free_chan_resources = d40_free_chan_resources;
2930 	dev->device_issue_pending = d40_issue_pending;
2931 	dev->device_tx_status = d40_tx_status;
2932 	dev->device_control = d40_control;
2933 	dev->dev = base->dev;
2934 }
2935 
2936 static int __init d40_dmaengine_init(struct d40_base *base,
2937 				     int num_reserved_chans)
2938 {
2939 	int err ;
2940 
2941 	d40_chan_init(base, &base->dma_slave, base->log_chans,
2942 		      0, base->num_log_chans);
2943 
2944 	dma_cap_zero(base->dma_slave.cap_mask);
2945 	dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2946 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2947 
2948 	d40_ops_init(base, &base->dma_slave);
2949 
2950 	err = dma_async_device_register(&base->dma_slave);
2951 
2952 	if (err) {
2953 		d40_err(base->dev, "Failed to register slave channels\n");
2954 		goto failure1;
2955 	}
2956 
2957 	d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2958 		      base->num_log_chans, base->num_memcpy_chans);
2959 
2960 	dma_cap_zero(base->dma_memcpy.cap_mask);
2961 	dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2962 	dma_cap_set(DMA_SG, base->dma_memcpy.cap_mask);
2963 
2964 	d40_ops_init(base, &base->dma_memcpy);
2965 
2966 	err = dma_async_device_register(&base->dma_memcpy);
2967 
2968 	if (err) {
2969 		d40_err(base->dev,
2970 			"Failed to regsiter memcpy only channels\n");
2971 		goto failure2;
2972 	}
2973 
2974 	d40_chan_init(base, &base->dma_both, base->phy_chans,
2975 		      0, num_reserved_chans);
2976 
2977 	dma_cap_zero(base->dma_both.cap_mask);
2978 	dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2979 	dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2980 	dma_cap_set(DMA_SG, base->dma_both.cap_mask);
2981 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2982 
2983 	d40_ops_init(base, &base->dma_both);
2984 	err = dma_async_device_register(&base->dma_both);
2985 
2986 	if (err) {
2987 		d40_err(base->dev,
2988 			"Failed to register logical and physical capable channels\n");
2989 		goto failure3;
2990 	}
2991 	return 0;
2992 failure3:
2993 	dma_async_device_unregister(&base->dma_memcpy);
2994 failure2:
2995 	dma_async_device_unregister(&base->dma_slave);
2996 failure1:
2997 	return err;
2998 }
2999 
3000 /* Suspend resume functionality */
3001 #ifdef CONFIG_PM
3002 static int dma40_pm_suspend(struct device *dev)
3003 {
3004 	struct platform_device *pdev = to_platform_device(dev);
3005 	struct d40_base *base = platform_get_drvdata(pdev);
3006 	int ret = 0;
3007 
3008 	if (base->lcpa_regulator)
3009 		ret = regulator_disable(base->lcpa_regulator);
3010 	return ret;
3011 }
3012 
3013 static int dma40_runtime_suspend(struct device *dev)
3014 {
3015 	struct platform_device *pdev = to_platform_device(dev);
3016 	struct d40_base *base = platform_get_drvdata(pdev);
3017 
3018 	d40_save_restore_registers(base, true);
3019 
3020 	/* Don't disable/enable clocks for v1 due to HW bugs */
3021 	if (base->rev != 1)
3022 		writel_relaxed(base->gcc_pwr_off_mask,
3023 			       base->virtbase + D40_DREG_GCC);
3024 
3025 	return 0;
3026 }
3027 
3028 static int dma40_runtime_resume(struct device *dev)
3029 {
3030 	struct platform_device *pdev = to_platform_device(dev);
3031 	struct d40_base *base = platform_get_drvdata(pdev);
3032 
3033 	if (base->initialized)
3034 		d40_save_restore_registers(base, false);
3035 
3036 	writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
3037 		       base->virtbase + D40_DREG_GCC);
3038 	return 0;
3039 }
3040 
