xref: /linux/drivers/dma/ste_dma40.c (revision 24168c5e6dfbdd5b414f048f47f75d64533296ca)
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_address.h>
23 #include <linux/of_dma.h>
24 #include <linux/amba/bus.h>
25 #include <linux/regulator/consumer.h>
26 
27 #include "dmaengine.h"
28 #include "ste_dma40.h"
29 #include "ste_dma40_ll.h"
30 
31 /**
32  * struct stedma40_platform_data - Configuration struct for the dma device.
33  *
34  * @disabled_channels: A vector, ending with -1, that marks physical channels
35  * that are for different reasons not available for the driver.
36  * @soft_lli_chans: A vector, that marks physical channels will use LLI by SW
37  * which avoids HW bug that exists in some versions of the controller.
38  * SoftLLI introduces relink overhead that could impact performance for
39  * certain use cases.
40  * @num_of_soft_lli_chans: The number of channels that needs to be configured
41  * to use SoftLLI.
42  * @use_esram_lcla: flag for mapping the lcla into esram region
43  * @num_of_memcpy_chans: The number of channels reserved for memcpy.
44  * @num_of_phy_chans: The number of physical channels implemented in HW.
45  * 0 means reading the number of channels from DMA HW but this is only valid
46  * for 'multiple of 4' channels, like 8.
47  */
48 struct stedma40_platform_data {
49 	int				 disabled_channels[STEDMA40_MAX_PHYS];
50 	int				*soft_lli_chans;
51 	int				 num_of_soft_lli_chans;
52 	bool				 use_esram_lcla;
53 	int				 num_of_memcpy_chans;
54 	int				 num_of_phy_chans;
55 };
56 
57 #define D40_NAME "dma40"
58 
59 #define D40_PHY_CHAN -1
60 
61 /* For masking out/in 2 bit channel positions */
62 #define D40_CHAN_POS(chan)  (2 * (chan / 2))
63 #define D40_CHAN_POS_MASK(chan) (0x3 << D40_CHAN_POS(chan))
64 
65 /* Maximum iterations taken before giving up suspending a channel */
66 #define D40_SUSPEND_MAX_IT 500
67 
68 /* Milliseconds */
69 #define DMA40_AUTOSUSPEND_DELAY	100
70 
71 /* Hardware requirement on LCLA alignment */
72 #define LCLA_ALIGNMENT 0x40000
73 
74 /* Max number of links per event group */
75 #define D40_LCLA_LINK_PER_EVENT_GRP 128
76 #define D40_LCLA_END D40_LCLA_LINK_PER_EVENT_GRP
77 
78 /* Max number of logical channels per physical channel */
79 #define D40_MAX_LOG_CHAN_PER_PHY 32
80 
81 /* Attempts before giving up to trying to get pages that are aligned */
82 #define MAX_LCLA_ALLOC_ATTEMPTS 256
83 
84 /* Bit markings for allocation map */
85 #define D40_ALLOC_FREE		BIT(31)
86 #define D40_ALLOC_PHY		BIT(30)
87 #define D40_ALLOC_LOG_FREE	0
88 
89 #define D40_MEMCPY_MAX_CHANS	8
90 
91 /* Reserved event lines for memcpy only. */
92 #define DB8500_DMA_MEMCPY_EV_0	51
93 #define DB8500_DMA_MEMCPY_EV_1	56
94 #define DB8500_DMA_MEMCPY_EV_2	57
95 #define DB8500_DMA_MEMCPY_EV_3	58
96 #define DB8500_DMA_MEMCPY_EV_4	59
97 #define DB8500_DMA_MEMCPY_EV_5	60
98 
99 static int dma40_memcpy_channels[] = {
100 	DB8500_DMA_MEMCPY_EV_0,
101 	DB8500_DMA_MEMCPY_EV_1,
102 	DB8500_DMA_MEMCPY_EV_2,
103 	DB8500_DMA_MEMCPY_EV_3,
104 	DB8500_DMA_MEMCPY_EV_4,
105 	DB8500_DMA_MEMCPY_EV_5,
106 };
107 
108 /* Default configuration for physical memcpy */
109 static const struct stedma40_chan_cfg dma40_memcpy_conf_phy = {
110 	.mode = STEDMA40_MODE_PHYSICAL,
111 	.dir = DMA_MEM_TO_MEM,
112 
113 	.src_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
114 	.src_info.psize = STEDMA40_PSIZE_PHY_1,
115 	.src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
116 
117 	.dst_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
118 	.dst_info.psize = STEDMA40_PSIZE_PHY_1,
119 	.dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
120 };
121 
122 /* Default configuration for logical memcpy */
123 static const struct stedma40_chan_cfg dma40_memcpy_conf_log = {
124 	.mode = STEDMA40_MODE_LOGICAL,
125 	.dir = DMA_MEM_TO_MEM,
126 
127 	.src_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
128 	.src_info.psize = STEDMA40_PSIZE_LOG_1,
129 	.src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
130 
131 	.dst_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
132 	.dst_info.psize = STEDMA40_PSIZE_LOG_1,
133 	.dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
134 };
135 
136 /**
137  * enum d40_command - The different commands and/or statuses.
