xref: /openbmc/linux/drivers/dma/ste_dma40.c (revision 11788d9b)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) Ericsson AB 2007-2008
4  * Copyright (C) ST-Ericsson SA 2008-2010
5  * Author: Per Forlin <per.forlin@stericsson.com> for ST-Ericsson
6  * Author: Jonas Aaberg <jonas.aberg@stericsson.com> for ST-Ericsson
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 const 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 const 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 __maybe_unused 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 __maybe_unused 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  * @dma_addr: DMA address, if mapped
385  * @base_unaligned: The orignal kmalloc pointer, if kmalloc is used.
386  * This pointer is only there for clean-up on error.
387  * @pages: The number of pages needed for all physical channels.
388  * Only used later for clean-up on error
389  * @lock: Lock to protect the content in this struct.
390  * @alloc_map: big map over which LCLA entry is own by which job.
391  */
392 struct d40_lcla_pool {
393 	void		*base;
394 	dma_addr_t	dma_addr;
395 	void		*base_unaligned;
396 	int		 pages;
397 	spinlock_t	 lock;
398 	struct d40_desc	**alloc_map;
399 };
400 
401 /**
402  * struct d40_phy_res - struct for handling eventlines mapped to physical
403  * channels.
404  *
405  * @lock: A lock protection this entity.
406  * @reserved: True if used by secure world or otherwise.
407  * @num: The physical channel number of this entity.
408  * @allocated_src: Bit mapped to show which src event line's are mapped to
409  * this physical channel. Can also be free or physically allocated.
410  * @allocated_dst: Same as for src but is dst.
411  * allocated_dst and allocated_src uses the D40_ALLOC* defines as well as
412  * event line number.
413  * @use_soft_lli: To mark if the linked lists of channel are managed by SW.
414  */
415 struct d40_phy_res {
416 	spinlock_t lock;
417 	bool	   reserved;
418 	int	   num;
419 	u32	   allocated_src;
420 	u32	   allocated_dst;
421 	bool	   use_soft_lli;
422 };
423 
424 struct d40_base;
425 
426 /**
427  * struct d40_chan - Struct that describes a channel.
428  *
429  * @lock: A spinlock to protect this struct.
430  * @log_num: The logical number, if any of this channel.
431  * @pending_tx: The number of pending transfers. Used between interrupt handler
432  * and tasklet.
433  * @busy: Set to true when transfer is ongoing on this channel.
434  * @phy_chan: Pointer to physical channel which this instance runs on. If this
435  * point is NULL, then the channel is not allocated.
436  * @chan: DMA engine handle.
437  * @tasklet: Tasklet that gets scheduled from interrupt context to complete a
438  * transfer and call client callback.
439  * @client: Cliented owned descriptor list.
440  * @pending_queue: Submitted jobs, to be issued by issue_pending()
441  * @active: Active descriptor.
442  * @done: Completed jobs
443  * @queue: Queued jobs.
444  * @prepare_queue: Prepared jobs.
445  * @dma_cfg: The client configuration of this dma channel.
446  * @slave_config: DMA slave configuration.
447  * @configured: whether the dma_cfg configuration is valid
448  * @base: Pointer to the device instance struct.
449  * @src_def_cfg: Default cfg register setting for src.
450  * @dst_def_cfg: Default cfg register setting for dst.
451  * @log_def: Default logical channel settings.
452  * @lcpa: Pointer to dst and src lcpa settings.
453  * @runtime_addr: runtime configured address.
454  * @runtime_direction: runtime configured direction.
455  *
456  * This struct can either "be" a logical or a physical channel.
457  */
458 struct d40_chan {
459 	spinlock_t			 lock;
460 	int				 log_num;
461 	int				 pending_tx;
462 	bool				 busy;
463 	struct d40_phy_res		*phy_chan;
464 	struct dma_chan			 chan;
465 	struct tasklet_struct		 tasklet;
466 	struct list_head		 client;
467 	struct list_head		 pending_queue;
468 	struct list_head		 active;
469 	struct list_head		 done;
470 	struct list_head		 queue;
471 	struct list_head		 prepare_queue;
472 	struct stedma40_chan_cfg	 dma_cfg;
473 	struct dma_slave_config		 slave_config;
474 	bool				 configured;
475 	struct d40_base			*base;
476 	/* Default register configurations */
477 	u32				 src_def_cfg;
478 	u32				 dst_def_cfg;
479 	struct d40_def_lcsp		 log_def;
480 	struct d40_log_lli_full		*lcpa;
481 	/* Runtime reconfiguration */
482 	dma_addr_t			runtime_addr;
483 	enum dma_transfer_direction	runtime_direction;
484 };
485 
486 /**
487  * struct d40_gen_dmac - generic values to represent u8500/u8540 DMA
488  * controller
489  *
490  * @backup: the pointer to the registers address array for backup
491  * @backup_size: the size of the registers address array for backup
492  * @realtime_en: the realtime enable register
493  * @realtime_clear: the realtime clear register
494  * @high_prio_en: the high priority enable register
495  * @high_prio_clear: the high priority clear register
496  * @interrupt_en: the interrupt enable register
497  * @interrupt_clear: the interrupt clear register
498  * @il: the pointer to struct d40_interrupt_lookup
499  * @il_size: the size of d40_interrupt_lookup array
500  * @init_reg: the pointer to the struct d40_reg_val
501  * @init_reg_size: the size of d40_reg_val array
502  */
503 struct d40_gen_dmac {
504 	u32				*backup;
505 	u32				 backup_size;
506 	u32				 realtime_en;
507 	u32				 realtime_clear;
508 	u32				 high_prio_en;
509 	u32				 high_prio_clear;
510 	u32				 interrupt_en;
511 	u32				 interrupt_clear;
512 	struct d40_interrupt_lookup	*il;
513 	u32				 il_size;
514 	struct d40_reg_val		*init_reg;
515 	u32				 init_reg_size;
516 };
517 
518 /**
519  * struct d40_base - The big global struct, one for each probe'd instance.
520  *
521  * @interrupt_lock: Lock used to make sure one interrupt is handle a time.
522  * @execmd_lock: Lock for execute command usage since several channels share
523  * the same physical register.
524  * @dev: The device structure.
525  * @virtbase: The virtual base address of the DMA's register.
526  * @rev: silicon revision detected.
527  * @clk: Pointer to the DMA clock structure.
528  * @phy_start: Physical memory start of the DMA registers.
529  * @phy_size: Size of the DMA register map.
530  * @irq: The IRQ number.
531  * @num_memcpy_chans: The number of channels used for memcpy (mem-to-mem
532  * transfers).
533  * @num_phy_chans: The number of physical channels. Read from HW. This
534  * is the number of available channels for this driver, not counting "Secure
535  * mode" allocated physical channels.
536  * @num_log_chans: The number of logical channels. Calculated from
537  * num_phy_chans.
538  * @dma_parms: DMA parameters for the channel
539  * @dma_both: dma_device channels that can do both memcpy and slave transfers.
540  * @dma_slave: dma_device channels that can do only do slave transfers.
541  * @dma_memcpy: dma_device channels that can do only do memcpy transfers.
542  * @phy_chans: Room for all possible physical channels in system.
543  * @log_chans: Room for all possible logical channels in system.
544  * @lookup_log_chans: Used to map interrupt number to logical channel. Points
545  * to log_chans entries.
546  * @lookup_phy_chans: Used to map interrupt number to physical channel. Points
547  * to phy_chans entries.
548  * @plat_data: Pointer to provided platform_data which is the driver
549  * configuration.
550  * @lcpa_regulator: Pointer to hold the regulator for the esram bank for lcla.
551  * @phy_res: Vector containing all physical channels.
552  * @lcla_pool: lcla pool settings and data.
553  * @lcpa_base: The virtual mapped address of LCPA.
554  * @phy_lcpa: The physical address of the LCPA.
555  * @lcpa_size: The size of the LCPA area.
556  * @desc_slab: cache for descriptors.
557  * @reg_val_backup: Here the values of some hardware registers are stored
558  * before the DMA is powered off. They are restored when the power is back on.
559  * @reg_val_backup_v4: Backup of registers that only exits on dma40 v3 and
560  * later
561  * @reg_val_backup_chan: Backup data for standard channel parameter registers.
562  * @regs_interrupt: Scratch space for registers during interrupt.
563  * @gcc_pwr_off_mask: Mask to maintain the channels that can be turned off.
564  * @gen_dmac: the struct for generic registers values to represent u8500/8540
565  * DMA controller
566  */
567 struct d40_base {
568 	spinlock_t			 interrupt_lock;
569 	spinlock_t			 execmd_lock;
570 	struct device			 *dev;
571 	void __iomem			 *virtbase;
572 	u8				  rev:4;
573 	struct clk			 *clk;
574 	phys_addr_t			  phy_start;
575 	resource_size_t			  phy_size;
576 	int				  irq;
577 	int				  num_memcpy_chans;
578 	int				  num_phy_chans;
579 	int				  num_log_chans;
580 	struct device_dma_parameters	  dma_parms;
581 	struct dma_device		  dma_both;
582 	struct dma_device		  dma_slave;
583 	struct dma_device		  dma_memcpy;
584 	struct d40_chan			 *phy_chans;
585 	struct d40_chan			 *log_chans;
586 	struct d40_chan			**lookup_log_chans;
587 	struct d40_chan			**lookup_phy_chans;
588 	struct stedma40_platform_data	 *plat_data;
589 	struct regulator		 *lcpa_regulator;
590 	/* Physical half channels */
591 	struct d40_phy_res		 *phy_res;
592 	struct d40_lcla_pool		  lcla_pool;
593 	void				 *lcpa_base;
594 	dma_addr_t			  phy_lcpa;
595 	resource_size_t			  lcpa_size;
596 	struct kmem_cache		 *desc_slab;
597 	u32				  reg_val_backup[BACKUP_REGS_SZ];
598 	u32				  reg_val_backup_v4[BACKUP_REGS_SZ_MAX];
599 	u32				 *reg_val_backup_chan;
600 	u32				 *regs_interrupt;
601 	u16				  gcc_pwr_off_mask;
602 	struct d40_gen_dmac		  gen_dmac;
603 };
604 
605 static struct device *chan2dev(struct d40_chan *d40c)
606 {
607 	return &d40c->chan.dev->device;
608 }
609 
610 static bool chan_is_physical(struct d40_chan *chan)
611 {
612 	return chan->log_num == D40_PHY_CHAN;
613 }
614 
615 static bool chan_is_logical(struct d40_chan *chan)
616 {
617 	return !chan_is_physical(chan);
618 }
619 
620 static void __iomem *chan_base(struct d40_chan *chan)
621 {
622 	return chan->base->virtbase + D40_DREG_PCBASE +
623 	       chan->phy_chan->num * D40_DREG_PCDELTA;
624 }
625 
626 #define d40_err(dev, format, arg...)		\
627 	dev_err(dev, "[%s] " format, __func__, ## arg)
628 
629 #define chan_err(d40c, format, arg...)		\
630 	d40_err(chan2dev(d40c), format, ## arg)
631 
632 static int d40_set_runtime_config_write(struct dma_chan *chan,
633 				  struct dma_slave_config *config,
634 				  enum dma_transfer_direction direction);
635 
636 static int d40_pool_lli_alloc(struct d40_chan *d40c, struct d40_desc *d40d,
637 			      int lli_len)
638 {
639 	bool is_log = chan_is_logical(d40c);
640 	u32 align;
641 	void *base;
642 
643 	if (is_log)
644 		align = sizeof(struct d40_log_lli);
645 	else
646 		align = sizeof(struct d40_phy_lli);
647 
648 	if (lli_len == 1) {
649 		base = d40d->lli_pool.pre_alloc_lli;
650 		d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
651 		d40d->lli_pool.base = NULL;
652 	} else {
653 		d40d->lli_pool.size = lli_len * 2 * align;
654 
655 		base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
656 		d40d->lli_pool.base = base;
657 
658 		if (d40d->lli_pool.base == NULL)
659 			return -ENOMEM;
660 	}
661 
662 	if (is_log) {
663 		d40d->lli_log.src = PTR_ALIGN(base, align);
664 		d40d->lli_log.dst = d40d->lli_log.src + lli_len;
665 
666 		d40d->lli_pool.dma_addr = 0;
667 	} else {
668 		d40d->lli_phy.src = PTR_ALIGN(base, align);
669 		d40d->lli_phy.dst = d40d->lli_phy.src + lli_len;
670 
671 		d40d->lli_pool.dma_addr = dma_map_single(d40c->base->dev,
672 							 d40d->lli_phy.src,
673 							 d40d->lli_pool.size,
674 							 DMA_TO_DEVICE);
675 
676 		if (dma_mapping_error(d40c->base->dev,
677 				      d40d->lli_pool.dma_addr)) {
678 			kfree(d40d->lli_pool.base);
679 			d40d->lli_pool.base = NULL;
680 			d40d->lli_pool.dma_addr = 0;
681 			return -ENOMEM;
682 		}
683 	}
684 
685 	return 0;
686 }
687 
688 static void d40_pool_lli_free(struct d40_chan *d40c, struct d40_desc *d40d)
689 {
690 	if (d40d->lli_pool.dma_addr)
691 		dma_unmap_single(d40c->base->dev, d40d->lli_pool.dma_addr,
692 				 d40d->lli_pool.size, DMA_TO_DEVICE);
693 
694 	kfree(d40d->lli_pool.base);
695 	d40d->lli_pool.base = NULL;
696 	d40d->lli_pool.size = 0;
697 	d40d->lli_log.src = NULL;
698 	d40d->lli_log.dst = NULL;
699 	d40d->lli_phy.src = NULL;
700 	d40d->lli_phy.dst = NULL;
701 }
702 
703 static int d40_lcla_alloc_one(struct d40_chan *d40c,
704 			      struct d40_desc *d40d)
705 {
706 	unsigned long flags;
707 	int i;
708 	int ret = -EINVAL;
709 
710 	spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
711 
712 	/*
713 	 * Allocate both src and dst at the same time, therefore the half
714 	 * start on 1 since 0 can't be used since zero is used as end marker.