3041 static int dma40_resume(struct device *dev)
3042 {
3043 	struct platform_device *pdev = to_platform_device(dev);
3044 	struct d40_base *base = platform_get_drvdata(pdev);
3045 	int ret = 0;
3046 
3047 	if (base->lcpa_regulator)
3048 		ret = regulator_enable(base->lcpa_regulator);
3049 
3050 	return ret;
3051 }
3052 
3053 static const struct dev_pm_ops dma40_pm_ops = {
3054 	.suspend		= dma40_pm_suspend,
3055 	.runtime_suspend	= dma40_runtime_suspend,
3056 	.runtime_resume		= dma40_runtime_resume,
3057 	.resume			= dma40_resume,
3058 };
3059 #define DMA40_PM_OPS	(&dma40_pm_ops)
3060 #else
3061 #define DMA40_PM_OPS	NULL
3062 #endif
3063 
3064 /* Initialization functions. */
3065 
3066 static int __init d40_phy_res_init(struct d40_base *base)
3067 {
3068 	int i;
3069 	int num_phy_chans_avail = 0;
3070 	u32 val[2];
3071 	int odd_even_bit = -2;
3072 	int gcc = D40_DREG_GCC_ENA;
3073 
3074 	val[0] = readl(base->virtbase + D40_DREG_PRSME);
3075 	val[1] = readl(base->virtbase + D40_DREG_PRSMO);
3076 
3077 	for (i = 0; i < base->num_phy_chans; i++) {
3078 		base->phy_res[i].num = i;
3079 		odd_even_bit += 2 * ((i % 2) == 0);
3080 		if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
3081 			/* Mark security only channels as occupied */
3082 			base->phy_res[i].allocated_src = D40_ALLOC_PHY;
3083 			base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
3084 			base->phy_res[i].reserved = true;
3085 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3086 						       D40_DREG_GCC_SRC);
3087 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3088 						       D40_DREG_GCC_DST);
3089 
3090 
3091 		} else {
3092 			base->phy_res[i].allocated_src = D40_ALLOC_FREE;
3093 			base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
3094 			base->phy_res[i].reserved = false;
3095 			num_phy_chans_avail++;
3096 		}
3097 		spin_lock_init(&base->phy_res[i].lock);
3098 	}
3099 
3100 	/* Mark disabled channels as occupied */
3101 	for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
3102 		int chan = base->plat_data->disabled_channels[i];
3103 
3104 		base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
3105 		base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
3106 		base->phy_res[chan].reserved = true;
3107 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3108 					       D40_DREG_GCC_SRC);
3109 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3110 					       D40_DREG_GCC_DST);
3111 		num_phy_chans_avail--;
3112 	}
3113 
3114 	/* Mark soft_lli channels */
3115 	for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) {
3116 		int chan = base->plat_data->soft_lli_chans[i];
3117 
3118 		base->phy_res[chan].use_soft_lli = true;
3119 	}
3120 
3121 	dev_info(base->dev, "%d of %d physical DMA channels available\n",
3122 		 num_phy_chans_avail, base->num_phy_chans);
3123 
3124 	/* Verify settings extended vs standard */
3125 	val[0] = readl(base->virtbase + D40_DREG_PRTYP);
3126 
3127 	for (i = 0; i < base->num_phy_chans; i++) {
3128 
3129 		if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
3130 		    (val[0] & 0x3) != 1)
3131 			dev_info(base->dev,
3132 				 "[%s] INFO: channel %d is misconfigured (%d)\n",
3133 				 __func__, i, val[0] & 0x3);
3134 
3135 		val[0] = val[0] >> 2;
3136 	}
3137 
3138 	/*
3139 	 * To keep things simple, Enable all clocks initially.
3140 	 * The clocks will get managed later post channel allocation.
3141 	 * The clocks for the event lines on which reserved channels exists
3142 	 * are not managed here.