138  *
139  * @D40_DMA_STOP: DMA channel command STOP or status STOPPED,
140  * @D40_DMA_RUN: The DMA channel is RUNNING of the command RUN.
141  * @D40_DMA_SUSPEND_REQ: Request the DMA to SUSPEND as soon as possible.
142  * @D40_DMA_SUSPENDED: The DMA channel is SUSPENDED.
143  */
144 enum d40_command {
145 	D40_DMA_STOP		= 0,
146 	D40_DMA_RUN		= 1,
147 	D40_DMA_SUSPEND_REQ	= 2,
148 	D40_DMA_SUSPENDED	= 3
149 };
150 
151 /*
152  * enum d40_events - The different Event Enables for the event lines.
153  *
154  * @D40_DEACTIVATE_EVENTLINE: De-activate Event line, stopping the logical chan.
155  * @D40_ACTIVATE_EVENTLINE: Activate the Event line, to start a logical chan.
156  * @D40_SUSPEND_REQ_EVENTLINE: Requesting for suspending a event line.
157  * @D40_ROUND_EVENTLINE: Status check for event line.
158  */
159 
160 enum d40_events {
161 	D40_DEACTIVATE_EVENTLINE	= 0,
162 	D40_ACTIVATE_EVENTLINE		= 1,
163 	D40_SUSPEND_REQ_EVENTLINE	= 2,
164 	D40_ROUND_EVENTLINE		= 3
165 };
166 
167 /*
168  * These are the registers that has to be saved and later restored
169  * when the DMA hw is powered off.
170  * TODO: Add save/restore of D40_DREG_GCC on dma40 v3 or later, if that works.
171  */
172 static __maybe_unused u32 d40_backup_regs[] = {
173 	D40_DREG_LCPA,
174 	D40_DREG_LCLA,
175 	D40_DREG_PRMSE,
176 	D40_DREG_PRMSO,
177 	D40_DREG_PRMOE,
178 	D40_DREG_PRMOO,
179 };
180 
181 #define BACKUP_REGS_SZ ARRAY_SIZE(d40_backup_regs)
182 
183 /*
184  * since 9540 and 8540 has the same HW revision
185  * use v4a for 9540 or earlier
186  * use v4b for 8540 or later
187  * HW revision:
188  * DB8500ed has revision 0
189  * DB8500v1 has revision 2
190  * DB8500v2 has revision 3
191  * AP9540v1 has revision 4
192  * DB8540v1 has revision 4
193  * TODO: Check if all these registers have to be saved/restored on dma40 v4a
194  */
195 static u32 d40_backup_regs_v4a[] = {
196 	D40_DREG_PSEG1,
197 	D40_DREG_PSEG2,
198 	D40_DREG_PSEG3,
199 	D40_DREG_PSEG4,
200 	D40_DREG_PCEG1,
201 	D40_DREG_PCEG2,
202 	D40_DREG_PCEG3,
203 	D40_DREG_PCEG4,
204 	D40_DREG_RSEG1,
205 	D40_DREG_RSEG2,
206 	D40_DREG_RSEG3,
207 	D40_DREG_RSEG4,
208 	D40_DREG_RCEG1,
209 	D40_DREG_RCEG2,
210 	D40_DREG_RCEG3,
211 	D40_DREG_RCEG4,
212 };
213 
214 #define BACKUP_REGS_SZ_V4A ARRAY_SIZE(d40_backup_regs_v4a)
215 
216 static u32 d40_backup_regs_v4b[] = {
217 	D40_DREG_CPSEG1,
218 	D40_DREG_CPSEG2,
219 	D40_DREG_CPSEG3,
220 	D40_DREG_CPSEG4,
221 	D40_DREG_CPSEG5,
222 	D40_DREG_CPCEG1,
223 	D40_DREG_CPCEG2,
224 	D40_DREG_CPCEG3,
225 	D40_DREG_CPCEG4,
226 	D40_DREG_CPCEG5,
227 	D40_DREG_CRSEG1,