715 	 */
716 	for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
717 		int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
718 
719 		if (!d40c->base->lcla_pool.alloc_map[idx]) {
720 			d40c->base->lcla_pool.alloc_map[idx] = d40d;
721 			d40d->lcla_alloc++;
722 			ret = i;
723 			break;
724 		}
725 	}
726 
727 	spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
728 
729 	return ret;
730 }
731 
732 static int d40_lcla_free_all(struct d40_chan *d40c,
733 			     struct d40_desc *d40d)
734 {
735 	unsigned long flags;
736 	int i;
737 	int ret = -EINVAL;
738 
739 	if (chan_is_physical(d40c))
740 		return 0;
741 
742 	spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
743 
744 	for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
745 		int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
746 
747 		if (d40c->base->lcla_pool.alloc_map[idx] == d40d) {
748 			d40c->base->lcla_pool.alloc_map[idx] = NULL;
749 			d40d->lcla_alloc--;
750 			if (d40d->lcla_alloc == 0) {
751 				ret = 0;
752 				break;
753 			}
754 		}
755 	}
756 
757 	spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
758 
759 	return ret;
760 
761 }
762 
763 static void d40_desc_remove(struct d40_desc *d40d)
764 {
765 	list_del(&d40d->node);
766 }
767 
768 static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
769 {
770 	struct d40_desc *desc = NULL;
771 
772 	if (!list_empty(&d40c->client)) {
773 		struct d40_desc *d;
774 		struct d40_desc *_d;
775 
776 		list_for_each_entry_safe(d, _d, &d40c->client, node) {
777 			if (async_tx_test_ack(&d->txd)) {
778 				d40_desc_remove(d);
779 				desc = d;
780 				memset(desc, 0, sizeof(*desc));
781 				break;
782 			}
783 		}
784 	}
785 
786 	if (!desc)
787 		desc = kmem_cache_zalloc(d40c->base->desc_slab, GFP_NOWAIT);
788 
789 	if (desc)
790 		INIT_LIST_HEAD(&desc->node);
791 
792 	return desc;
793 }
794 
795 static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
796 {
797 
798 	d40_pool_lli_free(d40c, d40d);
799 	d40_lcla_free_all(d40c, d40d);
800 	kmem_cache_free(d40c->base->desc_slab, d40d);
801 }
802 
803 static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
804 {
805 	list_add_tail(&desc->node, &d40c->active);
806 }
807 
808 static void d40_phy_lli_load(struct d40_chan *chan, struct d40_desc *desc)
809 {
810 	struct d40_phy_lli *lli_dst = desc->lli_phy.dst;
811 	struct d40_phy_lli *lli_src = desc->lli_phy.src;
812 	void __iomem *base = chan_base(chan);
813 
814 	writel(lli_src->reg_cfg, base + D40_CHAN_REG_SSCFG);
815 	writel(lli_src->reg_elt, base + D40_CHAN_REG_SSELT);
816 	writel(lli_src->reg_ptr, base + D40_CHAN_REG_SSPTR);
817 	writel(lli_src->reg_lnk, base + D40_CHAN_REG_SSLNK);
818 
819 	writel(lli_dst->reg_cfg, base + D40_CHAN_REG_SDCFG);
820 	writel(lli_dst->reg_elt, base + D40_CHAN_REG_SDELT);
821 	writel(lli_dst->reg_ptr, base + D40_CHAN_REG_SDPTR);
822 	writel(lli_dst->reg_lnk, base + D40_CHAN_REG_SDLNK);
823 }
824 
825 static void d40_desc_done(struct d40_chan *d40c, struct d40_desc *desc)
826 {
827 	list_add_tail(&desc->node, &d40c->done);
828 }
829 
830 static void d40_log_lli_to_lcxa(struct d40_chan *chan, struct d40_desc *desc)
831 {
832 	struct d40_lcla_pool *pool = &chan->base->lcla_pool;
833 	struct d40_log_lli_bidir *lli = &desc->lli_log;
834 	int lli_current = desc->lli_current;
835 	int lli_len = desc->lli_len;
836 	bool cyclic = desc->cyclic;
837 	int curr_lcla = -EINVAL;
838 	int first_lcla = 0;
839 	bool use_esram_lcla = chan->base->plat_data->use_esram_lcla;
840 	bool linkback;
841 
842 	/*
843 	 * We may have partially running cyclic transfers, in case we did't get
844 	 * enough LCLA entries.
845 	 */
846 	linkback = cyclic && lli_current == 0;
847 
848 	/*
849 	 * For linkback, we need one LCLA even with only one link, because we
850 	 * can't link back to the one in LCPA space
851 	 */
852 	if (linkback || (lli_len - lli_current > 1)) {
853 		/*
854 		 * If the channel is expected to use only soft_lli don't
855 		 * allocate a lcla. This is to avoid a HW issue that exists
856 		 * in some controller during a peripheral to memory transfer
857 		 * that uses linked lists.
858 		 */
859 		if (!(chan->phy_chan->use_soft_lli &&
860 			chan->dma_cfg.dir == DMA_DEV_TO_MEM))
861 			curr_lcla = d40_lcla_alloc_one(chan, desc);
862 
863 		first_lcla = curr_lcla;
864 	}
865 
866 	/*
867 	 * For linkback, we normally load the LCPA in the loop since we need to
868 	 * link it to the second LCLA and not the first.  However, if we
869 	 * couldn't even get a first LCLA, then we have to run in LCPA and
870 	 * reload manually.
871 	 */
872 	if (!linkback || curr_lcla == -EINVAL) {
873 		unsigned int flags = 0;
874 
875 		if (curr_lcla == -EINVAL)
876 			flags |= LLI_TERM_INT;
877 
878 		d40_log_lli_lcpa_write(chan->lcpa,
879 				       &lli->dst[lli_current],
880 				       &lli->src[lli_current],
881 				       curr_lcla,
882 				       flags);
883 		lli_current++;
884 	}
885 
886 	if (curr_lcla < 0)
887 		goto set_current;
888 
889 	for (; lli_current < lli_len; lli_current++) {
890 		unsigned int lcla_offset = chan->phy_chan->num * 1024 +
891 					   8 * curr_lcla * 2;
892 		struct d40_log_lli *lcla = pool->base + lcla_offset;
893 		unsigned int flags = 0;
894 		int next_lcla;
895 
896 		if (lli_current + 1 < lli_len)
897 			next_lcla = d40_lcla_alloc_one(chan, desc);
898 		else
899 			next_lcla = linkback ? first_lcla : -EINVAL;
900 
901 		if (cyclic || next_lcla == -EINVAL)
902 			flags |= LLI_TERM_INT;
903 
904 		if (linkback && curr_lcla == first_lcla) {
905 			/* First link goes in both LCPA and LCLA */
906 			d40_log_lli_lcpa_write(chan->lcpa,
907 					       &lli->dst[lli_current],
908 					       &lli->src[lli_current],
909 					       next_lcla, flags);
910 		}
911 
912 		/*
913 		 * One unused LCLA in the cyclic case if the very first
914 		 * next_lcla fails...
915 		 */
916 		d40_log_lli_lcla_write(lcla,
917 				       &lli->dst[lli_current],
918 				       &lli->src[lli_current],
919 				       next_lcla, flags);
920 
921 		/*
922 		 * Cache maintenance is not needed if lcla is
923 		 * mapped in esram
924 		 */
925 		if (!use_esram_lcla) {
926 			dma_sync_single_range_for_device(chan->base->dev,
927 						pool->dma_addr, lcla_offset,
928 						2 * sizeof(struct d40_log_lli),
929 						DMA_TO_DEVICE);
930 		}
931 		curr_lcla = next_lcla;
932 
933 		if (curr_lcla == -EINVAL || curr_lcla == first_lcla) {
934 			lli_current++;
935 			break;
936 		}
937 	}
938  set_current:
939 	desc->lli_current = lli_current;
940 }
941 
942 static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
943 {
944 	if (chan_is_physical(d40c)) {
945 		d40_phy_lli_load(d40c, d40d);
946 		d40d->lli_current = d40d->lli_len;
947 	} else
948 		d40_log_lli_to_lcxa(d40c, d40d);
949 }
950 
951 static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
952 {
953 	return list_first_entry_or_null(&d40c->active, struct d40_desc, node);
954 }
955 
956 /* remove desc from current queue and add it to the pending_queue */
957 static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
958 {
959 	d40_desc_remove(desc);
960 	desc->is_in_client_list = false;
961 	list_add_tail(&desc->node, &d40c->pending_queue);
962 }
963 
964 static struct d40_desc *d40_first_pending(struct d40_chan *d40c)
965 {
966 	return list_first_entry_or_null(&d40c->pending_queue, struct d40_desc,
967 					node);
968 }
969 
970 static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
971 {
972 	return list_first_entry_or_null(&d40c->queue, struct d40_desc, node);
973 }
974 
975 static struct d40_desc *d40_first_done(struct d40_chan *d40c)
976 {
977 	return list_first_entry_or_null(&d40c->done, struct d40_desc, node);
978 }
979 
980 static int d40_psize_2_burst_size(bool is_log, int psize)
981 {
982 	if (is_log) {
983 		if (psize == STEDMA40_PSIZE_LOG_1)
984 			return 1;
985 	} else {
986 		if (psize == STEDMA40_PSIZE_PHY_1)
987 			return 1;
988 	}
989 
990 	return 2 << psize;
991 }
992 
993 /*
994  * The dma only supports transmitting packages up to
995  * STEDMA40_MAX_SEG_SIZE * data_width, where data_width is stored in Bytes.
996  *
997  * Calculate the total number of dma elements required to send the entire sg list.
998  */
999 static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2)
1000 {
1001 	int dmalen;
1002 	u32 max_w = max(data_width1, data_width2);
1003 	u32 min_w = min(data_width1, data_width2);
1004 	u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE * min_w, max_w);
1005 
1006 	if (seg_max > STEDMA40_MAX_SEG_SIZE)
1007 		seg_max -= max_w;
1008 
1009 	if (!IS_ALIGNED(size, max_w))
1010 		return -EINVAL;
1011 
1012 	if (size <= seg_max)
1013 		dmalen = 1;
1014 	else {
1015 		dmalen = size / seg_max;
1016 		if (dmalen * seg_max < size)
1017 			dmalen++;
1018 	}
1019 	return dmalen;
1020 }
1021 
1022 static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len,
1023 			   u32 data_width1, u32 data_width2)
1024 {
1025 	struct scatterlist *sg;
1026 	int i;
1027 	int len = 0;
1028 	int ret;
1029 
1030 	for_each_sg(sgl, sg, sg_len, i) {
1031 		ret = d40_size_2_dmalen(sg_dma_len(sg),
1032 					data_width1, data_width2);
1033 		if (ret < 0)
1034 			return ret;
1035 		len += ret;
1036 	}
1037 	return len;
1038 }
1039 
1040 static int __d40_execute_command_phy(struct d40_chan *d40c,
1041 				     enum d40_command command)
1042 {
1043 	u32 status;
1044 	int i;
1045 	void __iomem *active_reg;
1046 	int ret = 0;
1047 	unsigned long flags;
1048 	u32 wmask;
1049 
1050 	if (command == D40_DMA_STOP) {
1051 		ret = __d40_execute_command_phy(d40c, D40_DMA_SUSPEND_REQ);
1052 		if (ret)
1053 			return ret;
1054 	}
1055 
1056 	spin_lock_irqsave(&d40c->base->execmd_lock, flags);
1057 
1058 	if (d40c->phy_chan->num % 2 == 0)
1059 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1060 	else
1061 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1062 
1063 	if (command == D40_DMA_SUSPEND_REQ) {
1064 		status = (readl(active_reg) &
1065 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1066 			D40_CHAN_POS(d40c->phy_chan->num);
1067 
1068 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1069 			goto unlock;
1070 	}
1071 
1072 	wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
1073 	writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
1074 	       active_reg);
1075 
1076 	if (command == D40_DMA_SUSPEND_REQ) {
1077 
1078 		for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
1079 			status = (readl(active_reg) &
1080 				  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1081 				D40_CHAN_POS(d40c->phy_chan->num);
1082 
1083 			cpu_relax();
1084 			/*
1085 			 * Reduce the number of bus accesses while
1086 			 * waiting for the DMA to suspend.
1087 			 */
1088 			udelay(3);
1089 
1090 			if (status == D40_DMA_STOP ||
1091 			    status == D40_DMA_SUSPENDED)
1092 				break;
1093 		}
1094 
1095 		if (i == D40_SUSPEND_MAX_IT) {
1096 			chan_err(d40c,
1097 				"unable to suspend the chl %d (log: %d) status %x\n",
1098 				d40c->phy_chan->num, d40c->log_num,
1099 				status);
1100 			dump_stack();
1101 			ret = -EBUSY;
1102 		}
1103 
1104 	}
1105  unlock:
1106 	spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
1107 	return ret;
1108 }
1109 
1110 static void d40_term_all(struct d40_chan *d40c)
1111 {
1112 	struct d40_desc *d40d;
1113 	struct d40_desc *_d;
1114 
1115 	/* Release completed descriptors */
1116 	while ((d40d = d40_first_done(d40c))) {
1117 		d40_desc_remove(d40d);
1118 		d40_desc_free(d40c, d40d);
1119 	}
1120 
1121 	/* Release active descriptors */
1122 	while ((d40d = d40_first_active_get(d40c))) {
1123 		d40_desc_remove(d40d);
1124 		d40_desc_free(d40c, d40d);
1125 	}
1126 
1127 	/* Release queued descriptors waiting for transfer */
1128 	while ((d40d = d40_first_queued(d40c))) {
1129 		d40_desc_remove(d40d);
1130 		d40_desc_free(d40c, d40d);
1131 	}
1132 
1133 	/* Release pending descriptors */
1134 	while ((d40d = d40_first_pending(d40c))) {
1135 		d40_desc_remove(d40d);
1136 		d40_desc_free(d40c, d40d);
1137 	}
1138 
1139 	/* Release client owned descriptors */
1140 	if (!list_empty(&d40c->client))
1141 		list_for_each_entry_safe(d40d, _d, &d40c->client, node) {
1142 			d40_desc_remove(d40d);
1143 			d40_desc_free(d40c, d40d);
1144 		}
1145 
1146 	/* Release descriptors in prepare queue */
1147 	if (!list_empty(&d40c->prepare_queue))
1148 		list_for_each_entry_safe(d40d, _d,
1149 					 &d40c->prepare_queue, node) {
1150 			d40_desc_remove(d40d);
1151 			d40_desc_free(d40c, d40d);
1152 		}
1153 
1154 	d40c->pending_tx = 0;
1155 }
1156 
1157 static void __d40_config_set_event(struct d40_chan *d40c,
1158 				   enum d40_events event_type, u32 event,
1159 				   int reg)
1160 {
1161 	void __iomem *addr = chan_base(d40c) + reg;
1162 	int tries;
1163 	u32 status;
1164 
1165 	switch (event_type) {
1166 
1167 	case D40_DEACTIVATE_EVENTLINE:
1168 
1169 		writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
1170 		       | ~D40_EVENTLINE_MASK(event), addr);
1171 		break;
1172 
1173 	case D40_SUSPEND_REQ_EVENTLINE:
1174 		status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1175 			  D40_EVENTLINE_POS(event);
1176 
1177 		if (status == D40_DEACTIVATE_EVENTLINE ||
1178 		    status == D40_SUSPEND_REQ_EVENTLINE)
1179 			break;
1180 
1181 		writel((D40_SUSPEND_REQ_EVENTLINE << D40_EVENTLINE_POS(event))
1182 		       | ~D40_EVENTLINE_MASK(event), addr);
1183 
1184 		for (tries = 0 ; tries < D40_SUSPEND_MAX_IT; tries++) {
1185 
1186 			status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1187 				  D40_EVENTLINE_POS(event);
1188 
1189 			cpu_relax();
1190 			/*
1191 			 * Reduce the number of bus accesses while
1192 			 * waiting for the DMA to suspend.