3143 	 */
3144 	writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3145 	base->gcc_pwr_off_mask = gcc;
3146 
3147 	return num_phy_chans_avail;
3148 }
3149 
3150 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
3151 {
3152 	struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3153 	struct clk *clk = NULL;
3154 	void __iomem *virtbase = NULL;
3155 	struct resource *res = NULL;
3156 	struct d40_base *base = NULL;
3157 	int num_log_chans = 0;
3158 	int num_phy_chans;
3159 	int num_memcpy_chans;
3160 	int clk_ret = -EINVAL;
3161 	int i;
3162 	u32 pid;
3163 	u32 cid;
3164 	u8 rev;
3165 
3166 	clk = clk_get(&pdev->dev, NULL);
3167 	if (IS_ERR(clk)) {
3168 		d40_err(&pdev->dev, "No matching clock found\n");
3169 		goto failure;
3170 	}
3171 
3172 	clk_ret = clk_prepare_enable(clk);
3173 	if (clk_ret) {
3174 		d40_err(&pdev->dev, "Failed to prepare/enable clock\n");
3175 		goto failure;
3176 	}
3177 
3178 	/* Get IO for DMAC base address */
3179 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
3180 	if (!res)
3181 		goto failure;
3182 
3183 	if (request_mem_region(res->start, resource_size(res),
3184 			       D40_NAME " I/O base") == NULL)
3185 		goto failure;
3186 
3187 	virtbase = ioremap(res->start, resource_size(res));
3188 	if (!virtbase)
3189 		goto failure;
3190 
3191 	/* This is just a regular AMBA PrimeCell ID actually */
3192 	for (pid = 0, i = 0; i < 4; i++)
3193 		pid |= (readl(virtbase + resource_size(res) - 0x20 + 4 * i)
3194 			& 255) << (i * 8);
3195 	for (cid = 0, i = 0; i < 4; i++)
3196 		cid |= (readl(virtbase + resource_size(res) - 0x10 + 4 * i)
3197 			& 255) << (i * 8);
3198 
3199 	if (cid != AMBA_CID) {
3200 		d40_err(&pdev->dev, "Unknown hardware! No PrimeCell ID\n");
3201 		goto failure;
3202 	}
3203 	if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
3204 		d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
3205 			AMBA_MANF_BITS(pid),
3206 			AMBA_VENDOR_ST);
3207 		goto failure;
3208 	}
3209 	/*
3210 	 * HW revision:
3211 	 * DB8500ed has revision 0
3212 	 * ? has revision 1
3213 	 * DB8500v1 has revision 2
3214 	 * DB8500v2 has revision 3
3215 	 * AP9540v1 has revision 4
3216 	 * DB8540v1 has revision 4
3217 	 */
3218 	rev = AMBA_REV_BITS(pid);
3219 	if (rev < 2) {
3220 		d40_err(&pdev->dev, "hardware revision: %d is not supported", rev);
3221 		goto failure;
3222 	}
3223 
3224 	/* The number of physical channels on this HW */
3225 	if (plat_data->num_of_phy_chans)
3226 		num_phy_chans = plat_data->num_of_phy_chans;
3227 	else
3228 		num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
3229 
3230 	/* The number of channels used for memcpy */
3231 	if (plat_data->num_of_memcpy_chans)
3232 		num_memcpy_chans = plat_data->num_of_memcpy_chans;
3233 	else
3234 		num_memcpy_chans = ARRAY_SIZE(dma40_memcpy_channels);
3235 
3236 	num_log_chans = num_phy_chans * D40_MAX_LOG_CHAN_PER_PHY;
3237 
3238 	dev_info(&pdev->dev,
3239 		 "hardware rev: %d @ %pa with %d physical and %d logical channels\n",
3240 		 rev, &res->start, num_phy_chans, num_log_chans);
3241 
3242 	base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
3243 		       (num_phy_chans + num_log_chans + num_memcpy_chans) *
3244 		       sizeof(struct d40_chan), GFP_KERNEL);
3245 
3246 	if (base == NULL) {
3247 		d40_err(&pdev->dev, "Out of memory\n");
3248 		goto failure;
3249 	}
3250 
3251 	base->rev = rev;
3252 	base->clk = clk;
3253 	base->num_memcpy_chans = num_memcpy_chans;