228 	D40_DREG_CRSEG2,
229 	D40_DREG_CRSEG3,
230 	D40_DREG_CRSEG4,
231 	D40_DREG_CRSEG5,
232 	D40_DREG_CRCEG1,
233 	D40_DREG_CRCEG2,
234 	D40_DREG_CRCEG3,
235 	D40_DREG_CRCEG4,
236 	D40_DREG_CRCEG5,
237 };
238 
239 #define BACKUP_REGS_SZ_V4B ARRAY_SIZE(d40_backup_regs_v4b)
240 
241 static __maybe_unused u32 d40_backup_regs_chan[] = {
242 	D40_CHAN_REG_SSCFG,
243 	D40_CHAN_REG_SSELT,
244 	D40_CHAN_REG_SSPTR,
245 	D40_CHAN_REG_SSLNK,
246 	D40_CHAN_REG_SDCFG,
247 	D40_CHAN_REG_SDELT,
248 	D40_CHAN_REG_SDPTR,
249 	D40_CHAN_REG_SDLNK,
250 };
251 
252 #define BACKUP_REGS_SZ_MAX ((BACKUP_REGS_SZ_V4A > BACKUP_REGS_SZ_V4B) ? \
253 			     BACKUP_REGS_SZ_V4A : BACKUP_REGS_SZ_V4B)
254 
255 /**
256  * struct d40_interrupt_lookup - lookup table for interrupt handler
257  *
258  * @src: Interrupt mask register.
259  * @clr: Interrupt clear register.
260  * @is_error: true if this is an error interrupt.
261  * @offset: start delta in the lookup_log_chans in d40_base. If equals to
262  * D40_PHY_CHAN, the lookup_phy_chans shall be used instead.
263  */
264 struct d40_interrupt_lookup {
265 	u32 src;
266 	u32 clr;
267 	bool is_error;
268 	int offset;
269 };
270 
271 
272 static struct d40_interrupt_lookup il_v4a[] = {
273 	{D40_DREG_LCTIS0, D40_DREG_LCICR0, false,  0},
274 	{D40_DREG_LCTIS1, D40_DREG_LCICR1, false, 32},
275 	{D40_DREG_LCTIS2, D40_DREG_LCICR2, false, 64},
276 	{D40_DREG_LCTIS3, D40_DREG_LCICR3, false, 96},
277 	{D40_DREG_LCEIS0, D40_DREG_LCICR0, true,   0},
278 	{D40_DREG_LCEIS1, D40_DREG_LCICR1, true,  32},
279 	{D40_DREG_LCEIS2, D40_DREG_LCICR2, true,  64},
280 	{D40_DREG_LCEIS3, D40_DREG_LCICR3, true,  96},
281 	{D40_DREG_PCTIS,  D40_DREG_PCICR,  false, D40_PHY_CHAN},
282 	{D40_DREG_PCEIS,  D40_DREG_PCICR,  true,  D40_PHY_CHAN},
283 };
284 
285 static struct d40_interrupt_lookup il_v4b[] = {
286 	{D40_DREG_CLCTIS1, D40_DREG_CLCICR1, false,  0},
287 	{D40_DREG_CLCTIS2, D40_DREG_CLCICR2, false, 32},
288 	{D40_DREG_CLCTIS3, D40_DREG_CLCICR3, false, 64},
289 	{D40_DREG_CLCTIS4, D40_DREG_CLCICR4, false, 96},
290 	{D40_DREG_CLCTIS5, D40_DREG_CLCICR5, false, 128},
291 	{D40_DREG_CLCEIS1, D40_DREG_CLCICR1, true,   0},
292 	{D40_DREG_CLCEIS2, D40_DREG_CLCICR2, true,  32},
293 	{D40_DREG_CLCEIS3, D40_DREG_CLCICR3, true,  64},
294 	{D40_DREG_CLCEIS4, D40_DREG_CLCICR4, true,  96},
295 	{D40_DREG_CLCEIS5, D40_DREG_CLCICR5, true,  128},
296 	{D40_DREG_CPCTIS,  D40_DREG_CPCICR,  false, D40_PHY_CHAN},
297 	{D40_DREG_CPCEIS,  D40_DREG_CPCICR,  true,  D40_PHY_CHAN},
298 };
299 
300 /**
301  * struct d40_reg_val - simple lookup struct
302  *
303  * @reg: The register.