1193 			 */
1194 			udelay(3);
1195 
1196 			if (status == D40_DEACTIVATE_EVENTLINE)
1197 				break;
1198 		}
1199 
1200 		if (tries == D40_SUSPEND_MAX_IT) {
1201 			chan_err(d40c,
1202 				"unable to stop the event_line chl %d (log: %d)"
1203 				"status %x\n", d40c->phy_chan->num,
1204 				 d40c->log_num, status);
1205 		}
1206 		break;
1207 
1208 	case D40_ACTIVATE_EVENTLINE:
1209 	/*
1210 	 * The hardware sometimes doesn't register the enable when src and dst
1211 	 * event lines are active on the same logical channel.  Retry to ensure
1212 	 * it does.  Usually only one retry is sufficient.
1213 	 */
1214 		tries = 100;
1215 		while (--tries) {
1216 			writel((D40_ACTIVATE_EVENTLINE <<
1217 				D40_EVENTLINE_POS(event)) |
1218 				~D40_EVENTLINE_MASK(event), addr);
1219 
1220 			if (readl(addr) & D40_EVENTLINE_MASK(event))
1221 				break;
1222 		}
1223 
1224 		if (tries != 99)
1225 			dev_dbg(chan2dev(d40c),
1226 				"[%s] workaround enable S%cLNK (%d tries)\n",
1227 				__func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D',
1228 				100 - tries);
1229 
1230 		WARN_ON(!tries);
1231 		break;
1232 
1233 	case D40_ROUND_EVENTLINE:
1234 		BUG();
1235 		break;
1236 
1237 	}
1238 }
1239 
1240 static void d40_config_set_event(struct d40_chan *d40c,
1241 				 enum d40_events event_type)
1242 {
1243 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
1244 
1245 	/* Enable event line connected to device (or memcpy) */
1246 	if ((d40c->dma_cfg.dir == DMA_DEV_TO_MEM) ||
1247 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
1248 		__d40_config_set_event(d40c, event_type, event,
1249 				       D40_CHAN_REG_SSLNK);
1250 
1251 	if (d40c->dma_cfg.dir !=  DMA_DEV_TO_MEM)
1252 		__d40_config_set_event(d40c, event_type, event,
1253 				       D40_CHAN_REG_SDLNK);
1254 }
1255 
1256 static u32 d40_chan_has_events(struct d40_chan *d40c)
1257 {
1258 	void __iomem *chanbase = chan_base(d40c);
1259 	u32 val;
1260 
1261 	val = readl(chanbase + D40_CHAN_REG_SSLNK);
1262 	val |= readl(chanbase + D40_CHAN_REG_SDLNK);
1263 
1264 	return val;
1265 }
1266 
1267 static int
1268 __d40_execute_command_log(struct d40_chan *d40c, enum d40_command command)
1269 {
1270 	unsigned long flags;
1271 	int ret = 0;
1272 	u32 active_status;
1273 	void __iomem *active_reg;
1274 
1275 	if (d40c->phy_chan->num % 2 == 0)
1276 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1277 	else
1278 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1279 
1280 
1281 	spin_lock_irqsave(&d40c->phy_chan->lock, flags);
1282 
1283 	switch (command) {
1284 	case D40_DMA_STOP:
1285 	case D40_DMA_SUSPEND_REQ:
1286 
1287 		active_status = (readl(active_reg) &
1288 				 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1289 				 D40_CHAN_POS(d40c->phy_chan->num);
1290 
1291 		if (active_status == D40_DMA_RUN)
1292 			d40_config_set_event(d40c, D40_SUSPEND_REQ_EVENTLINE);
1293 		else
1294 			d40_config_set_event(d40c, D40_DEACTIVATE_EVENTLINE);
1295 
1296 		if (!d40_chan_has_events(d40c) && (command == D40_DMA_STOP))
1297 			ret = __d40_execute_command_phy(d40c, command);
1298 
1299 		break;
1300 
1301 	case D40_DMA_RUN:
1302 
1303 		d40_config_set_event(d40c, D40_ACTIVATE_EVENTLINE);
1304 		ret = __d40_execute_command_phy(d40c, command);
1305 		break;
1306 
1307 	case D40_DMA_SUSPENDED:
1308 		BUG();
1309 		break;
1310 	}
1311 
1312 	spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
1313 	return ret;
1314 }
1315 
1316 static int d40_channel_execute_command(struct d40_chan *d40c,
1317 				       enum d40_command command)
1318 {
1319 	if (chan_is_logical(d40c))
1320 		return __d40_execute_command_log(d40c, command);
1321 	else
1322 		return __d40_execute_command_phy(d40c, command);
1323 }
1324 
1325 static u32 d40_get_prmo(struct d40_chan *d40c)
1326 {
1327 	static const unsigned int phy_map[] = {
1328 		[STEDMA40_PCHAN_BASIC_MODE]
1329 			= D40_DREG_PRMO_PCHAN_BASIC,
1330 		[STEDMA40_PCHAN_MODULO_MODE]
1331 			= D40_DREG_PRMO_PCHAN_MODULO,
1332 		[STEDMA40_PCHAN_DOUBLE_DST_MODE]
1333 			= D40_DREG_PRMO_PCHAN_DOUBLE_DST,
1334 	};
1335 	static const unsigned int log_map[] = {
1336 		[STEDMA40_LCHAN_SRC_PHY_DST_LOG]
1337 			= D40_DREG_PRMO_LCHAN_SRC_PHY_DST_LOG,
1338 		[STEDMA40_LCHAN_SRC_LOG_DST_PHY]
1339 			= D40_DREG_PRMO_LCHAN_SRC_LOG_DST_PHY,
1340 		[STEDMA40_LCHAN_SRC_LOG_DST_LOG]
1341 			= D40_DREG_PRMO_LCHAN_SRC_LOG_DST_LOG,
1342 	};
1343 
1344 	if (chan_is_physical(d40c))
1345 		return phy_map[d40c->dma_cfg.mode_opt];
1346 	else
1347 		return log_map[d40c->dma_cfg.mode_opt];
1348 }
1349 
1350 static void d40_config_write(struct d40_chan *d40c)
1351 {
1352 	u32 addr_base;
1353 	u32 var;
1354 
1355 	/* Odd addresses are even addresses + 4 */
1356 	addr_base = (d40c->phy_chan->num % 2) * 4;
1357 	/* Setup channel mode to logical or physical */
1358 	var = ((u32)(chan_is_logical(d40c)) + 1) <<
1359 		D40_CHAN_POS(d40c->phy_chan->num);
1360 	writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
1361 
1362 	/* Setup operational mode option register */
1363 	var = d40_get_prmo(d40c) << D40_CHAN_POS(d40c->phy_chan->num);
1364 
1365 	writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
1366 
1367 	if (chan_is_logical(d40c)) {
1368 		int lidx = (d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS)
1369 			   & D40_SREG_ELEM_LOG_LIDX_MASK;
1370 		void __iomem *chanbase = chan_base(d40c);
1371 
1372 		/* Set default config for CFG reg */
1373 		writel(d40c->src_def_cfg, chanbase + D40_CHAN_REG_SSCFG);
1374 		writel(d40c->dst_def_cfg, chanbase + D40_CHAN_REG_SDCFG);
1375 
1376 		/* Set LIDX for lcla */
1377 		writel(lidx, chanbase + D40_CHAN_REG_SSELT);
1378 		writel(lidx, chanbase + D40_CHAN_REG_SDELT);
1379 
1380 		/* Clear LNK which will be used by d40_chan_has_events() */
1381 		writel(0, chanbase + D40_CHAN_REG_SSLNK);
1382 		writel(0, chanbase + D40_CHAN_REG_SDLNK);
1383 	}
1384 }
1385 
1386 static u32 d40_residue(struct d40_chan *d40c)
1387 {
1388 	u32 num_elt;
1389 
1390 	if (chan_is_logical(d40c))
1391 		num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
1392 			>> D40_MEM_LCSP2_ECNT_POS;
1393 	else {
1394 		u32 val = readl(chan_base(d40c) + D40_CHAN_REG_SDELT);
1395 		num_elt = (val & D40_SREG_ELEM_PHY_ECNT_MASK)
1396 			  >> D40_SREG_ELEM_PHY_ECNT_POS;
1397 	}
1398 
1399 	return num_elt * d40c->dma_cfg.dst_info.data_width;
1400 }
1401 
1402 static bool d40_tx_is_linked(struct d40_chan *d40c)
1403 {
1404 	bool is_link;
1405 
1406 	if (chan_is_logical(d40c))
1407 		is_link = readl(&d40c->lcpa->lcsp3) &  D40_MEM_LCSP3_DLOS_MASK;
1408 	else
1409 		is_link = readl(chan_base(d40c) + D40_CHAN_REG_SDLNK)
1410 			  & D40_SREG_LNK_PHYS_LNK_MASK;
1411 
1412 	return is_link;
1413 }
1414 
1415 static int d40_pause(struct dma_chan *chan)
1416 {
1417 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
1418 	int res = 0;
1419 	unsigned long flags;
1420 
1421 	if (d40c->phy_chan == NULL) {
1422 		chan_err(d40c, "Channel is not allocated!\n");
1423 		return -EINVAL;
1424 	}
1425 
1426 	if (!d40c->busy)
1427 		return 0;
1428 
1429 	spin_lock_irqsave(&d40c->lock, flags);
1430 	pm_runtime_get_sync(d40c->base->dev);
1431 
1432 	res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1433 
1434 	pm_runtime_mark_last_busy(d40c->base->dev);
1435 	pm_runtime_put_autosuspend(d40c->base->dev);
1436 	spin_unlock_irqrestore(&d40c->lock, flags);
1437 	return res;
1438 }
1439 
1440 static int d40_resume(struct dma_chan *chan)
1441 {
1442 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
1443 	int res = 0;
1444 	unsigned long flags;
1445 
1446 	if (d40c->phy_chan == NULL) {
1447 		chan_err(d40c, "Channel is not allocated!\n");
1448 		return -EINVAL;
1449 	}
1450 
1451 	if (!d40c->busy)
1452 		return 0;
1453 
1454 	spin_lock_irqsave(&d40c->lock, flags);
1455 	pm_runtime_get_sync(d40c->base->dev);
1456 
1457 	/* If bytes left to transfer or linked tx resume job */
1458 	if (d40_residue(d40c) || d40_tx_is_linked(d40c))
1459 		res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1460 
1461 	pm_runtime_mark_last_busy(d40c->base->dev);
1462 	pm_runtime_put_autosuspend(d40c->base->dev);
1463 	spin_unlock_irqrestore(&d40c->lock, flags);
1464 	return res;
1465 }
1466 
1467 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
1468 {
1469 	struct d40_chan *d40c = container_of(tx->chan,
1470 					     struct d40_chan,
1471 					     chan);
1472 	struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
1473 	unsigned long flags;
1474 	dma_cookie_t cookie;
1475 
1476 	spin_lock_irqsave(&d40c->lock, flags);
1477 	cookie = dma_cookie_assign(tx);
1478 	d40_desc_queue(d40c, d40d);
1479 	spin_unlock_irqrestore(&d40c->lock, flags);
1480 
1481 	return cookie;
1482 }
1483 
1484 static int d40_start(struct d40_chan *d40c)
1485 {
1486 	return d40_channel_execute_command(d40c, D40_DMA_RUN);
1487 }
1488 
1489 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
1490 {
1491 	struct d40_desc *d40d;
1492 	int err;
1493 
1494 	/* Start queued jobs, if any */
1495 	d40d = d40_first_queued(d40c);
1496 
1497 	if (d40d != NULL) {
1498 		if (!d40c->busy) {
1499 			d40c->busy = true;
1500 			pm_runtime_get_sync(d40c->base->dev);
1501 		}
1502 
1503 		/* Remove from queue */
1504 		d40_desc_remove(d40d);
1505 
1506 		/* Add to active queue */
1507 		d40_desc_submit(d40c, d40d);
1508 
1509 		/* Initiate DMA job */
1510 		d40_desc_load(d40c, d40d);
1511 
1512 		/* Start dma job */
1513 		err = d40_start(d40c);
1514 
1515 		if (err)
1516 			return NULL;
1517 	}
1518 
1519 	return d40d;
1520 }
1521 
1522 /* called from interrupt context */
1523 static void dma_tc_handle(struct d40_chan *d40c)
1524 {
1525 	struct d40_desc *d40d;
1526 
1527 	/* Get first active entry from list */
1528 	d40d = d40_first_active_get(d40c);
1529 
1530 	if (d40d == NULL)
1531 		return;
1532 
1533 	if (d40d->cyclic) {
1534 		/*
1535 		 * If this was a paritially loaded list, we need to reloaded
1536 		 * it, and only when the list is completed.  We need to check
1537 		 * for done because the interrupt will hit for every link, and
1538 		 * not just the last one.
1539 		 */
1540 		if (d40d->lli_current < d40d->lli_len
1541 		    && !d40_tx_is_linked(d40c)
1542 		    && !d40_residue(d40c)) {
1543 			d40_lcla_free_all(d40c, d40d);
1544 			d40_desc_load(d40c, d40d);
1545 			(void) d40_start(d40c);
1546 
1547 			if (d40d->lli_current == d40d->lli_len)
1548 				d40d->lli_current = 0;
1549 		}
1550 	} else {
1551 		d40_lcla_free_all(d40c, d40d);
1552 
1553 		if (d40d->lli_current < d40d->lli_len) {
1554 			d40_desc_load(d40c, d40d);
1555 			/* Start dma job */
1556 			(void) d40_start(d40c);
1557 			return;
1558 		}
1559 
1560 		if (d40_queue_start(d40c) == NULL) {
1561 			d40c->busy = false;
1562 
1563 			pm_runtime_mark_last_busy(d40c->base->dev);
1564 			pm_runtime_put_autosuspend(d40c->base->dev);
1565 		}
1566 
1567 		d40_desc_remove(d40d);
1568 		d40_desc_done(d40c, d40d);
1569 	}
1570 
1571 	d40c->pending_tx++;
1572 	tasklet_schedule(&d40c->tasklet);
1573 
1574 }
1575 
1576 static void dma_tasklet(unsigned long data)
1577 {
1578 	struct d40_chan *d40c = (struct d40_chan *) data;
1579 	struct d40_desc *d40d;
1580 	unsigned long flags;
1581 	bool callback_active;
1582 	struct dmaengine_desc_callback cb;
1583 
1584 	spin_lock_irqsave(&d40c->lock, flags);
1585 
1586 	/* Get first entry from the done list */
1587 	d40d = d40_first_done(d40c);
1588 	if (d40d == NULL) {
1589 		/* Check if we have reached here for cyclic job */
1590 		d40d = d40_first_active_get(d40c);
1591 		if (d40d == NULL || !d40d->cyclic)
1592 			goto check_pending_tx;
1593 	}
1594 
1595 	if (!d40d->cyclic)
1596 		dma_cookie_complete(&d40d->txd);
1597 
1598 	/*
1599 	 * If terminating a channel pending_tx is set to zero.
1600 	 * This prevents any finished active jobs to return to the client.