3254 	base->num_phy_chans = num_phy_chans;
3255 	base->num_log_chans = num_log_chans;
3256 	base->phy_start = res->start;
3257 	base->phy_size = resource_size(res);
3258 	base->virtbase = virtbase;
3259 	base->plat_data = plat_data;
3260 	base->dev = &pdev->dev;
3261 	base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3262 	base->log_chans = &base->phy_chans[num_phy_chans];
3263 
3264 	if (base->plat_data->num_of_phy_chans == 14) {
3265 		base->gen_dmac.backup = d40_backup_regs_v4b;
3266 		base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
3267 		base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
3268 		base->gen_dmac.interrupt_clear = D40_DREG_CPCICR;
3269 		base->gen_dmac.realtime_en = D40_DREG_CRSEG1;
3270 		base->gen_dmac.realtime_clear = D40_DREG_CRCEG1;
3271 		base->gen_dmac.high_prio_en = D40_DREG_CPSEG1;
3272 		base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1;
3273 		base->gen_dmac.il = il_v4b;
3274 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4b);
3275 		base->gen_dmac.init_reg = dma_init_reg_v4b;
3276 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
3277 	} else {
3278 		if (base->rev >= 3) {
3279 			base->gen_dmac.backup = d40_backup_regs_v4a;
3280 			base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
3281 		}
3282 		base->gen_dmac.interrupt_en = D40_DREG_PCMIS;
3283 		base->gen_dmac.interrupt_clear = D40_DREG_PCICR;
3284 		base->gen_dmac.realtime_en = D40_DREG_RSEG1;
3285 		base->gen_dmac.realtime_clear = D40_DREG_RCEG1;
3286 		base->gen_dmac.high_prio_en = D40_DREG_PSEG1;
3287 		base->gen_dmac.high_prio_clear = D40_DREG_PCEG1;
3288 		base->gen_dmac.il = il_v4a;
3289 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4a);
3290 		base->gen_dmac.init_reg = dma_init_reg_v4a;
3291 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a);
3292 	}
3293 
3294 	base->phy_res = kzalloc(num_phy_chans * sizeof(struct d40_phy_res),
3295 				GFP_KERNEL);
3296 	if (!base->phy_res)
3297 		goto failure;
3298 
3299 	base->lookup_phy_chans = kzalloc(num_phy_chans *
3300 					 sizeof(struct d40_chan *),
3301 					 GFP_KERNEL);
3302 	if (!base->lookup_phy_chans)
3303 		goto failure;
3304 
3305 	base->lookup_log_chans = kzalloc(num_log_chans *
3306 					 sizeof(struct d40_chan *),
3307 					 GFP_KERNEL);
3308 	if (!base->lookup_log_chans)
3309 		goto failure;
3310 
3311 	base->reg_val_backup_chan = kmalloc(base->num_phy_chans *
3312 					    sizeof(d40_backup_regs_chan),
3313 					    GFP_KERNEL);
3314 	if (!base->reg_val_backup_chan)
3315 		goto failure;
3316 
3317 	base->lcla_pool.alloc_map =
3318 		kzalloc(num_phy_chans * sizeof(struct d40_desc *)
3319 			* D40_LCLA_LINK_PER_EVENT_GRP, GFP_KERNEL);
3320 	if (!base->lcla_pool.alloc_map)
3321 		goto failure;
3322 
3323 	base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3324 					    0, SLAB_HWCACHE_ALIGN,
3325 					    NULL);
3326 	if (base->desc_slab == NULL)
3327 		goto failure;
3328 
3329 	return base;
3330 
3331 failure:
3332 	if (!clk_ret)
3333 		clk_disable_unprepare(clk);
3334 	if (!IS_ERR(clk))
3335 		clk_put(clk);
3336 	if (virtbase)
3337 		iounmap(virtbase);
3338 	if (res)
3339 		release_mem_region(res->start,
3340 				   resource_size(res));
3341 	if (virtbase)
3342 		iounmap(virtbase);
3343 
3344 	if (base) {
3345 		kfree(base->lcla_pool.alloc_map);
3346 		kfree(base->reg_val_backup_chan);
3347 		kfree(base->lookup_log_chans);