304  * @val: The value that belongs to the register in reg.
305  */
306 struct d40_reg_val {
307 	unsigned int reg;
308 	unsigned int val;
309 };
310 
311 static __initdata struct d40_reg_val dma_init_reg_v4a[] = {
312 	/* Clock every part of the DMA block from start */
313 	{ .reg = D40_DREG_GCC,    .val = D40_DREG_GCC_ENABLE_ALL},
314 
315 	/* Interrupts on all logical channels */
316 	{ .reg = D40_DREG_LCMIS0, .val = 0xFFFFFFFF},
317 	{ .reg = D40_DREG_LCMIS1, .val = 0xFFFFFFFF},
318 	{ .reg = D40_DREG_LCMIS2, .val = 0xFFFFFFFF},
319 	{ .reg = D40_DREG_LCMIS3, .val = 0xFFFFFFFF},
320 	{ .reg = D40_DREG_LCICR0, .val = 0xFFFFFFFF},
321 	{ .reg = D40_DREG_LCICR1, .val = 0xFFFFFFFF},
322 	{ .reg = D40_DREG_LCICR2, .val = 0xFFFFFFFF},
323 	{ .reg = D40_DREG_LCICR3, .val = 0xFFFFFFFF},
324 	{ .reg = D40_DREG_LCTIS0, .val = 0xFFFFFFFF},
325 	{ .reg = D40_DREG_LCTIS1, .val = 0xFFFFFFFF},
326 	{ .reg = D40_DREG_LCTIS2, .val = 0xFFFFFFFF},
327 	{ .reg = D40_DREG_LCTIS3, .val = 0xFFFFFFFF}
328 };
329 static __initdata struct d40_reg_val dma_init_reg_v4b[] = {
330 	/* Clock every part of the DMA block from start */
331 	{ .reg = D40_DREG_GCC,    .val = D40_DREG_GCC_ENABLE_ALL},
332 
333 	/* Interrupts on all logical channels */
334 	{ .reg = D40_DREG_CLCMIS1, .val = 0xFFFFFFFF},
335 	{ .reg = D40_DREG_CLCMIS2, .val = 0xFFFFFFFF},
336 	{ .reg = D40_DREG_CLCMIS3, .val = 0xFFFFFFFF},
337 	{ .reg = D40_DREG_CLCMIS4, .val = 0xFFFFFFFF},
338 	{ .reg = D40_DREG_CLCMIS5, .val = 0xFFFFFFFF},
339 	{ .reg = D40_DREG_CLCICR1, .val = 0xFFFFFFFF},
340 	{ .reg = D40_DREG_CLCICR2, .val = 0xFFFFFFFF},
341 	{ .reg = D40_DREG_CLCICR3, .val = 0xFFFFFFFF},
342 	{ .reg = D40_DREG_CLCICR4, .val = 0xFFFFFFFF},
343 	{ .reg = D40_DREG_CLCICR5, .val = 0xFFFFFFFF},
344 	{ .reg = D40_DREG_CLCTIS1, .val = 0xFFFFFFFF},
345 	{ .reg = D40_DREG_CLCTIS2, .val = 0xFFFFFFFF},
346 	{ .reg = D40_DREG_CLCTIS3, .val = 0xFFFFFFFF},
347 	{ .reg = D40_DREG_CLCTIS4, .val = 0xFFFFFFFF},
348 	{ .reg = D40_DREG_CLCTIS5, .val = 0xFFFFFFFF}
349 };
350 
351 /**
352  * struct d40_lli_pool - Structure for keeping LLIs in memory
353  *
354  * @base: Pointer to memory area when the pre_alloc_lli's are not large
355  * enough, IE bigger than the most common case, 1 dst and 1 src. NULL if
356  * pre_alloc_lli is used.
357  * @dma_addr: DMA address, if mapped
358  * @size: The size in bytes of the memory at base or the size of pre_alloc_lli.