1601 	 */
1602 	if (d40c->pending_tx == 0) {
1603 		spin_unlock_irqrestore(&d40c->lock, flags);
1604 		return;
1605 	}
1606 
1607 	/* Callback to client */
1608 	callback_active = !!(d40d->txd.flags & DMA_PREP_INTERRUPT);
1609 	dmaengine_desc_get_callback(&d40d->txd, &cb);
1610 
1611 	if (!d40d->cyclic) {
1612 		if (async_tx_test_ack(&d40d->txd)) {
1613 			d40_desc_remove(d40d);
1614 			d40_desc_free(d40c, d40d);
1615 		} else if (!d40d->is_in_client_list) {
1616 			d40_desc_remove(d40d);
1617 			d40_lcla_free_all(d40c, d40d);
1618 			list_add_tail(&d40d->node, &d40c->client);
1619 			d40d->is_in_client_list = true;
1620 		}
1621 	}
1622 
1623 	d40c->pending_tx--;
1624 
1625 	if (d40c->pending_tx)
1626 		tasklet_schedule(&d40c->tasklet);
1627 
1628 	spin_unlock_irqrestore(&d40c->lock, flags);
1629 
1630 	if (callback_active)
1631 		dmaengine_desc_callback_invoke(&cb, NULL);
1632 
1633 	return;
1634  check_pending_tx:
1635 	/* Rescue manouver if receiving double interrupts */
1636 	if (d40c->pending_tx > 0)
1637 		d40c->pending_tx--;
1638 	spin_unlock_irqrestore(&d40c->lock, flags);
1639 }
1640 
1641 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1642 {
1643 	int i;
1644 	u32 idx;
1645 	u32 row;
1646 	long chan = -1;
1647 	struct d40_chan *d40c;
1648 	unsigned long flags;
1649 	struct d40_base *base = data;
1650 	u32 *regs = base->regs_interrupt;
1651 	struct d40_interrupt_lookup *il = base->gen_dmac.il;
1652 	u32 il_size = base->gen_dmac.il_size;
1653 
1654 	spin_lock_irqsave(&base->interrupt_lock, flags);
1655 
1656 	/* Read interrupt status of both logical and physical channels */
1657 	for (i = 0; i < il_size; i++)
1658 		regs[i] = readl(base->virtbase + il[i].src);
1659 
1660 	for (;;) {
1661 
1662 		chan = find_next_bit((unsigned long *)regs,
1663 				     BITS_PER_LONG * il_size, chan + 1);
1664 
1665 		/* No more set bits found? */
1666 		if (chan == BITS_PER_LONG * il_size)
1667 			break;
1668 
1669 		row = chan / BITS_PER_LONG;
1670 		idx = chan & (BITS_PER_LONG - 1);
1671 
1672 		if (il[row].offset == D40_PHY_CHAN)
1673 			d40c = base->lookup_phy_chans[idx];
1674 		else
1675 			d40c = base->lookup_log_chans[il[row].offset + idx];
1676 
1677 		if (!d40c) {
1678 			/*
1679 			 * No error because this can happen if something else
1680 			 * in the system is using the channel.
1681 			 */
1682 			continue;
1683 		}
1684 
1685 		/* ACK interrupt */
1686 		writel(BIT(idx), base->virtbase + il[row].clr);
1687 
1688 		spin_lock(&d40c->lock);
1689 
1690 		if (!il[row].is_error)
1691 			dma_tc_handle(d40c);
1692 		else
1693 			d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1694 				chan, il[row].offset, idx);
1695 
1696 		spin_unlock(&d40c->lock);
1697 	}
1698 
1699 	spin_unlock_irqrestore(&base->interrupt_lock, flags);
1700 
1701 	return IRQ_HANDLED;
1702 }
1703 
1704 static int d40_validate_conf(struct d40_chan *d40c,
1705 			     struct stedma40_chan_cfg *conf)
1706 {
1707 	int res = 0;
1708 	bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1709 
1710 	if (!conf->dir) {
1711 		chan_err(d40c, "Invalid direction.\n");
1712 		res = -EINVAL;
1713 	}
1714 
1715 	if ((is_log && conf->dev_type > d40c->base->num_log_chans)  ||
1716 	    (!is_log && conf->dev_type > d40c->base->num_phy_chans) ||
1717 	    (conf->dev_type < 0)) {
1718 		chan_err(d40c, "Invalid device type (%d)\n", conf->dev_type);
1719 		res = -EINVAL;
1720 	}
1721 
1722 	if (conf->dir == DMA_DEV_TO_DEV) {
1723 		/*
1724 		 * DMAC HW supports it. Will be added to this driver,
1725 		 * in case any dma client requires it.
1726 		 */
1727 		chan_err(d40c, "periph to periph not supported\n");
1728 		res = -EINVAL;
1729 	}
1730 
1731 	if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1732 	    conf->src_info.data_width !=
1733 	    d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1734 	    conf->dst_info.data_width) {
1735 		/*
1736 		 * The DMAC hardware only supports
1737 		 * src (burst x width) == dst (burst x width)
1738 		 */
1739 
1740 		chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1741 		res = -EINVAL;
1742 	}
1743 
1744 	return res;
1745 }
1746 
1747 static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1748 			       bool is_src, int log_event_line, bool is_log,
1749 			       bool *first_user)
1750 {
1751 	unsigned long flags;
1752 	spin_lock_irqsave(&phy->lock, flags);
1753 
1754 	*first_user = ((phy->allocated_src | phy->allocated_dst)
1755 			== D40_ALLOC_FREE);
1756 
1757 	if (!is_log) {
1758 		/* Physical interrupts are masked per physical full channel */
1759 		if (phy->allocated_src == D40_ALLOC_FREE &&
1760 		    phy->allocated_dst == D40_ALLOC_FREE) {
1761 			phy->allocated_dst = D40_ALLOC_PHY;
1762 			phy->allocated_src = D40_ALLOC_PHY;
1763 			goto found_unlock;
1764 		} else
1765 			goto not_found_unlock;
1766 	}
1767 
1768 	/* Logical channel */
1769 	if (is_src) {
1770 		if (phy->allocated_src == D40_ALLOC_PHY)
1771 			goto not_found_unlock;
1772 
1773 		if (phy->allocated_src == D40_ALLOC_FREE)
1774 			phy->allocated_src = D40_ALLOC_LOG_FREE;
1775 
1776 		if (!(phy->allocated_src & BIT(log_event_line))) {
1777 			phy->allocated_src |= BIT(log_event_line);
1778 			goto found_unlock;
1779 		} else
1780 			goto not_found_unlock;
1781 	} else {
1782 		if (phy->allocated_dst == D40_ALLOC_PHY)
1783 			goto not_found_unlock;
1784 
1785 		if (phy->allocated_dst == D40_ALLOC_FREE)
1786 			phy->allocated_dst = D40_ALLOC_LOG_FREE;
1787 
1788 		if (!(phy->allocated_dst & BIT(log_event_line))) {
1789 			phy->allocated_dst |= BIT(log_event_line);
1790 			goto found_unlock;
1791 		}
1792 	}
1793  not_found_unlock:
1794 	spin_unlock_irqrestore(&phy->lock, flags);
1795 	return false;
1796  found_unlock:
1797 	spin_unlock_irqrestore(&phy->lock, flags);
1798 	return true;
1799 }
1800 
1801 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1802 			       int log_event_line)
1803 {
1804 	unsigned long flags;
1805 	bool is_free = false;
1806 
1807 	spin_lock_irqsave(&phy->lock, flags);
1808 	if (!log_event_line) {
1809 		phy->allocated_dst = D40_ALLOC_FREE;
1810 		phy->allocated_src = D40_ALLOC_FREE;
1811 		is_free = true;
1812 		goto unlock;
1813 	}
1814 
1815 	/* Logical channel */
1816 	if (is_src) {
1817 		phy->allocated_src &= ~BIT(log_event_line);
1818 		if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1819 			phy->allocated_src = D40_ALLOC_FREE;
1820 	} else {
1821 		phy->allocated_dst &= ~BIT(log_event_line);
1822 		if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1823 			phy->allocated_dst = D40_ALLOC_FREE;
1824 	}
1825 
1826 	is_free = ((phy->allocated_src | phy->allocated_dst) ==
1827 		   D40_ALLOC_FREE);
1828  unlock:
1829 	spin_unlock_irqrestore(&phy->lock, flags);
1830 
1831 	return is_free;
1832 }
1833 
1834 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1835 {
1836 	int dev_type = d40c->dma_cfg.dev_type;
1837 	int event_group;
1838 	int event_line;
1839 	struct d40_phy_res *phys;
1840 	int i;
1841 	int j;
1842 	int log_num;
1843 	int num_phy_chans;
1844 	bool is_src;
1845 	bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1846 
1847 	phys = d40c->base->phy_res;
1848 	num_phy_chans = d40c->base->num_phy_chans;
1849 
1850 	if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
1851 		log_num = 2 * dev_type;
1852 		is_src = true;
1853 	} else if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
1854 		   d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1855 		/* dst event lines are used for logical memcpy */
1856 		log_num = 2 * dev_type + 1;
1857 		is_src = false;
1858 	} else
1859 		return -EINVAL;
1860 
1861 	event_group = D40_TYPE_TO_GROUP(dev_type);
1862 	event_line = D40_TYPE_TO_EVENT(dev_type);
1863 
1864 	if (!is_log) {
1865 		if (d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1866 			/* Find physical half channel */
1867 			if (d40c->dma_cfg.use_fixed_channel) {
1868 				i = d40c->dma_cfg.phy_channel;
1869 				if (d40_alloc_mask_set(&phys[i], is_src,
1870 						       0, is_log,
1871 						       first_phy_user))
1872 					goto found_phy;
1873 			} else {
1874 				for (i = 0; i < num_phy_chans; i++) {
1875 					if (d40_alloc_mask_set(&phys[i], is_src,
1876 						       0, is_log,
1877 						       first_phy_user))
1878 						goto found_phy;
1879 				}
1880 			}
1881 		} else
1882 			for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1883 				int phy_num = j  + event_group * 2;
1884 				for (i = phy_num; i < phy_num + 2; i++) {
1885 					if (d40_alloc_mask_set(&phys[i],
1886 							       is_src,
1887 							       0,
1888 							       is_log,
1889 							       first_phy_user))
1890 						goto found_phy;
1891 				}
1892 			}
1893 		return -EINVAL;
1894 found_phy:
1895 		d40c->phy_chan = &phys[i];
1896 		d40c->log_num = D40_PHY_CHAN;
1897 		goto out;
1898 	}
1899 	if (dev_type == -1)
1900 		return -EINVAL;
1901 
1902 	/* Find logical channel */
1903 	for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1904 		int phy_num = j + event_group * 2;
1905 
1906 		if (d40c->dma_cfg.use_fixed_channel) {
1907 			i = d40c->dma_cfg.phy_channel;
1908 
1909 			if ((i != phy_num) && (i != phy_num + 1)) {
1910 				dev_err(chan2dev(d40c),
1911 					"invalid fixed phy channel %d\n", i);
1912 				return -EINVAL;
1913 			}
1914 
1915 			if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1916 					       is_log, first_phy_user))
1917 				goto found_log;
1918 
1919 			dev_err(chan2dev(d40c),
1920 				"could not allocate fixed phy channel %d\n", i);
1921 			return -EINVAL;
1922 		}
1923 
1924 		/*
1925 		 * Spread logical channels across all available physical rather
1926 		 * than pack every logical channel at the first available phy
1927 		 * channels.