3348 		kfree(base->lookup_phy_chans);
3349 		kfree(base->phy_res);
3350 		kfree(base);
3351 	}
3352 
3353 	return NULL;
3354 }
3355 
3356 static void __init d40_hw_init(struct d40_base *base)
3357 {
3358 
3359 	int i;
3360 	u32 prmseo[2] = {0, 0};
3361 	u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3362 	u32 pcmis = 0;
3363 	u32 pcicr = 0;
3364 	struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg;
3365 	u32 reg_size = base->gen_dmac.init_reg_size;
3366 
3367 	for (i = 0; i < reg_size; i++)
3368 		writel(dma_init_reg[i].val,
3369 		       base->virtbase + dma_init_reg[i].reg);
3370 
3371 	/* Configure all our dma channels to default settings */
3372 	for (i = 0; i < base->num_phy_chans; i++) {
3373 
3374 		activeo[i % 2] = activeo[i % 2] << 2;
3375 
3376 		if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3377 		    == D40_ALLOC_PHY) {
3378 			activeo[i % 2] |= 3;
3379 			continue;
3380 		}
3381 
3382 		/* Enable interrupt # */
3383 		pcmis = (pcmis << 1) | 1;
3384 
3385 		/* Clear interrupt # */
3386 		pcicr = (pcicr << 1) | 1;
3387 
3388 		/* Set channel to physical mode */
3389 		prmseo[i % 2] = prmseo[i % 2] << 2;
3390 		prmseo[i % 2] |= 1;
3391 
3392 	}
3393 
3394 	writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3395 	writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3396 	writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3397 	writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3398 
3399 	/* Write which interrupt to enable */
3400 	writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en);
3401 
3402 	/* Write which interrupt to clear */
3403 	writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear);
3404 
3405 	/* These are __initdata and cannot be accessed after init */
3406 	base->gen_dmac.init_reg = NULL;
3407 	base->gen_dmac.init_reg_size = 0;
3408 }
3409 
3410 static int __init d40_lcla_allocate(struct d40_base *base)
3411 {
3412 	struct d40_lcla_pool *pool = &base->lcla_pool;
3413 	unsigned long *page_list;
3414 	int i, j;
3415 	int ret = 0;
3416 
3417 	/*
3418 	 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3419 	 * To full fill this hardware requirement without wasting 256 kb
3420 	 * we allocate pages until we get an aligned one.
3421 	 */
3422 	page_list = kmalloc(sizeof(unsigned long) * MAX_LCLA_ALLOC_ATTEMPTS,
3423 			    GFP_KERNEL);
3424 
3425 	if (!page_list) {
3426 		ret = -ENOMEM;
3427 		goto failure;
3428 	}
3429 
3430 	/* Calculating how many pages that are required */
3431 	base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3432 
3433 	for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3434 		page_list[i] = __get_free_pages(GFP_KERNEL,
3435 						base->lcla_pool.pages);
3436 		if (!page_list[i]) {
3437 
3438 			d40_err(base->dev, "Failed to allocate %d pages.\n",
3439 				base->lcla_pool.pages);
3440 
3441 			for (j = 0; j < i; j++)
3442 				free_pages(page_list[j], base->lcla_pool.pages);
3443 			goto failure;
3444 		}
3445 
3446 		if ((virt_to_phys((void *)page_list[i]) &
3447 		     (LCLA_ALIGNMENT - 1)) == 0)
3448 			break;
3449 	}
3450 
3451 	for (j = 0; j < i; j++)
3452 		free_pages(page_list[j], base->lcla_pool.pages);
3453 
3454 	if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3455 		base->lcla_pool.base = (void *)page_list[i];
3456 	} else {
3457 		/*
3458 		 * After many attempts and no succees with finding the correct
3459 		 * alignment, try with allocating a big buffer.