359  * @pre_alloc_lli: Pre allocated area for the most common case of transfers,
360  * one buffer to one buffer.
361  */
362 struct d40_lli_pool {
363 	void	*base;
364 	int	 size;
365 	dma_addr_t	dma_addr;
366 	/* Space for dst and src, plus an extra for padding */
367 	u8	 pre_alloc_lli[3 * sizeof(struct d40_phy_lli)];
368 };
369 
370 /**
371  * struct d40_desc - A descriptor is one DMA job.
372  *
373  * @lli_phy: LLI settings for physical channel. Both src and dst=
374  * points into the lli_pool, to base if lli_len > 1 or to pre_alloc_lli if
375  * lli_len equals one.
376  * @lli_log: Same as above but for logical channels.
377  * @lli_pool: The pool with two entries pre-allocated.
378  * @lli_len: Number of llis of current descriptor.
379  * @lli_current: Number of transferred llis.
380  * @lcla_alloc: Number of LCLA entries allocated.
381  * @txd: DMA engine struct. Used for among other things for communication
382  * during a transfer.
383  * @node: List entry.
384  * @is_in_client_list: true if the client owns this descriptor.
385  * @cyclic: true if this is a cyclic job
386  *
387  * This descriptor is used for both logical and physical transfers.
388  */
389 struct d40_desc {
390 	/* LLI physical */
391 	struct d40_phy_lli_bidir	 lli_phy;
392 	/* LLI logical */
393 	struct d40_log_lli_bidir	 lli_log;
394 
395 	struct d40_lli_pool		 lli_pool;
396 	int				 lli_len;
397 	int				 lli_current;
398 	int				 lcla_alloc;
399 
400 	struct dma_async_tx_descriptor	 txd;
401 	struct list_head		 node;
402 
403 	bool				 is_in_client_list;
404 	bool				 cyclic;
405 };
406 
407 /**
408  * struct d40_lcla_pool - LCLA pool settings and data.
409  *
410  * @base: The virtual address of LCLA. 18 bit aligned.
411  * @dma_addr: DMA address, if mapped
412  * @base_unaligned: The original kmalloc pointer, if kmalloc is used.
413  * This pointer is only there for clean-up on error.
414  * @pages: The number of pages needed for all physical channels.
415  * Only used later for clean-up on error
416  * @lock: Lock to protect the content in this struct.
417  * @alloc_map: big map over which LCLA entry is own by which job.
418  */
419 struct d40_lcla_pool {
420 	void		*base;
421 	dma_addr_t	dma_addr;
422 	void		*base_unaligned;
423 	int		 pages;
424 	spinlock_t	 lock;
425 	struct d40_desc	**alloc_map;
426 };
427 
428 /**
429  * struct d40_phy_res - struct for handling eventlines mapped to physical
430  * channels.
431  *
432  * @lock: A lock protection this entity.
433  * @reserved: True if used by secure world or otherwise.
434  * @num: The physical channel number of this entity.
435  * @allocated_src: Bit mapped to show which src event line's are mapped to
436  * this physical channel. Can also be free or physically allocated.
437  * @allocated_dst: Same as for src but is dst.
438  * allocated_dst and allocated_src uses the D40_ALLOC* defines as well as
439  * event line number.
440  * @use_soft_lli: To mark if the linked lists of channel are managed by SW.
441  */
442 struct d40_phy_res {
443 	spinlock_t lock;
444 	bool	   reserved;
445 	int	   num;
446 	u32	   allocated_src;
447 	u32	   allocated_dst;
448 	bool	   use_soft_lli;
449 };
450 
451 struct d40_base;
452 
453 /**
454  * struct d40_chan - Struct that describes a channel.
455  *
456  * @lock: A spinlock to protect this struct.
457  * @log_num: The logical number, if any of this channel.
458  * @pending_tx: The number of pending transfers. Used between interrupt handler
459  * and tasklet.
460  * @busy: Set to true when transfer is ongoing on this channel.
461  * @phy_chan: Pointer to physical channel which this instance runs on. If this
462  * point is NULL, then the channel is not allocated.
463  * @chan: DMA engine handle.
464  * @tasklet: Tasklet that gets scheduled from interrupt context to complete a
465  * transfer and call client callback.
466  * @client: Cliented owned descriptor list.
467  * @pending_queue: Submitted jobs, to be issued by issue_pending()
468  * @active: Active descriptor.