1928 		 */
1929 		if (is_src) {
1930 			for (i = phy_num; i < phy_num + 2; i++) {
1931 				if (d40_alloc_mask_set(&phys[i], is_src,
1932 						       event_line, is_log,
1933 						       first_phy_user))
1934 					goto found_log;
1935 			}
1936 		} else {
1937 			for (i = phy_num + 1; i >= phy_num; i--) {
1938 				if (d40_alloc_mask_set(&phys[i], is_src,
1939 						       event_line, is_log,
1940 						       first_phy_user))
1941 					goto found_log;
1942 			}
1943 		}
1944 	}
1945 	return -EINVAL;
1946 
1947 found_log:
1948 	d40c->phy_chan = &phys[i];
1949 	d40c->log_num = log_num;
1950 out:
1951 
1952 	if (is_log)
1953 		d40c->base->lookup_log_chans[d40c->log_num] = d40c;
1954 	else
1955 		d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
1956 
1957 	return 0;
1958 
1959 }
1960 
1961 static int d40_config_memcpy(struct d40_chan *d40c)
1962 {
1963 	dma_cap_mask_t cap = d40c->chan.device->cap_mask;
1964 
1965 	if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
1966 		d40c->dma_cfg = dma40_memcpy_conf_log;
1967 		d40c->dma_cfg.dev_type = dma40_memcpy_channels[d40c->chan.chan_id];
1968 
1969 		d40_log_cfg(&d40c->dma_cfg,
1970 			    &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
1971 
1972 	} else if (dma_has_cap(DMA_MEMCPY, cap) &&
1973 		   dma_has_cap(DMA_SLAVE, cap)) {
1974 		d40c->dma_cfg = dma40_memcpy_conf_phy;
1975 
1976 		/* Generate interrrupt at end of transfer or relink. */
1977 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_TIM_POS);
1978 
1979 		/* Generate interrupt on error. */
1980 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
1981 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
1982 
1983 	} else {
1984 		chan_err(d40c, "No memcpy\n");
1985 		return -EINVAL;
1986 	}
1987 
1988 	return 0;
1989 }
1990 
1991 static int d40_free_dma(struct d40_chan *d40c)
1992 {
1993 
1994 	int res = 0;
1995 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
1996 	struct d40_phy_res *phy = d40c->phy_chan;
1997 	bool is_src;
1998 
1999 	/* Terminate all queued and active transfers */
2000 	d40_term_all(d40c);
2001 
2002 	if (phy == NULL) {
2003 		chan_err(d40c, "phy == null\n");
2004 		return -EINVAL;
2005 	}
2006 
2007 	if (phy->allocated_src == D40_ALLOC_FREE &&
2008 	    phy->allocated_dst == D40_ALLOC_FREE) {
2009 		chan_err(d40c, "channel already free\n");
2010 		return -EINVAL;
2011 	}
2012 
2013 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2014 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM)
2015 		is_src = false;
2016 	else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2017 		is_src = true;
2018 	else {
2019 		chan_err(d40c, "Unknown direction\n");
2020 		return -EINVAL;
2021 	}
2022 
2023 	pm_runtime_get_sync(d40c->base->dev);
2024 	res = d40_channel_execute_command(d40c, D40_DMA_STOP);
2025 	if (res) {
2026 		chan_err(d40c, "stop failed\n");
2027 		goto mark_last_busy;
2028 	}
2029 
2030 	d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
2031 
2032 	if (chan_is_logical(d40c))
2033 		d40c->base->lookup_log_chans[d40c->log_num] = NULL;
2034 	else
2035 		d40c->base->lookup_phy_chans[phy->num] = NULL;
2036 
2037 	if (d40c->busy) {
2038 		pm_runtime_mark_last_busy(d40c->base->dev);
2039 		pm_runtime_put_autosuspend(d40c->base->dev);
2040 	}
2041 
2042 	d40c->busy = false;
2043 	d40c->phy_chan = NULL;
2044 	d40c->configured = false;
2045  mark_last_busy:
2046 	pm_runtime_mark_last_busy(d40c->base->dev);
2047 	pm_runtime_put_autosuspend(d40c->base->dev);
2048 	return res;
2049 }
2050 
2051 static bool d40_is_paused(struct d40_chan *d40c)
2052 {
2053 	void __iomem *chanbase = chan_base(d40c);
2054 	bool is_paused = false;
2055 	unsigned long flags;
2056 	void __iomem *active_reg;
2057 	u32 status;
2058 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2059 
2060 	spin_lock_irqsave(&d40c->lock, flags);
2061 
2062 	if (chan_is_physical(d40c)) {
2063 		if (d40c->phy_chan->num % 2 == 0)
2064 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
2065 		else
2066 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
2067 
2068 		status = (readl(active_reg) &
2069 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
2070 			D40_CHAN_POS(d40c->phy_chan->num);
2071 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
2072 			is_paused = true;
2073 		goto unlock;
2074 	}
2075 
2076 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2077 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
2078 		status = readl(chanbase + D40_CHAN_REG_SDLNK);
2079 	} else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
2080 		status = readl(chanbase + D40_CHAN_REG_SSLNK);
2081 	} else {
2082 		chan_err(d40c, "Unknown direction\n");
2083 		goto unlock;
2084 	}
2085 
2086 	status = (status & D40_EVENTLINE_MASK(event)) >>
2087 		D40_EVENTLINE_POS(event);
2088 
2089 	if (status != D40_DMA_RUN)
2090 		is_paused = true;
2091  unlock:
2092 	spin_unlock_irqrestore(&d40c->lock, flags);
2093 	return is_paused;
2094 
2095 }
2096 
2097 static u32 stedma40_residue(struct dma_chan *chan)
2098 {
2099 	struct d40_chan *d40c =
2100 		container_of(chan, struct d40_chan, chan);
2101 	u32 bytes_left;
2102 	unsigned long flags;
2103 
2104 	spin_lock_irqsave(&d40c->lock, flags);
2105 	bytes_left = d40_residue(d40c);
2106 	spin_unlock_irqrestore(&d40c->lock, flags);
2107 
2108 	return bytes_left;
2109 }
2110 
2111 static int
2112 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
2113 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2114 		unsigned int sg_len, dma_addr_t src_dev_addr,
2115 		dma_addr_t dst_dev_addr)
2116 {
2117 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2118 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2119 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2120 	int ret;
2121 
2122 	ret = d40_log_sg_to_lli(sg_src, sg_len,
2123 				src_dev_addr,
2124 				desc->lli_log.src,
2125 				chan->log_def.lcsp1,
2126 				src_info->data_width,
2127 				dst_info->data_width);
2128 
2129 	ret = d40_log_sg_to_lli(sg_dst, sg_len,
2130 				dst_dev_addr,
2131 				desc->lli_log.dst,
2132 				chan->log_def.lcsp3,
2133 				dst_info->data_width,
2134 				src_info->data_width);
2135 
2136 	return ret < 0 ? ret : 0;
2137 }
2138 
2139 static int
2140 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2141 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2142 		unsigned int sg_len, dma_addr_t src_dev_addr,
2143 		dma_addr_t dst_dev_addr)
2144 {
2145 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2146 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2147 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2148 	unsigned long flags = 0;
2149 	int ret;
2150 
2151 	if (desc->cyclic)
2152 		flags |= LLI_CYCLIC | LLI_TERM_INT;
2153 
2154 	ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2155 				desc->lli_phy.src,
2156 				virt_to_phys(desc->lli_phy.src),
2157 				chan->src_def_cfg,
2158 				src_info, dst_info, flags);
2159 
2160 	ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2161 				desc->lli_phy.dst,
2162 				virt_to_phys(desc->lli_phy.dst),
2163 				chan->dst_def_cfg,
2164 				dst_info, src_info, flags);
2165 
2166 	dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2167 				   desc->lli_pool.size, DMA_TO_DEVICE);
2168 
2169 	return ret < 0 ? ret : 0;
2170 }
2171 
2172 static struct d40_desc *
2173 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2174 	      unsigned int sg_len, unsigned long dma_flags)
2175 {
2176 	struct stedma40_chan_cfg *cfg;
2177 	struct d40_desc *desc;
2178 	int ret;
2179 
2180 	desc = d40_desc_get(chan);
2181 	if (!desc)
2182 		return NULL;
2183 
2184 	cfg = &chan->dma_cfg;
2185 	desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2186 					cfg->dst_info.data_width);
2187 	if (desc->lli_len < 0) {
2188 		chan_err(chan, "Unaligned size\n");
2189 		goto free_desc;
2190 	}
2191 
2192 	ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2193 	if (ret < 0) {
2194 		chan_err(chan, "Could not allocate lli\n");
2195 		goto free_desc;
2196 	}
2197 
2198 	desc->lli_current = 0;
2199 	desc->txd.flags = dma_flags;
2200 	desc->txd.tx_submit = d40_tx_submit;
2201 
2202 	dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2203 
2204 	return desc;
2205  free_desc:
2206 	d40_desc_free(chan, desc);
2207 	return NULL;
2208 }
2209 
2210 static struct dma_async_tx_descriptor *
2211 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2212 	    struct scatterlist *sg_dst, unsigned int sg_len,
2213 	    enum dma_transfer_direction direction, unsigned long dma_flags)
2214 {
2215 	struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2216 	dma_addr_t src_dev_addr;
2217 	dma_addr_t dst_dev_addr;
2218 	struct d40_desc *desc;
2219 	unsigned long flags;
2220 	int ret;
2221 
2222 	if (!chan->phy_chan) {
2223 		chan_err(chan, "Cannot prepare unallocated channel\n");
2224 		return NULL;
2225 	}
2226 
2227 	d40_set_runtime_config_write(dchan, &chan->slave_config, direction);
2228 
2229 	spin_lock_irqsave(&chan->lock, flags);
2230 
2231 	desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2232 	if (desc == NULL)
2233 		goto unlock;
2234 
2235 	if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2236 		desc->cyclic = true;
2237 
2238 	src_dev_addr = 0;
2239 	dst_dev_addr = 0;
2240 	if (direction == DMA_DEV_TO_MEM)
2241 		src_dev_addr = chan->runtime_addr;
2242 	else if (direction == DMA_MEM_TO_DEV)
2243 		dst_dev_addr = chan->runtime_addr;
2244 
2245 	if (chan_is_logical(chan))
2246 		ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2247 				      sg_len, src_dev_addr, dst_dev_addr);
2248 	else
2249 		ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2250 				      sg_len, src_dev_addr, dst_dev_addr);
2251 
2252 	if (ret) {
2253 		chan_err(chan, "Failed to prepare %s sg job: %d\n",
2254 			 chan_is_logical(chan) ? "log" : "phy", ret);
2255 		goto free_desc;
2256 	}
2257 
2258 	/*
2259 	 * add descriptor to the prepare queue in order to be able
2260 	 * to free them later in terminate_all
2261 	 */
2262 	list_add_tail(&desc->node, &chan->prepare_queue);
2263 
2264 	spin_unlock_irqrestore(&chan->lock, flags);
2265 
2266 	return &desc->txd;
2267  free_desc:
2268 	d40_desc_free(chan, desc);
2269  unlock:
2270 	spin_unlock_irqrestore(&chan->lock, flags);
2271 	return NULL;
2272 }
2273 
2274 bool stedma40_filter(struct dma_chan *chan, void *data)
2275 {
2276 	struct stedma40_chan_cfg *info = data;
2277 	struct d40_chan *d40c =
2278 		container_of(chan, struct d40_chan, chan);
2279 	int err;
2280 
2281 	if (data) {
2282 		err = d40_validate_conf(d40c, info);
2283 		if (!err)
2284 			d40c->dma_cfg = *info;
2285 	} else
2286 		err = d40_config_memcpy(d40c);
2287 
2288 	if (!err)
2289 		d40c->configured = true;
2290 
2291 	return err == 0;
2292 }
2293 EXPORT_SYMBOL(stedma40_filter);
2294 
2295 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2296 {
2297 	bool realtime = d40c->dma_cfg.realtime;
2298 	bool highprio = d40c->dma_cfg.high_priority;
2299 	u32 rtreg;
2300 	u32 event = D40_TYPE_TO_EVENT(dev_type);
2301 	u32 group = D40_TYPE_TO_GROUP(dev_type);
2302 	u32 bit = BIT(event);
2303 	u32 prioreg;
2304 	struct d40_gen_dmac *dmac = &d40c->base->gen_dmac;
2305 
2306 	rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear;
2307 	/*
2308 	 * Due to a hardware bug, in some cases a logical channel triggered by
2309 	 * a high priority destination event line can generate extra packet
2310 	 * transactions.
2311 	 *
2312 	 * The workaround is to not set the high priority level for the
2313 	 * destination event lines that trigger logical channels.
2314 	 */
2315 	if (!src && chan_is_logical(d40c))
2316 		highprio = false;
2317 
2318 	prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear;
2319 
2320 	/* Destination event lines are stored in the upper halfword */
2321 	if (!src)
2322 		bit <<= 16;
2323 
2324 	writel(bit, d40c->base->virtbase + prioreg + group * 4);
2325 	writel(bit, d40c->base->virtbase + rtreg + group * 4);
2326 }
2327 
2328 static void d40_set_prio_realtime(struct d40_chan *d40c)
2329 {
2330 	if (d40c->base->rev < 3)
2331 		return;
2332 
2333 	if ((d40c->dma_cfg.dir ==  DMA_DEV_TO_MEM) ||
2334 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2335 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, true);
2336 
2337 	if ((d40c->dma_cfg.dir ==  DMA_MEM_TO_DEV) ||
2338 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2339 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, false);
2340 }
2341 
2342 #define D40_DT_FLAGS_MODE(flags)       ((flags >> 0) & 0x1)
2343 #define D40_DT_FLAGS_DIR(flags)        ((flags >> 1) & 0x1)
2344 #define D40_DT_FLAGS_BIG_ENDIAN(flags) ((flags >> 2) & 0x1)
2345 #define D40_DT_FLAGS_FIXED_CHAN(flags) ((flags >> 3) & 0x1)
2346 #define D40_DT_FLAGS_HIGH_PRIO(flags)  ((flags >> 4) & 0x1)
2347 
2348 static struct dma_chan *d40_xlate(struct of_phandle_args *dma_spec,
2349 				  struct of_dma *ofdma)
2350 {
2351 	struct stedma40_chan_cfg cfg;
2352 	dma_cap_mask_t cap;
2353 	u32 flags;
2354 
2355 	memset(&cfg, 0, sizeof(struct stedma40_chan_cfg));
2356 
2357 	dma_cap_zero(cap);
2358 	dma_cap_set(DMA_SLAVE, cap);
2359 
2360 	cfg.dev_type = dma_spec->args[0];
2361 	flags = dma_spec->args[2];
2362 
2363 	switch (D40_DT_FLAGS_MODE(flags)) {
2364 	case 0: cfg.mode = STEDMA40_MODE_LOGICAL; break;
2365 	case 1: cfg.mode = STEDMA40_MODE_PHYSICAL; break;
2366 	}
2367 
2368 	switch (D40_DT_FLAGS_DIR(flags)) {
2369 	case 0:
2370 		cfg.dir = DMA_MEM_TO_DEV;
2371 		cfg.dst_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2372 		break;
2373 	case 1:
2374 		cfg.dir = DMA_DEV_TO_MEM;
2375 		cfg.src_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2376 		break;
2377 	}
2378 
2379 	if (D40_DT_FLAGS_FIXED_CHAN(flags)) {
2380 		cfg.phy_channel = dma_spec->args[1];
2381 		cfg.use_fixed_channel = true;
2382 	}
2383 
2384 	if (D40_DT_FLAGS_HIGH_PRIO(flags))
2385 		cfg.high_priority = true;
2386 
2387 	return dma_request_channel(cap, stedma40_filter, &cfg);
2388 }
2389 
2390 /* DMA ENGINE functions */
2391 static int d40_alloc_chan_resources(struct dma_chan *chan)
2392 {
2393 	int err;
2394 	unsigned long flags;
2395 	struct d40_chan *d40c =
2396 		container_of(chan, struct d40_chan, chan);
2397 	bool is_free_phy;
2398 	spin_lock_irqsave(&d40c->lock, flags);
2399 
2400 	dma_cookie_init(chan);
2401 
2402 	/* If no dma configuration is set use default configuration (memcpy) */
2403 	if (!d40c->configured) {
2404 		err = d40_config_memcpy(d40c);
2405 		if (err) {
2406 			chan_err(d40c, "Failed to configure memcpy channel\n");
2407 			goto mark_last_busy;
2408 		}
2409 	}
2410 
2411 	err = d40_allocate_channel(d40c, &is_free_phy);
2412 	if (err) {
2413 		chan_err(d40c, "Failed to allocate channel\n");
2414 		d40c->configured = false;
2415 		goto mark_last_busy;
2416 	}
2417 
2418 	pm_runtime_get_sync(d40c->base->dev);
2419 
2420 	d40_set_prio_realtime(d40c);
2421 
2422 	if (chan_is_logical(d40c)) {
2423 		if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2424 			d40c->lcpa = d40c->base->lcpa_base +
2425 				d40c->dma_cfg.dev_type * D40_LCPA_CHAN_SIZE;
2426 		else
2427 			d40c->lcpa = d40c->base->lcpa_base +
2428 				d40c->dma_cfg.dev_type *
2429 				D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2430 
2431 		/* Unmask the Global Interrupt Mask. */
2432 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2433 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2434 	}
2435 
2436 	dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2437 		 chan_is_logical(d40c) ? "logical" : "physical",
2438 		 d40c->phy_chan->num,
2439 		 d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2440 
2441 
2442 	/*
2443 	 * Only write channel configuration to the DMA if the physical
2444 	 * resource is free. In case of multiple logical channels
2445 	 * on the same physical resource, only the first write is necessary.