3460 		 */
3461 		dev_warn(base->dev,
3462 			 "[%s] Failed to get %d pages @ 18 bit align.\n",
3463 			 __func__, base->lcla_pool.pages);
3464 		base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3465 							 base->num_phy_chans +
3466 							 LCLA_ALIGNMENT,
3467 							 GFP_KERNEL);
3468 		if (!base->lcla_pool.base_unaligned) {
3469 			ret = -ENOMEM;
3470 			goto failure;
3471 		}
3472 
3473 		base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3474 						 LCLA_ALIGNMENT);
3475 	}
3476 
3477 	pool->dma_addr = dma_map_single(base->dev, pool->base,
3478 					SZ_1K * base->num_phy_chans,
3479 					DMA_TO_DEVICE);
3480 	if (dma_mapping_error(base->dev, pool->dma_addr)) {
3481 		pool->dma_addr = 0;
3482 		ret = -ENOMEM;
3483 		goto failure;
3484 	}
3485 
3486 	writel(virt_to_phys(base->lcla_pool.base),
3487 	       base->virtbase + D40_DREG_LCLA);
3488 failure:
3489 	kfree(page_list);
3490 	return ret;
3491 }
3492 
3493 static int __init d40_of_probe(struct platform_device *pdev,
3494 			       struct device_node *np)
3495 {
3496 	struct stedma40_platform_data *pdata;
3497 	int num_phy = 0, num_memcpy = 0, num_disabled = 0;
3498 	const __be32 *list;
3499 
3500 	pdata = devm_kzalloc(&pdev->dev,
3501 			     sizeof(struct stedma40_platform_data),
3502 			     GFP_KERNEL);
3503 	if (!pdata)
3504 		return -ENOMEM;
3505 
3506 	/* If absent this value will be obtained from h/w. */
3507 	of_property_read_u32(np, "dma-channels", &num_phy);
3508 	if (num_phy > 0)
3509 		pdata->num_of_phy_chans = num_phy;
3510 
3511 	list = of_get_property(np, "memcpy-channels", &num_memcpy);
3512 	num_memcpy /= sizeof(*list);
3513 
3514 	if (num_memcpy > D40_MEMCPY_MAX_CHANS || num_memcpy <= 0) {
3515 		d40_err(&pdev->dev,
3516 			"Invalid number of memcpy channels specified (%d)\n",
3517 			num_memcpy);
3518 		return -EINVAL;
3519 	}
3520 	pdata->num_of_memcpy_chans = num_memcpy;
3521 
3522 	of_property_read_u32_array(np, "memcpy-channels",
3523 				   dma40_memcpy_channels,
3524 				   num_memcpy);
3525 
3526 	list = of_get_property(np, "disabled-channels", &num_disabled);
3527 	num_disabled /= sizeof(*list);
3528 
3529 	if (num_disabled >= STEDMA40_MAX_PHYS || num_disabled < 0) {
3530 		d40_err(&pdev->dev,
3531 			"Invalid number of disabled channels specified (%d)\n",
3532 			num_disabled);
3533 		return -EINVAL;
3534 	}
3535 
3536 	of_property_read_u32_array(np, "disabled-channels",
3537 				   pdata->disabled_channels,
3538 				   num_disabled);
3539 	pdata->disabled_channels[num_disabled] = -1;
3540 
3541 	pdev->dev.platform_data = pdata;
3542 
3543 	return 0;
3544 }
3545 
3546 static int __init d40_probe(struct platform_device *pdev)
3547 {
3548 	struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3549 	struct device_node *np = pdev->dev.of_node;
3550 	int ret = -ENOENT;
3551 	struct d40_base *base = NULL;
3552 	struct resource *res = NULL;
3553 	int num_reserved_chans;
3554 	u32 val;
3555 
3556 	if (!plat_data) {
3557 		if (np) {
3558 			if(d40_of_probe(pdev, np)) {
3559 				ret = -ENOMEM;