469  * @done: Completed jobs
470  * @queue: Queued jobs.
471  * @prepare_queue: Prepared jobs.
472  * @dma_cfg: The client configuration of this dma channel.
473  * @slave_config: DMA slave configuration.
474  * @configured: whether the dma_cfg configuration is valid
475  * @base: Pointer to the device instance struct.
476  * @src_def_cfg: Default cfg register setting for src.
477  * @dst_def_cfg: Default cfg register setting for dst.
478  * @log_def: Default logical channel settings.
479  * @lcpa: Pointer to dst and src lcpa settings.
480  * @runtime_addr: runtime configured address.
481  * @runtime_direction: runtime configured direction.
482  *
483  * This struct can either "be" a logical or a physical channel.
484  */
485 struct d40_chan {
486 	spinlock_t			 lock;
487 	int				 log_num;
488 	int				 pending_tx;
489 	bool				 busy;
490 	struct d40_phy_res		*phy_chan;
491 	struct dma_chan			 chan;
492 	struct tasklet_struct		 tasklet;
493 	struct list_head		 client;
494 	struct list_head		 pending_queue;
495 	struct list_head		 active;
496 	struct list_head		 done;
497 	struct list_head		 queue;
498 	struct list_head		 prepare_queue;
499 	struct stedma40_chan_cfg	 dma_cfg;
500 	struct dma_slave_config		 slave_config;
501 	bool				 configured;
502 	struct d40_base			*base;
503 	/* Default register configurations */
504 	u32				 src_def_cfg;
505 	u32				 dst_def_cfg;
506 	struct d40_def_lcsp		 log_def;
507 	struct d40_log_lli_full		*lcpa;
508 	/* Runtime reconfiguration */
509 	dma_addr_t			runtime_addr;
510 	enum dma_transfer_direction	runtime_direction;
511 };
512 
513 /**
514  * struct d40_gen_dmac - generic values to represent u8500/u8540 DMA
515  * controller
516  *
517  * @backup: the pointer to the registers address array for backup
518  * @backup_size: the size of the registers address array for backup
519  * @realtime_en: the realtime enable register
520  * @realtime_clear: the realtime clear register
521  * @high_prio_en: the high priority enable register
522  * @high_prio_clear: the high priority clear register
523  * @interrupt_en: the interrupt enable register
524  * @interrupt_clear: the interrupt clear register
525  * @il: the pointer to struct d40_interrupt_lookup
526  * @il_size: the size of d40_interrupt_lookup array
527  * @init_reg: the pointer to the struct d40_reg_val
528  * @init_reg_size: the size of d40_reg_val array
529  */
530 struct d40_gen_dmac {
531 	u32				*backup;
532 	u32				 backup_size;
533 	u32				 realtime_en;
534 	u32				 realtime_clear;
535 	u32				 high_prio_en;
536 	u32				 high_prio_clear;
537 	u32				 interrupt_en;
538 	u32				 interrupt_clear;
539 	struct d40_interrupt_lookup	*il;
540 	u32				 il_size;
541 	struct d40_reg_val		*init_reg;
542 	u32				 init_reg_size;
543 };
544 
545 /**
546  * struct d40_base - The big global struct, one for each probe'd instance.
547  *
548  * @interrupt_lock: Lock used to make sure one interrupt is handle a time.
549  * @execmd_lock: Lock for execute command usage since several channels share
550  * the same physical register.
551  * @dev: The device structure.
552  * @virtbase: The virtual base address of the DMA's register.
553  * @rev: silicon revision detected.
554  * @clk: Pointer to the DMA clock structure.
555  * @irq: The IRQ number.
556  * @num_memcpy_chans: The number of channels used for memcpy (mem-to-mem
557  * transfers).
558  * @num_phy_chans: The number of physical channels. Read from HW. This
559  * is the number of available channels for this driver, not counting "Secure
560  * mode" allocated physical channels.
561  * @num_log_chans: The number of logical channels. Calculated from
562  * num_phy_chans.
563  * @dma_both: dma_device channels that can do both memcpy and slave transfers.
564  * @dma_slave: dma_device channels that can do only do slave transfers.
565  * @dma_memcpy: dma_device channels that can do only do memcpy transfers.
566  * @phy_chans: Room for all possible physical channels in system.
567  * @log_chans: Room for all possible logical channels in system.