2446 	 */
2447 	if (is_free_phy)
2448 		d40_config_write(d40c);
2449  mark_last_busy:
2450 	pm_runtime_mark_last_busy(d40c->base->dev);
2451 	pm_runtime_put_autosuspend(d40c->base->dev);
2452 	spin_unlock_irqrestore(&d40c->lock, flags);
2453 	return err;
2454 }
2455 
2456 static void d40_free_chan_resources(struct dma_chan *chan)
2457 {
2458 	struct d40_chan *d40c =
2459 		container_of(chan, struct d40_chan, chan);
2460 	int err;
2461 	unsigned long flags;
2462 
2463 	if (d40c->phy_chan == NULL) {
2464 		chan_err(d40c, "Cannot free unallocated channel\n");
2465 		return;
2466 	}
2467 
2468 	spin_lock_irqsave(&d40c->lock, flags);
2469 
2470 	err = d40_free_dma(d40c);
2471 
2472 	if (err)
2473 		chan_err(d40c, "Failed to free channel\n");
2474 	spin_unlock_irqrestore(&d40c->lock, flags);
2475 }
2476 
2477 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2478 						       dma_addr_t dst,
2479 						       dma_addr_t src,
2480 						       size_t size,
2481 						       unsigned long dma_flags)
2482 {
2483 	struct scatterlist dst_sg;
2484 	struct scatterlist src_sg;
2485 
2486 	sg_init_table(&dst_sg, 1);
2487 	sg_init_table(&src_sg, 1);
2488 
2489 	sg_dma_address(&dst_sg) = dst;
2490 	sg_dma_address(&src_sg) = src;
2491 
2492 	sg_dma_len(&dst_sg) = size;
2493 	sg_dma_len(&src_sg) = size;
2494 
2495 	return d40_prep_sg(chan, &src_sg, &dst_sg, 1,
2496 			   DMA_MEM_TO_MEM, dma_flags);
2497 }
2498 
2499 static struct dma_async_tx_descriptor *
2500 d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2501 		  unsigned int sg_len, enum dma_transfer_direction direction,
2502 		  unsigned long dma_flags, void *context)
2503 {
2504 	if (!is_slave_direction(direction))
2505 		return NULL;
2506 
2507 	return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2508 }
2509 
2510 static struct dma_async_tx_descriptor *
2511 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2512 		     size_t buf_len, size_t period_len,
2513 		     enum dma_transfer_direction direction, unsigned long flags)
2514 {
2515 	unsigned int periods = buf_len / period_len;
2516 	struct dma_async_tx_descriptor *txd;
2517 	struct scatterlist *sg;
2518 	int i;
2519 
2520 	sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2521 	if (!sg)
2522 		return NULL;
2523 
2524 	for (i = 0; i < periods; i++) {
2525 		sg_dma_address(&sg[i]) = dma_addr;
2526 		sg_dma_len(&sg[i]) = period_len;
2527 		dma_addr += period_len;
2528 	}
2529 
2530 	sg_chain(sg, periods + 1, sg);
2531 
2532 	txd = d40_prep_sg(chan, sg, sg, periods, direction,
2533 			  DMA_PREP_INTERRUPT);
2534 
2535 	kfree(sg);
2536 
2537 	return txd;
2538 }
2539 
2540 static enum dma_status d40_tx_status(struct dma_chan *chan,
2541 				     dma_cookie_t cookie,
2542 				     struct dma_tx_state *txstate)
2543 {
2544 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2545 	enum dma_status ret;
2546 
2547 	if (d40c->phy_chan == NULL) {
2548 		chan_err(d40c, "Cannot read status of unallocated channel\n");
2549 		return -EINVAL;
2550 	}
2551 
2552 	ret = dma_cookie_status(chan, cookie, txstate);
2553 	if (ret != DMA_COMPLETE && txstate)
2554 		dma_set_residue(txstate, stedma40_residue(chan));
2555 
2556 	if (d40_is_paused(d40c))
2557 		ret = DMA_PAUSED;
2558 
2559 	return ret;
2560 }
2561 
2562 static void d40_issue_pending(struct dma_chan *chan)
2563 {
2564 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2565 	unsigned long flags;
2566 
2567 	if (d40c->phy_chan == NULL) {
2568 		chan_err(d40c, "Channel is not allocated!\n");
2569 		return;
2570 	}
2571 
2572 	spin_lock_irqsave(&d40c->lock, flags);
2573 
2574 	list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2575 
2576 	/* Busy means that queued jobs are already being processed */
2577 	if (!d40c->busy)
2578 		(void) d40_queue_start(d40c);
2579 
2580 	spin_unlock_irqrestore(&d40c->lock, flags);
2581 }
2582 
2583 static int d40_terminate_all(struct dma_chan *chan)
2584 {
2585 	unsigned long flags;
2586 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2587 	int ret;
2588 
2589 	if (d40c->phy_chan == NULL) {
2590 		chan_err(d40c, "Channel is not allocated!\n");
2591 		return -EINVAL;
2592 	}
2593 
2594 	spin_lock_irqsave(&d40c->lock, flags);
2595 
2596 	pm_runtime_get_sync(d40c->base->dev);
2597 	ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2598 	if (ret)
2599 		chan_err(d40c, "Failed to stop channel\n");
2600 
2601 	d40_term_all(d40c);
2602 	pm_runtime_mark_last_busy(d40c->base->dev);
2603 	pm_runtime_put_autosuspend(d40c->base->dev);
2604 	if (d40c->busy) {
2605 		pm_runtime_mark_last_busy(d40c->base->dev);
2606 		pm_runtime_put_autosuspend(d40c->base->dev);
2607 	}
2608 	d40c->busy = false;
2609 
2610 	spin_unlock_irqrestore(&d40c->lock, flags);
2611 	return 0;
2612 }
2613 
2614 static int
2615 dma40_config_to_halfchannel(struct d40_chan *d40c,
2616 			    struct stedma40_half_channel_info *info,
2617 			    u32 maxburst)
2618 {
2619 	int psize;
2620 
2621 	if (chan_is_logical(d40c)) {
2622 		if (maxburst >= 16)
2623 			psize = STEDMA40_PSIZE_LOG_16;
2624 		else if (maxburst >= 8)
2625 			psize = STEDMA40_PSIZE_LOG_8;
2626 		else if (maxburst >= 4)
2627 			psize = STEDMA40_PSIZE_LOG_4;
2628 		else
2629 			psize = STEDMA40_PSIZE_LOG_1;
2630 	} else {
2631 		if (maxburst >= 16)
2632 			psize = STEDMA40_PSIZE_PHY_16;
2633 		else if (maxburst >= 8)
2634 			psize = STEDMA40_PSIZE_PHY_8;
2635 		else if (maxburst >= 4)
2636 			psize = STEDMA40_PSIZE_PHY_4;
2637 		else
2638 			psize = STEDMA40_PSIZE_PHY_1;
2639 	}
2640 
2641 	info->psize = psize;
2642 	info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2643 
2644 	return 0;
2645 }
2646 
2647 static int d40_set_runtime_config(struct dma_chan *chan,
2648 				  struct dma_slave_config *config)
2649 {
2650 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2651 
2652 	memcpy(&d40c->slave_config, config, sizeof(*config));
2653 
2654 	return 0;
2655 }
2656 
2657 /* Runtime reconfiguration extension */
2658 static int d40_set_runtime_config_write(struct dma_chan *chan,
2659 				  struct dma_slave_config *config,
2660 				  enum dma_transfer_direction direction)
2661 {
2662 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2663 	struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2664 	enum dma_slave_buswidth src_addr_width, dst_addr_width;
2665 	dma_addr_t config_addr;
2666 	u32 src_maxburst, dst_maxburst;
2667 	int ret;
2668 
2669 	if (d40c->phy_chan == NULL) {
2670 		chan_err(d40c, "Channel is not allocated!\n");
2671 		return -EINVAL;
2672 	}
2673 
2674 	src_addr_width = config->src_addr_width;
2675 	src_maxburst = config->src_maxburst;
2676 	dst_addr_width = config->dst_addr_width;
2677 	dst_maxburst = config->dst_maxburst;
2678 
2679 	if (direction == DMA_DEV_TO_MEM) {
2680 		config_addr = config->src_addr;
2681 
2682 		if (cfg->dir != DMA_DEV_TO_MEM)
2683 			dev_dbg(d40c->base->dev,
2684 				"channel was not configured for peripheral "
2685 				"to memory transfer (%d) overriding\n",
2686 				cfg->dir);
2687 		cfg->dir = DMA_DEV_TO_MEM;
2688 
2689 		/* Configure the memory side */
2690 		if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2691 			dst_addr_width = src_addr_width;
2692 		if (dst_maxburst == 0)
2693 			dst_maxburst = src_maxburst;
2694 
2695 	} else if (direction == DMA_MEM_TO_DEV) {
2696 		config_addr = config->dst_addr;
2697 
2698 		if (cfg->dir != DMA_MEM_TO_DEV)
2699 			dev_dbg(d40c->base->dev,
2700 				"channel was not configured for memory "
2701 				"to peripheral transfer (%d) overriding\n",
2702 				cfg->dir);
2703 		cfg->dir = DMA_MEM_TO_DEV;
2704 
2705 		/* Configure the memory side */
2706 		if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2707 			src_addr_width = dst_addr_width;
2708 		if (src_maxburst == 0)
2709 			src_maxburst = dst_maxburst;
2710 	} else {
2711 		dev_err(d40c->base->dev,
2712 			"unrecognized channel direction %d\n",
2713 			direction);
2714 		return -EINVAL;
2715 	}
2716 
2717 	if (config_addr <= 0) {
2718 		dev_err(d40c->base->dev, "no address supplied\n");
2719 		return -EINVAL;
2720 	}
2721 
2722 	if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2723 		dev_err(d40c->base->dev,
2724 			"src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2725 			src_maxburst,
2726 			src_addr_width,
2727 			dst_maxburst,
2728 			dst_addr_width);
2729 		return -EINVAL;
2730 	}
2731 
2732 	if (src_maxburst > 16) {
2733 		src_maxburst = 16;
2734 		dst_maxburst = src_maxburst * src_addr_width / dst_addr_width;
2735 	} else if (dst_maxburst > 16) {
2736 		dst_maxburst = 16;
2737 		src_maxburst = dst_maxburst * dst_addr_width / src_addr_width;
2738 	}
2739 
2740 	/* Only valid widths are; 1, 2, 4 and 8. */
2741 	if (src_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2742 	    src_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2743 	    dst_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2744 	    dst_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2745 	    !is_power_of_2(src_addr_width) ||
2746 	    !is_power_of_2(dst_addr_width))
2747 		return -EINVAL;
2748 
2749 	cfg->src_info.data_width = src_addr_width;
2750 	cfg->dst_info.data_width = dst_addr_width;
2751 
2752 	ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2753 					  src_maxburst);
2754 	if (ret)
2755 		return ret;
2756 
2757 	ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2758 					  dst_maxburst);
2759 	if (ret)
2760 		return ret;
2761 
2762 	/* Fill in register values */
2763 	if (chan_is_logical(d40c))
2764 		d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2765 	else
2766 		d40_phy_cfg(cfg, &d40c->src_def_cfg, &d40c->dst_def_cfg);
2767 
2768 	/* These settings will take precedence later */
2769 	d40c->runtime_addr = config_addr;
2770 	d40c->runtime_direction = direction;
2771 	dev_dbg(d40c->base->dev,
2772 		"configured channel %s for %s, data width %d/%d, "
2773 		"maxburst %d/%d elements, LE, no flow control\n",
2774 		dma_chan_name(chan),
2775 		(direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2776 		src_addr_width, dst_addr_width,
2777 		src_maxburst, dst_maxburst);
2778 
2779 	return 0;
2780 }
2781 
2782 /* Initialization functions */
2783 
2784 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2785 				 struct d40_chan *chans, int offset,
2786 				 int num_chans)
2787 {
2788 	int i = 0;
2789 	struct d40_chan *d40c;
2790 
2791 	INIT_LIST_HEAD(&dma->channels);
2792 
2793 	for (i = offset; i < offset + num_chans; i++) {
2794 		d40c = &chans[i];
2795 		d40c->base = base;
2796 		d40c->chan.device = dma;
2797 
2798 		spin_lock_init(&d40c->lock);
2799 
2800 		d40c->log_num = D40_PHY_CHAN;
2801 
2802 		INIT_LIST_HEAD(&d40c->done);
2803 		INIT_LIST_HEAD(&d40c->active);
2804 		INIT_LIST_HEAD(&d40c->queue);
2805 		INIT_LIST_HEAD(&d40c->pending_queue);
2806 		INIT_LIST_HEAD(&d40c->client);
2807 		INIT_LIST_HEAD(&d40c->prepare_queue);
2808 
2809 		tasklet_init(&d40c->tasklet, dma_tasklet,
2810 			     (unsigned long) d40c);
2811 
2812 		list_add_tail(&d40c->chan.device_node,
2813 			      &dma->channels);
2814 	}
2815 }
2816 
2817 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2818 {
2819 	if (dma_has_cap(DMA_SLAVE, dev->cap_mask)) {
2820 		dev->device_prep_slave_sg = d40_prep_slave_sg;
2821 		dev->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
2822 	}
2823 
2824 	if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2825 		dev->device_prep_dma_memcpy = d40_prep_memcpy;
2826 		dev->directions = BIT(DMA_MEM_TO_MEM);
2827 		/*
2828 		 * This controller can only access address at even
2829 		 * 32bit boundaries, i.e. 2^2
2830 		 */
2831 		dev->copy_align = DMAENGINE_ALIGN_4_BYTES;
2832 	}
2833 
2834 	if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2835 		dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2836 
2837 	dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2838 	dev->device_free_chan_resources = d40_free_chan_resources;
2839 	dev->device_issue_pending = d40_issue_pending;
2840 	dev->device_tx_status = d40_tx_status;
2841 	dev->device_config = d40_set_runtime_config;
2842 	dev->device_pause = d40_pause;
2843 	dev->device_resume = d40_resume;
2844 	dev->device_terminate_all = d40_terminate_all;
2845 	dev->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
2846 	dev->dev = base->dev;
2847 }
2848 
2849 static int __init d40_dmaengine_init(struct d40_base *base,
2850 				     int num_reserved_chans)
2851 {
2852 	int err ;
2853 
2854 	d40_chan_init(base, &base->dma_slave, base->log_chans,
2855 		      0, base->num_log_chans);
2856 
2857 	dma_cap_zero(base->dma_slave.cap_mask);
2858 	dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2859 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2860 
2861 	d40_ops_init(base, &base->dma_slave);
2862 
2863 	err = dmaenginem_async_device_register(&base->dma_slave);
2864 
2865 	if (err) {
2866 		d40_err(base->dev, "Failed to register slave channels\n");
2867 		goto exit;
2868 	}
2869 
2870 	d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2871 		      base->num_log_chans, base->num_memcpy_chans);
2872 
2873 	dma_cap_zero(base->dma_memcpy.cap_mask);
2874 	dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2875 
2876 	d40_ops_init(base, &base->dma_memcpy);
2877 
2878 	err = dmaenginem_async_device_register(&base->dma_memcpy);
2879 
2880 	if (err) {
2881 		d40_err(base->dev,
2882 			"Failed to register memcpy only channels\n");
2883 		goto exit;
2884 	}
2885 
2886 	d40_chan_init(base, &base->dma_both, base->phy_chans,
2887 		      0, num_reserved_chans);
2888 
2889 	dma_cap_zero(base->dma_both.cap_mask);
2890 	dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2891 	dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2892 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2893 
2894 	d40_ops_init(base, &base->dma_both);
2895 	err = dmaenginem_async_device_register(&base->dma_both);
2896 
2897 	if (err) {
2898 		d40_err(base->dev,
2899 			"Failed to register logical and physical capable channels\n");
2900 		goto exit;
2901 	}
2902 	return 0;
2903  exit:
2904 	return err;
2905 }
2906 
2907 /* Suspend resume functionality */
2908 #ifdef CONFIG_PM_SLEEP
2909 static int dma40_suspend(struct device *dev)
2910 {
2911 	struct d40_base *base = dev_get_drvdata(dev);
2912 	int ret;
2913 
2914 	ret = pm_runtime_force_suspend(dev);
2915 	if (ret)
2916 		return ret;
2917 
2918 	if (base->lcpa_regulator)
2919 		ret = regulator_disable(base->lcpa_regulator);
2920 	return ret;
2921 }
2922 
2923 static int dma40_resume(struct device *dev)
2924 {
2925 	struct d40_base *base = dev_get_drvdata(dev);
2926 	int ret = 0;
2927 
2928 	if (base->lcpa_regulator) {
2929 		ret = regulator_enable(base->lcpa_regulator);
2930 		if (ret)
2931 			return ret;
2932 	}
2933 
2934 	return pm_runtime_force_resume(dev);