3560 				goto failure;
3561 			}
3562 		} else {
3563 			d40_err(&pdev->dev, "No pdata or Device Tree provided\n");
3564 			goto failure;
3565 		}
3566 	}
3567 
3568 	base = d40_hw_detect_init(pdev);
3569 	if (!base)
3570 		goto failure;
3571 
3572 	num_reserved_chans = d40_phy_res_init(base);
3573 
3574 	platform_set_drvdata(pdev, base);
3575 
3576 	spin_lock_init(&base->interrupt_lock);
3577 	spin_lock_init(&base->execmd_lock);
3578 
3579 	/* Get IO for logical channel parameter address */
3580 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
3581 	if (!res) {
3582 		ret = -ENOENT;
3583 		d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
3584 		goto failure;
3585 	}
3586 	base->lcpa_size = resource_size(res);
3587 	base->phy_lcpa = res->start;
3588 
3589 	if (request_mem_region(res->start, resource_size(res),
3590 			       D40_NAME " I/O lcpa") == NULL) {
3591 		ret = -EBUSY;
3592 		d40_err(&pdev->dev, "Failed to request LCPA region %pR\n", res);
3593 		goto failure;
3594 	}
3595 
3596 	/* We make use of ESRAM memory for this. */
3597 	val = readl(base->virtbase + D40_DREG_LCPA);
3598 	if (res->start != val && val != 0) {
3599 		dev_warn(&pdev->dev,
3600 			 "[%s] Mismatch LCPA dma 0x%x, def %pa\n",
3601 			 __func__, val, &res->start);
3602 	} else
3603 		writel(res->start, base->virtbase + D40_DREG_LCPA);
3604 
3605 	base->lcpa_base = ioremap(res->start, resource_size(res));
3606 	if (!base->lcpa_base) {
3607 		ret = -ENOMEM;
3608 		d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
3609 		goto failure;
3610 	}
3611 	/* If lcla has to be located in ESRAM we don't need to allocate */
3612 	if (base->plat_data->use_esram_lcla) {
3613 		res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3614 							"lcla_esram");
3615 		if (!res) {
3616 			ret = -ENOENT;
3617 			d40_err(&pdev->dev,
3618 				"No \"lcla_esram\" memory resource\n");
3619 			goto failure;
3620 		}
3621 		base->lcla_pool.base = ioremap(res->start,
3622 						resource_size(res));
3623 		if (!base->lcla_pool.base) {
3624 			ret = -ENOMEM;
3625 			d40_err(&pdev->dev, "Failed to ioremap LCLA region\n");
3626 			goto failure;
3627 		}
3628 		writel(res->start, base->virtbase + D40_DREG_LCLA);
3629 
3630 	} else {
3631 		ret = d40_lcla_allocate(base);
3632 		if (ret) {
3633 			d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
3634 			goto failure;
3635 		}
3636 	}
3637 
3638 	spin_lock_init(&base->lcla_pool.lock);
3639 
3640 	base->irq = platform_get_irq(pdev, 0);
3641 
3642 	ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3643 	if (ret) {
3644 		d40_err(&pdev->dev, "No IRQ defined\n");
3645 		goto failure;
3646 	}
3647 
3648 	pm_runtime_irq_safe(base->dev);
3649 	pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3650 	pm_runtime_use_autosuspend(base->dev);
3651 	pm_runtime_enable(base->dev);
3652 	pm_runtime_resume(base->dev);
3653 
3654 	if (base->plat_data->use_esram_lcla) {
3655 
3656 		base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3657 		if (IS_ERR(base->lcpa_regulator)) {