568  * @lookup_log_chans: Used to map interrupt number to logical channel. Points
569  * to log_chans entries.
570  * @lookup_phy_chans: Used to map interrupt number to physical channel. Points
571  * to phy_chans entries.
572  * @plat_data: Pointer to provided platform_data which is the driver
573  * configuration.
574  * @lcpa_regulator: Pointer to hold the regulator for the esram bank for lcla.
575  * @phy_res: Vector containing all physical channels.
576  * @lcla_pool: lcla pool settings and data.
577  * @lcpa_base: The virtual mapped address of LCPA.
578  * @phy_lcpa: The physical address of the LCPA.
579  * @lcpa_size: The size of the LCPA area.
580  * @desc_slab: cache for descriptors.
581  * @reg_val_backup: Here the values of some hardware registers are stored
582  * before the DMA is powered off. They are restored when the power is back on.
583  * @reg_val_backup_v4: Backup of registers that only exits on dma40 v3 and
584  * later
585  * @reg_val_backup_chan: Backup data for standard channel parameter registers.
586  * @regs_interrupt: Scratch space for registers during interrupt.
587  * @gcc_pwr_off_mask: Mask to maintain the channels that can be turned off.
588  * @gen_dmac: the struct for generic registers values to represent u8500/8540
589  * DMA controller
590  */
591 struct d40_base {
592 	spinlock_t			 interrupt_lock;
593 	spinlock_t			 execmd_lock;
594 	struct device			 *dev;
595 	void __iomem			 *virtbase;
596 	u8				  rev:4;
597 	struct clk			 *clk;
598 	int				  irq;
599 	int				  num_memcpy_chans;
600 	int				  num_phy_chans;
601 	int				  num_log_chans;
602 	struct dma_device		  dma_both;
603 	struct dma_device		  dma_slave;
604 	struct dma_device		  dma_memcpy;
605 	struct d40_chan			 *phy_chans;
606 	struct d40_chan			 *log_chans;
607 	struct d40_chan			**lookup_log_chans;
608 	struct d40_chan			**lookup_phy_chans;
609 	struct stedma40_platform_data	 *plat_data;
610 	struct regulator		 *lcpa_regulator;
611 	/* Physical half channels */
612 	struct d40_phy_res		 *phy_res;
613 	struct d40_lcla_pool		  lcla_pool;
614 	void				 *lcpa_base;
615 	dma_addr_t			  phy_lcpa;
616 	resource_size_t			  lcpa_size;
617 	struct kmem_cache		 *desc_slab;
618 	u32				  reg_val_backup[BACKUP_REGS_SZ];
619 	u32				  reg_val_backup_v4[BACKUP_REGS_SZ_MAX];
620 	u32				 *reg_val_backup_chan;
621 	u32				 *regs_interrupt;
622 	u16				  gcc_pwr_off_mask;
623 	struct d40_gen_dmac		  gen_dmac;
624 };
625 
626 static struct device *chan2dev(struct d40_chan *d40c)
627 {
628 	return &d40c->chan.dev->device;
629 }
630 
631 static bool chan_is_physical(struct d40_chan *chan)
632 {
633 	return chan->log_num == D40_PHY_CHAN;
634 }
635 
636 static bool chan_is_logical(struct d40_chan *chan)
637 {
638 	return !chan_is_physical(chan);
639 }
640 
641 static void __iomem *chan_base(struct d40_chan *chan)
642 {
643 	return chan->base->virtbase + D40_DREG_PCBASE +
644 	       chan->phy_chan->num * D40_DREG_PCDELTA;
645 }
646 
647 #define d40_err(dev, format, arg...)		\
648 	dev_err(dev, "[%s] " format, __func__, ## arg)
649 
650 #define chan_err(d40c, format, arg...)		\
651 	d40_err(chan2dev(d40c), format, ## arg)
652 
653 static int d40_set_runtime_config_write(struct dma_chan *chan,
654 				  struct dma_slave_config *config,
655 				  enum dma_transfer_direction direction);
656 
657 static int d40_pool_lli_alloc(struct d40_chan *d40c, struct d40_desc *d40d,
658 			      int lli_len)
659 {
660 	bool is_log = chan_is_logical(d40c);
661 	u32 align;
662 	void *base;
663 
664 	if (is_log)
665 		align = sizeof(struct d40_log_lli);
666 	else
667 		align = sizeof(struct d40_phy_lli);