2935 }
2936 #endif
2937 
2938 #ifdef CONFIG_PM
2939 static void dma40_backup(void __iomem *baseaddr, u32 *backup,
2940 			 u32 *regaddr, int num, bool save)
2941 {
2942 	int i;
2943 
2944 	for (i = 0; i < num; i++) {
2945 		void __iomem *addr = baseaddr + regaddr[i];
2946 
2947 		if (save)
2948 			backup[i] = readl_relaxed(addr);
2949 		else
2950 			writel_relaxed(backup[i], addr);
2951 	}
2952 }
2953 
2954 static void d40_save_restore_registers(struct d40_base *base, bool save)
2955 {
2956 	int i;
2957 
2958 	/* Save/Restore channel specific registers */
2959 	for (i = 0; i < base->num_phy_chans; i++) {
2960 		void __iomem *addr;
2961 		int idx;
2962 
2963 		if (base->phy_res[i].reserved)
2964 			continue;
2965 
2966 		addr = base->virtbase + D40_DREG_PCBASE + i * D40_DREG_PCDELTA;
2967 		idx = i * ARRAY_SIZE(d40_backup_regs_chan);
2968 
2969 		dma40_backup(addr, &base->reg_val_backup_chan[idx],
2970 			     d40_backup_regs_chan,
2971 			     ARRAY_SIZE(d40_backup_regs_chan),
2972 			     save);
2973 	}
2974 
2975 	/* Save/Restore global registers */
2976 	dma40_backup(base->virtbase, base->reg_val_backup,
2977 		     d40_backup_regs, ARRAY_SIZE(d40_backup_regs),
2978 		     save);
2979 
2980 	/* Save/Restore registers only existing on dma40 v3 and later */
2981 	if (base->gen_dmac.backup)
2982 		dma40_backup(base->virtbase, base->reg_val_backup_v4,
2983 			     base->gen_dmac.backup,
2984 			base->gen_dmac.backup_size,
2985 			save);
2986 }
2987 
2988 static int dma40_runtime_suspend(struct device *dev)
2989 {
2990 	struct d40_base *base = dev_get_drvdata(dev);
2991 
2992 	d40_save_restore_registers(base, true);
2993 
2994 	/* Don't disable/enable clocks for v1 due to HW bugs */
2995 	if (base->rev != 1)
2996 		writel_relaxed(base->gcc_pwr_off_mask,
2997 			       base->virtbase + D40_DREG_GCC);
2998 
2999 	return 0;
3000 }
3001 
3002 static int dma40_runtime_resume(struct device *dev)
3003 {
3004 	struct d40_base *base = dev_get_drvdata(dev);
3005 
3006 	d40_save_restore_registers(base, false);
3007 
3008 	writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
3009 		       base->virtbase + D40_DREG_GCC);
3010 	return 0;
3011 }
3012 #endif
3013 
3014 static const struct dev_pm_ops dma40_pm_ops = {
3015 	SET_LATE_SYSTEM_SLEEP_PM_OPS(dma40_suspend, dma40_resume)
3016 	SET_RUNTIME_PM_OPS(dma40_runtime_suspend,
3017 				dma40_runtime_resume,
3018 				NULL)
3019 };
3020 
3021 /* Initialization functions. */
3022 
3023 static int __init d40_phy_res_init(struct d40_base *base)
3024 {
3025 	int i;
3026 	int num_phy_chans_avail = 0;
3027 	u32 val[2];
3028 	int odd_even_bit = -2;
3029 	int gcc = D40_DREG_GCC_ENA;
3030 
3031 	val[0] = readl(base->virtbase + D40_DREG_PRSME);
3032 	val[1] = readl(base->virtbase + D40_DREG_PRSMO);
3033 
3034 	for (i = 0; i < base->num_phy_chans; i++) {
3035 		base->phy_res[i].num = i;
3036 		odd_even_bit += 2 * ((i % 2) == 0);
3037 		if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
3038 			/* Mark security only channels as occupied */
3039 			base->phy_res[i].allocated_src = D40_ALLOC_PHY;
3040 			base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
3041 			base->phy_res[i].reserved = true;
3042 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3043 						       D40_DREG_GCC_SRC);
3044 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3045 						       D40_DREG_GCC_DST);
3046 
3047 
3048 		} else {
3049 			base->phy_res[i].allocated_src = D40_ALLOC_FREE;
3050 			base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
3051 			base->phy_res[i].reserved = false;
3052 			num_phy_chans_avail++;
3053 		}
3054 		spin_lock_init(&base->phy_res[i].lock);
3055 	}
3056 
3057 	/* Mark disabled channels as occupied */
3058 	for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
3059 		int chan = base->plat_data->disabled_channels[i];
3060 
3061 		base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
3062 		base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
3063 		base->phy_res[chan].reserved = true;
3064 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3065 					       D40_DREG_GCC_SRC);
3066 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3067 					       D40_DREG_GCC_DST);
3068 		num_phy_chans_avail--;
3069 	}
3070 
3071 	/* Mark soft_lli channels */
3072 	for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) {
3073 		int chan = base->plat_data->soft_lli_chans[i];
3074 
3075 		base->phy_res[chan].use_soft_lli = true;
3076 	}
3077 
3078 	dev_info(base->dev, "%d of %d physical DMA channels available\n",
3079 		 num_phy_chans_avail, base->num_phy_chans);
3080 
3081 	/* Verify settings extended vs standard */
3082 	val[0] = readl(base->virtbase + D40_DREG_PRTYP);
3083 
3084 	for (i = 0; i < base->num_phy_chans; i++) {
3085 
3086 		if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
3087 		    (val[0] & 0x3) != 1)
3088 			dev_info(base->dev,
3089 				 "[%s] INFO: channel %d is misconfigured (%d)\n",
3090 				 __func__, i, val[0] & 0x3);
3091 
3092 		val[0] = val[0] >> 2;
3093 	}
3094 
3095 	/*
3096 	 * To keep things simple, Enable all clocks initially.
3097 	 * The clocks will get managed later post channel allocation.
3098 	 * The clocks for the event lines on which reserved channels exists
3099 	 * are not managed here.
3100 	 */
3101 	writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3102 	base->gcc_pwr_off_mask = gcc;
3103 
3104 	return num_phy_chans_avail;
3105 }
3106 
3107 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
3108 {
3109 	struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3110 	struct clk *clk;
3111 	void __iomem *virtbase;
3112 	struct resource *res;
3113 	struct d40_base *base;
3114 	int num_log_chans;
3115 	int num_phy_chans;
3116 	int num_memcpy_chans;
3117 	int clk_ret = -EINVAL;
3118 	int i;
3119 	u32 pid;
3120 	u32 cid;
3121 	u8 rev;
3122 
3123 	clk = clk_get(&pdev->dev, NULL);
3124 	if (IS_ERR(clk)) {
3125 		d40_err(&pdev->dev, "No matching clock found\n");
3126 		goto check_prepare_enabled;
3127 	}
3128 
3129 	clk_ret = clk_prepare_enable(clk);
3130 	if (clk_ret) {
3131 		d40_err(&pdev->dev, "Failed to prepare/enable clock\n");
3132 		goto disable_unprepare;
3133 	}
3134 
3135 	/* Get IO for DMAC base address */
3136 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
3137 	if (!res)
3138 		goto disable_unprepare;
3139 
3140 	if (request_mem_region(res->start, resource_size(res),
3141 			       D40_NAME " I/O base") == NULL)
3142 		goto release_region;
3143 
3144 	virtbase = ioremap(res->start, resource_size(res));
3145 	if (!virtbase)
3146 		goto release_region;
3147 
3148 	/* This is just a regular AMBA PrimeCell ID actually */
3149 	for (pid = 0, i = 0; i < 4; i++)
3150 		pid |= (readl(virtbase + resource_size(res) - 0x20 + 4 * i)
3151 			& 255) << (i * 8);
3152 	for (cid = 0, i = 0; i < 4; i++)
3153 		cid |= (readl(virtbase + resource_size(res) - 0x10 + 4 * i)
3154 			& 255) << (i * 8);
3155 
3156 	if (cid != AMBA_CID) {
3157 		d40_err(&pdev->dev, "Unknown hardware! No PrimeCell ID\n");
3158 		goto unmap_io;
3159 	}
3160 	if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
3161 		d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
3162 			AMBA_MANF_BITS(pid),
3163 			AMBA_VENDOR_ST);
3164 		goto unmap_io;
3165 	}
3166 	/*
3167 	 * HW revision:
3168 	 * DB8500ed has revision 0
3169 	 * ? has revision 1
3170 	 * DB8500v1 has revision 2
3171 	 * DB8500v2 has revision 3
3172 	 * AP9540v1 has revision 4
3173 	 * DB8540v1 has revision 4
3174 	 */
3175 	rev = AMBA_REV_BITS(pid);
3176 	if (rev < 2) {
3177 		d40_err(&pdev->dev, "hardware revision: %d is not supported", rev);
3178 		goto unmap_io;
3179 	}
3180 
3181 	/* The number of physical channels on this HW */
3182 	if (plat_data->num_of_phy_chans)
3183 		num_phy_chans = plat_data->num_of_phy_chans;
3184 	else
3185 		num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
3186 
3187 	/* The number of channels used for memcpy */
3188 	if (plat_data->num_of_memcpy_chans)
3189 		num_memcpy_chans = plat_data->num_of_memcpy_chans;
3190 	else
3191 		num_memcpy_chans = ARRAY_SIZE(dma40_memcpy_channels);
3192 
3193 	num_log_chans = num_phy_chans * D40_MAX_LOG_CHAN_PER_PHY;
3194 
3195 	dev_info(&pdev->dev,
3196 		 "hardware rev: %d @ %pa with %d physical and %d logical channels\n",
3197 		 rev, &res->start, num_phy_chans, num_log_chans);
3198 
3199 	base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
3200 		       (num_phy_chans + num_log_chans + num_memcpy_chans) *
3201 		       sizeof(struct d40_chan), GFP_KERNEL);
3202 
3203 	if (base == NULL)
3204 		goto unmap_io;
3205 
3206 	base->rev = rev;
3207 	base->clk = clk;
3208 	base->num_memcpy_chans = num_memcpy_chans;
3209 	base->num_phy_chans = num_phy_chans;
3210 	base->num_log_chans = num_log_chans;
3211 	base->phy_start = res->start;
3212 	base->phy_size = resource_size(res);
3213 	base->virtbase = virtbase;
3214 	base->plat_data = plat_data;
3215 	base->dev = &pdev->dev;
3216 	base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3217 	base->log_chans = &base->phy_chans[num_phy_chans];
3218 
3219 	if (base->plat_data->num_of_phy_chans == 14) {
3220 		base->gen_dmac.backup = d40_backup_regs_v4b;
3221 		base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
3222 		base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
3223 		base->gen_dmac.interrupt_clear = D40_DREG_CPCICR;
3224 		base->gen_dmac.realtime_en = D40_DREG_CRSEG1;
3225 		base->gen_dmac.realtime_clear = D40_DREG_CRCEG1;
3226 		base->gen_dmac.high_prio_en = D40_DREG_CPSEG1;
3227 		base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1;
3228 		base->gen_dmac.il = il_v4b;
3229 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4b);
3230 		base->gen_dmac.init_reg = dma_init_reg_v4b;
3231 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
3232 	} else {
3233 		if (base->rev >= 3) {
3234 			base->gen_dmac.backup = d40_backup_regs_v4a;
3235 			base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
3236 		}
3237 		base->gen_dmac.interrupt_en = D40_DREG_PCMIS;
3238 		base->gen_dmac.interrupt_clear = D40_DREG_PCICR;
3239 		base->gen_dmac.realtime_en = D40_DREG_RSEG1;
3240 		base->gen_dmac.realtime_clear = D40_DREG_RCEG1;
3241 		base->gen_dmac.high_prio_en = D40_DREG_PSEG1;
3242 		base->gen_dmac.high_prio_clear = D40_DREG_PCEG1;
3243 		base->gen_dmac.il = il_v4a;
3244 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4a);
3245 		base->gen_dmac.init_reg = dma_init_reg_v4a;
3246 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a);
3247 	}
3248 
3249 	base->phy_res = kcalloc(num_phy_chans,
3250 				sizeof(*base->phy_res),
3251 				GFP_KERNEL);
3252 	if (!base->phy_res)
3253 		goto free_base;
3254 
3255 	base->lookup_phy_chans = kcalloc(num_phy_chans,
3256 					 sizeof(*base->lookup_phy_chans),
3257 					 GFP_KERNEL);
3258 	if (!base->lookup_phy_chans)
3259 		goto free_phy_res;
3260 
3261 	base->lookup_log_chans = kcalloc(num_log_chans,
3262 					 sizeof(*base->lookup_log_chans),
3263 					 GFP_KERNEL);
3264 	if (!base->lookup_log_chans)
3265 		goto free_phy_chans;
3266 
3267 	base->reg_val_backup_chan = kmalloc_array(base->num_phy_chans,
3268 						  sizeof(d40_backup_regs_chan),
3269 						  GFP_KERNEL);
3270 	if (!base->reg_val_backup_chan)
3271 		goto free_log_chans;
3272 
3273 	base->lcla_pool.alloc_map = kcalloc(num_phy_chans
3274 					    * D40_LCLA_LINK_PER_EVENT_GRP,
3275 					    sizeof(*base->lcla_pool.alloc_map),
3276 					    GFP_KERNEL);
3277 	if (!base->lcla_pool.alloc_map)
3278 		goto free_backup_chan;
3279 
3280 	base->regs_interrupt = kmalloc_array(base->gen_dmac.il_size,
3281 					     sizeof(*base->regs_interrupt),
3282 					     GFP_KERNEL);
3283 	if (!base->regs_interrupt)
3284 		goto free_map;
3285 
3286 	base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3287 					    0, SLAB_HWCACHE_ALIGN,
3288 					    NULL);
3289 	if (base->desc_slab == NULL)
3290 		goto free_regs;
3291 
3292 
3293 	return base;
3294  free_regs:
3295 	kfree(base->regs_interrupt);
3296  free_map:
3297 	kfree(base->lcla_pool.alloc_map);
3298  free_backup_chan:
3299 	kfree(base->reg_val_backup_chan);
3300  free_log_chans:
3301 	kfree(base->lookup_log_chans);
3302  free_phy_chans:
3303 	kfree(base->lookup_phy_chans);
3304  free_phy_res:
3305 	kfree(base->phy_res);
3306  free_base:
3307 	kfree(base);
3308  unmap_io:
3309 	iounmap(virtbase);
3310  release_region:
3311 	release_mem_region(res->start, resource_size(res));
3312  check_prepare_enabled:
3313 	if (!clk_ret)
3314  disable_unprepare:
3315 		clk_disable_unprepare(clk);
3316 	if (!IS_ERR(clk))
3317 		clk_put(clk);
3318 	return NULL;
3319 }
3320 
3321 static void __init d40_hw_init(struct d40_base *base)
3322 {
3323 
3324 	int i;
3325 	u32 prmseo[2] = {0, 0};
3326 	u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3327 	u32 pcmis = 0;
3328 	u32 pcicr = 0;
3329 	struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg;
3330 	u32 reg_size = base->gen_dmac.init_reg_size;
3331 
3332 	for (i = 0; i < reg_size; i++)
3333 		writel(dma_init_reg[i].val,
3334 		       base->virtbase + dma_init_reg[i].reg);
3335 
3336 	/* Configure all our dma channels to default settings */
3337 	for (i = 0; i < base->num_phy_chans; i++) {
3338 
3339 		activeo[i % 2] = activeo[i % 2] << 2;
3340 
3341 		if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3342 		    == D40_ALLOC_PHY) {
3343 			activeo[i % 2] |= 3;
3344 			continue;
3345 		}
3346 
3347 		/* Enable interrupt # */
3348 		pcmis = (pcmis << 1) | 1;
3349 
3350 		/* Clear interrupt # */
3351 		pcicr = (pcicr << 1) | 1;
3352 
3353 		/* Set channel to physical mode */
3354 		prmseo[i % 2] = prmseo[i % 2] << 2;
3355 		prmseo[i % 2] |= 1;
3356 
3357 	}
3358 
3359 	writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3360 	writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3361 	writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3362 	writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3363 
3364 	/* Write which interrupt to enable */
3365 	writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en);
3366 
3367 	/* Write which interrupt to clear */
3368 	writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear);
3369 
3370 	/* These are __initdata and cannot be accessed after init */
3371 	base->gen_dmac.init_reg = NULL;
3372 	base->gen_dmac.init_reg_size = 0;
3373 }
3374 
3375 static int __init d40_lcla_allocate(struct d40_base *base)
3376 {
3377 	struct d40_lcla_pool *pool = &base->lcla_pool;
3378 	unsigned long *page_list;
3379 	int i, j;
3380 	int ret;
3381 
3382 	/*
3383 	 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3384 	 * To full fill this hardware requirement without wasting 256 kb
3385 	 * we allocate pages until we get an aligned one.