3658 			d40_err(&pdev->dev, "Failed to get lcpa_regulator\n");
3659 			ret = PTR_ERR(base->lcpa_regulator);
3660 			base->lcpa_regulator = NULL;
3661 			goto failure;
3662 		}
3663 
3664 		ret = regulator_enable(base->lcpa_regulator);
3665 		if (ret) {
3666 			d40_err(&pdev->dev,
3667 				"Failed to enable lcpa_regulator\n");
3668 			regulator_put(base->lcpa_regulator);
3669 			base->lcpa_regulator = NULL;
3670 			goto failure;
3671 		}
3672 	}
3673 
3674 	base->initialized = true;
3675 	ret = d40_dmaengine_init(base, num_reserved_chans);
3676 	if (ret)
3677 		goto failure;
3678 
3679 	base->dev->dma_parms = &base->dma_parms;
3680 	ret = dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
3681 	if (ret) {
3682 		d40_err(&pdev->dev, "Failed to set dma max seg size\n");
3683 		goto failure;
3684 	}
3685 
3686 	d40_hw_init(base);
3687 
3688 	if (np) {
3689 		ret = of_dma_controller_register(np, d40_xlate, NULL);
3690 		if (ret)
3691 			dev_err(&pdev->dev,
3692 				"could not register of_dma_controller\n");
3693 	}
3694 
3695 	dev_info(base->dev, "initialized\n");
3696 	return 0;
3697 
3698 failure:
3699 	if (base) {
3700 		if (base->desc_slab)
3701 			kmem_cache_destroy(base->desc_slab);
3702 		if (base->virtbase)
3703 			iounmap(base->virtbase);
3704 
3705 		if (base->lcla_pool.base && base->plat_data->use_esram_lcla) {
3706 			iounmap(base->lcla_pool.base);
3707 			base->lcla_pool.base = NULL;
3708 		}
3709 
3710 		if (base->lcla_pool.dma_addr)
3711 			dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3712 					 SZ_1K * base->num_phy_chans,
3713 					 DMA_TO_DEVICE);
3714 
3715 		if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3716 			free_pages((unsigned long)base->lcla_pool.base,
3717 				   base->lcla_pool.pages);
3718 
3719 		kfree(base->lcla_pool.base_unaligned);
3720 
3721 		if (base->phy_lcpa)
3722 			release_mem_region(base->phy_lcpa,
3723 					   base->lcpa_size);
3724 		if (base->phy_start)
3725 			release_mem_region(base->phy_start,
3726 					   base->phy_size);
3727 		if (base->clk) {
3728 			clk_disable_unprepare(base->clk);
3729 			clk_put(base->clk);
3730 		}
3731 
3732 		if (base->lcpa_regulator) {
3733 			regulator_disable(base->lcpa_regulator);
3734 			regulator_put(base->lcpa_regulator);
3735 		}
3736 
3737 		kfree(base->lcla_pool.alloc_map);
3738 		kfree(base->lookup_log_chans);
3739 		kfree(base->lookup_phy_chans);
3740 		kfree(base->phy_res);
3741 		kfree(base);
3742 	}
3743 
3744 	d40_err(&pdev->dev, "probe failed\n");
3745 	return ret;
3746 }
3747 
3748 static const struct of_device_id d40_match[] = {
3749         { .compatible = "stericsson,dma40", },
3750         {}
3751 };
3752 
3753 static struct platform_driver d40_driver = {
3754 	.driver = {
3755 		.owner = THIS_MODULE,
3756 		.name  = D40_NAME,
3757 		.pm = DMA40_PM_OPS,
3758 		.of_match_table = d40_match,
3759 	},
3760 };
3761 
3762 static int __init stedma40_init(void)
3763 {
3764 	return platform_driver_probe(&d40_driver, d40_probe);
3765 }
3766 subsys_initcall(stedma40_init);
3767