3386 	 */
3387 	page_list = kmalloc_array(MAX_LCLA_ALLOC_ATTEMPTS,
3388 				  sizeof(*page_list),
3389 				  GFP_KERNEL);
3390 	if (!page_list)
3391 		return -ENOMEM;
3392 
3393 	/* Calculating how many pages that are required */
3394 	base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3395 
3396 	for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3397 		page_list[i] = __get_free_pages(GFP_KERNEL,
3398 						base->lcla_pool.pages);
3399 		if (!page_list[i]) {
3400 
3401 			d40_err(base->dev, "Failed to allocate %d pages.\n",
3402 				base->lcla_pool.pages);
3403 			ret = -ENOMEM;
3404 
3405 			for (j = 0; j < i; j++)
3406 				free_pages(page_list[j], base->lcla_pool.pages);
3407 			goto free_page_list;
3408 		}
3409 
3410 		if ((virt_to_phys((void *)page_list[i]) &
3411 		     (LCLA_ALIGNMENT - 1)) == 0)
3412 			break;
3413 	}
3414 
3415 	for (j = 0; j < i; j++)
3416 		free_pages(page_list[j], base->lcla_pool.pages);
3417 
3418 	if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3419 		base->lcla_pool.base = (void *)page_list[i];
3420 	} else {
3421 		/*
3422 		 * After many attempts and no succees with finding the correct
3423 		 * alignment, try with allocating a big buffer.
3424 		 */
3425 		dev_warn(base->dev,
3426 			 "[%s] Failed to get %d pages @ 18 bit align.\n",
3427 			 __func__, base->lcla_pool.pages);
3428 		base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3429 							 base->num_phy_chans +
3430 							 LCLA_ALIGNMENT,
3431 							 GFP_KERNEL);
3432 		if (!base->lcla_pool.base_unaligned) {
3433 			ret = -ENOMEM;
3434 			goto free_page_list;
3435 		}
3436 
3437 		base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3438 						 LCLA_ALIGNMENT);
3439 	}
3440 
3441 	pool->dma_addr = dma_map_single(base->dev, pool->base,
3442 					SZ_1K * base->num_phy_chans,
3443 					DMA_TO_DEVICE);
3444 	if (dma_mapping_error(base->dev, pool->dma_addr)) {
3445 		pool->dma_addr = 0;
3446 		ret = -ENOMEM;
3447 		goto free_page_list;
3448 	}
3449 
3450 	writel(virt_to_phys(base->lcla_pool.base),
3451 	       base->virtbase + D40_DREG_LCLA);
3452 	ret = 0;
3453  free_page_list:
3454 	kfree(page_list);
3455 	return ret;
3456 }
3457 
3458 static int __init d40_of_probe(struct platform_device *pdev,
3459 			       struct device_node *np)
3460 {
3461 	struct stedma40_platform_data *pdata;
3462 	int num_phy = 0, num_memcpy = 0, num_disabled = 0;
3463 	const __be32 *list;
3464 
3465 	pdata = devm_kzalloc(&pdev->dev, sizeof(*pdata), GFP_KERNEL);
3466 	if (!pdata)
3467 		return -ENOMEM;
3468 
3469 	/* If absent this value will be obtained from h/w. */
3470 	of_property_read_u32(np, "dma-channels", &num_phy);
3471 	if (num_phy > 0)
3472 		pdata->num_of_phy_chans = num_phy;
3473 
3474 	list = of_get_property(np, "memcpy-channels", &num_memcpy);
3475 	num_memcpy /= sizeof(*list);
3476 
3477 	if (num_memcpy > D40_MEMCPY_MAX_CHANS || num_memcpy <= 0) {
3478 		d40_err(&pdev->dev,
3479 			"Invalid number of memcpy channels specified (%d)\n",
3480 			num_memcpy);
3481 		return -EINVAL;
3482 	}
3483 	pdata->num_of_memcpy_chans = num_memcpy;
3484 
3485 	of_property_read_u32_array(np, "memcpy-channels",
3486 				   dma40_memcpy_channels,
3487 				   num_memcpy);
3488 
3489 	list = of_get_property(np, "disabled-channels", &num_disabled);
3490 	num_disabled /= sizeof(*list);
3491 
3492 	if (num_disabled >= STEDMA40_MAX_PHYS || num_disabled < 0) {
3493 		d40_err(&pdev->dev,
3494 			"Invalid number of disabled channels specified (%d)\n",
3495 			num_disabled);
3496 		return -EINVAL;
3497 	}
3498 
3499 	of_property_read_u32_array(np, "disabled-channels",
3500 				   pdata->disabled_channels,
3501 				   num_disabled);
3502 	pdata->disabled_channels[num_disabled] = -1;
3503 
3504 	pdev->dev.platform_data = pdata;
3505 
3506 	return 0;
3507 }
3508 
3509 static int __init d40_probe(struct platform_device *pdev)
3510 {
3511 	struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3512 	struct device_node *np = pdev->dev.of_node;
3513 	int ret = -ENOENT;
3514 	struct d40_base *base;
3515 	struct resource *res;
3516 	int num_reserved_chans;
3517 	u32 val;
3518 
3519 	if (!plat_data) {
3520 		if (np) {
3521 			if (d40_of_probe(pdev, np)) {
3522 				ret = -ENOMEM;
3523 				goto report_failure;
3524 			}
3525 		} else {
3526 			d40_err(&pdev->dev, "No pdata or Device Tree provided\n");
3527 			goto report_failure;
3528 		}
3529 	}
3530 
3531 	base = d40_hw_detect_init(pdev);
3532 	if (!base)
3533 		goto report_failure;
3534 
3535 	num_reserved_chans = d40_phy_res_init(base);
3536 
3537 	platform_set_drvdata(pdev, base);
3538 
3539 	spin_lock_init(&base->interrupt_lock);
3540 	spin_lock_init(&base->execmd_lock);
3541 
3542 	/* Get IO for logical channel parameter address */
3543 	res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
3544 	if (!res) {
3545 		ret = -ENOENT;
3546 		d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
3547 		goto destroy_cache;
3548 	}
3549 	base->lcpa_size = resource_size(res);
3550 	base->phy_lcpa = res->start;
3551 
3552 	if (request_mem_region(res->start, resource_size(res),
3553 			       D40_NAME " I/O lcpa") == NULL) {
3554 		ret = -EBUSY;
3555 		d40_err(&pdev->dev, "Failed to request LCPA region %pR\n", res);
3556 		goto destroy_cache;
3557 	}
3558 
3559 	/* We make use of ESRAM memory for this. */
3560 	val = readl(base->virtbase + D40_DREG_LCPA);
3561 	if (res->start != val && val != 0) {
3562 		dev_warn(&pdev->dev,
3563 			 "[%s] Mismatch LCPA dma 0x%x, def %pa\n",
3564 			 __func__, val, &res->start);
3565 	} else
3566 		writel(res->start, base->virtbase + D40_DREG_LCPA);
3567 
3568 	base->lcpa_base = ioremap(res->start, resource_size(res));
3569 	if (!base->lcpa_base) {
3570 		ret = -ENOMEM;
3571 		d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
3572 		goto destroy_cache;
3573 	}
3574 	/* If lcla has to be located in ESRAM we don't need to allocate */
3575 	if (base->plat_data->use_esram_lcla) {
3576 		res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3577 							"lcla_esram");
3578 		if (!res) {
3579 			ret = -ENOENT;
3580 			d40_err(&pdev->dev,
3581 				"No \"lcla_esram\" memory resource\n");
3582 			goto destroy_cache;
3583 		}
3584 		base->lcla_pool.base = ioremap(res->start,
3585 						resource_size(res));
3586 		if (!base->lcla_pool.base) {
3587 			ret = -ENOMEM;
3588 			d40_err(&pdev->dev, "Failed to ioremap LCLA region\n");
3589 			goto destroy_cache;
3590 		}
3591 		writel(res->start, base->virtbase + D40_DREG_LCLA);
3592 
3593 	} else {
3594 		ret = d40_lcla_allocate(base);
3595 		if (ret) {
3596 			d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
3597 			goto destroy_cache;
3598 		}
3599 	}
3600 
3601 	spin_lock_init(&base->lcla_pool.lock);
3602 
3603 	base->irq = platform_get_irq(pdev, 0);
3604 
3605 	ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3606 	if (ret) {
3607 		d40_err(&pdev->dev, "No IRQ defined\n");
3608 		goto destroy_cache;
3609 	}
3610 
3611 	if (base->plat_data->use_esram_lcla) {
3612 
3613 		base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3614 		if (IS_ERR(base->lcpa_regulator)) {
3615 			d40_err(&pdev->dev, "Failed to get lcpa_regulator\n");
3616 			ret = PTR_ERR(base->lcpa_regulator);
3617 			base->lcpa_regulator = NULL;
3618 			goto destroy_cache;
3619 		}
3620 
3621 		ret = regulator_enable(base->lcpa_regulator);
3622 		if (ret) {
3623 			d40_err(&pdev->dev,
3624 				"Failed to enable lcpa_regulator\n");
3625 			regulator_put(base->lcpa_regulator);
3626 			base->lcpa_regulator = NULL;
3627 			goto destroy_cache;
3628 		}
3629 	}
3630 
3631 	writel_relaxed(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3632 
3633 	pm_runtime_irq_safe(base->dev);
3634 	pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3635 	pm_runtime_use_autosuspend(base->dev);
3636 	pm_runtime_mark_last_busy(base->dev);
3637 	pm_runtime_set_active(base->dev);
3638 	pm_runtime_enable(base->dev);
3639 
3640 	ret = d40_dmaengine_init(base, num_reserved_chans);
3641 	if (ret)
3642 		goto destroy_cache;
3643 
3644 	base->dev->dma_parms = &base->dma_parms;
3645 	ret = dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
3646 	if (ret) {
3647 		d40_err(&pdev->dev, "Failed to set dma max seg size\n");
3648 		goto destroy_cache;
3649 	}
3650 
3651 	d40_hw_init(base);
3652 
3653 	if (np) {
3654 		ret = of_dma_controller_register(np, d40_xlate, NULL);
3655 		if (ret)
3656 			dev_err(&pdev->dev,
3657 				"could not register of_dma_controller\n");
3658 	}
3659 
3660 	dev_info(base->dev, "initialized\n");
3661 	return 0;
3662  destroy_cache:
3663 	kmem_cache_destroy(base->desc_slab);
3664 	if (base->virtbase)
3665 		iounmap(base->virtbase);
3666 
3667 	if (base->lcla_pool.base && base->plat_data->use_esram_lcla) {
3668 		iounmap(base->lcla_pool.base);
3669 		base->lcla_pool.base = NULL;
3670 	}
3671 
3672 	if (base->lcla_pool.dma_addr)
3673 		dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3674 				 SZ_1K * base->num_phy_chans,
3675 				 DMA_TO_DEVICE);
3676 
3677 	if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3678 		free_pages((unsigned long)base->lcla_pool.base,
3679 			   base->lcla_pool.pages);
3680 
3681 	kfree(base->lcla_pool.base_unaligned);
3682 
3683 	if (base->phy_lcpa)
3684 		release_mem_region(base->phy_lcpa,
3685 				   base->lcpa_size);
3686 	if (base->phy_start)
3687 		release_mem_region(base->phy_start,
3688 				   base->phy_size);
3689 	if (base->clk) {
3690 		clk_disable_unprepare(base->clk);
3691 		clk_put(base->clk);
3692 	}
3693 
3694 	if (base->lcpa_regulator) {
3695 		regulator_disable(base->lcpa_regulator);
3696 		regulator_put(base->lcpa_regulator);
3697 	}
3698 
3699 	kfree(base->lcla_pool.alloc_map);
3700 	kfree(base->lookup_log_chans);
3701 	kfree(base->lookup_phy_chans);
3702 	kfree(base->phy_res);
3703 	kfree(base);
3704  report_failure:
3705 	d40_err(&pdev->dev, "probe failed\n");
3706 	return ret;
3707 }
3708 
3709 static const struct of_device_id d40_match[] = {
3710         { .compatible = "stericsson,dma40", },
3711         {}
3712 };
3713 
3714 static struct platform_driver d40_driver = {
3715 	.driver = {
3716 		.name  = D40_NAME,
3717 		.pm = &dma40_pm_ops,
3718 		.of_match_table = d40_match,
3719 	},
3720 };
3721 
3722 static int __init stedma40_init(void)
3723 {
3724 	return platform_driver_probe(&d40_driver, d40_probe);
3725 }
3726 subsys_initcall(stedma40_init);
3727