668 
669 	if (lli_len == 1) {
670 		base = d40d->lli_pool.pre_alloc_lli;
671 		d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
672 		d40d->lli_pool.base = NULL;
673 	} else {
674 		d40d->lli_pool.size = lli_len * 2 * align;
675 
676 		base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
677 		d40d->lli_pool.base = base;
678 
679 		if (d40d->lli_pool.base == NULL)
680 			return -ENOMEM;
681 	}
682 
683 	if (is_log) {
684 		d40d->lli_log.src = PTR_ALIGN(base, align);
685 		d40d->lli_log.dst = d40d->lli_log.src + lli_len;
686 
687 		d40d->lli_pool.dma_addr = 0;
688 	} else {
689 		d40d->lli_phy.src = PTR_ALIGN(base, align);
690 		d40d->lli_phy.dst = d40d->lli_phy.src + lli_len;
691 
692 		d40d->lli_pool.dma_addr = dma_map_single(d40c->base->dev,
693 							 d40d->lli_phy.src,
694 							 d40d->lli_pool.size,
695 							 DMA_TO_DEVICE);
696 
697 		if (dma_mapping_error(d40c->base->dev,
698 				      d40d->lli_pool.dma_addr)) {
699 			kfree(d40d->lli_pool.base);
700 			d40d->lli_pool.base = NULL;
701 			d40d->lli_pool.dma_addr = 0;
702 			return -ENOMEM;
703 		}
704 	}
705 
706 	return 0;
707 }
708 
709 static void d40_pool_lli_free(struct d40_chan *d40c, struct d40_desc *d40d)
710 {
711 	if (d40d->lli_pool.dma_addr)
712 		dma_unmap_single(d40c->base->dev, d40d->lli_pool.dma_addr,
713 				 d40d->lli_pool.size, DMA_TO_DEVICE);
714 
715 	kfree(d40d->lli_pool.base);
716 	d40d->lli_pool.base = NULL;
717 	d40d->lli_pool.size = 0;
718 	d40d->lli_log.src = NULL;
719 	d40d->lli_log.dst = NULL;
720 	d40d->lli_phy.src = NULL;
721 	d40d->lli_phy.dst = NULL;
722 }
723 
724 static int d40_lcla_alloc_one(struct d40_chan *d40c,
725 			      struct d40_desc *d40d)
726 {
727 	unsigned long flags;
728 	int i;
729 	int ret = -EINVAL;
730 
731 	spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
732 
733 	/*
734 	 * Allocate both src and dst at the same time, therefore the half
735 	 * start on 1 since 0 can't be used since zero is used as end marker.
736 	 */
737 	for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
738 		int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
739 
740 		if (!d40c->base->lcla_pool.alloc_map[idx]) {
741 			d40c->base->lcla_pool.alloc_map[idx] = d40d;
742 			d40d->lcla_alloc++;
743 			ret = i;
744 			break;
745 		}
746 	}
747 
748 	spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
749 
750 	return ret;
751 }
752 
753 static int d40_lcla_free_all(struct d40_chan *d40c,
754 			     struct d40_desc *d40d)
755 {
756 	unsigned long flags;
757 	int i;
758 	int ret = -EINVAL;
759 
760 	if (chan_is_physical(d40c))
761 		return 0;
762 
763 	spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
764 
765 	for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
766 		int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
767 
768 		if (d40c->base->lcla_pool.alloc_map[idx] == d40d) {
769 			d40c->base->lcla_pool.alloc_map[idx] = NULL;
770 			d40d->lcla_alloc--;
771 			if (d40d->lcla_alloc == 0) {
772 				ret = 0;
773 				break;
774 			}
775 		}
776 	}
777 
778 	spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
779 
780 	return ret;
781 
782 }
783 
784 static void d40_desc_remove(struct d40_desc *d40d)
785 {
786 	list_del(&d40d->node);
787 }
788 
789 static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
790 {
791 	struct d40_desc *desc = NULL;
792 
793 	if (!list_empty(&d40c->client)) {
794 		struct d40_desc *d;
795 		struct d40_desc *_d;
796 
797 		list_for_each_entry_safe(d, _d, &d40c->client, node) {
798 			if (async_tx_test_ack(&d->txd)) {
799 				d40_desc_remove(d);
800 				desc = d;
801 				memset(desc, 0, sizeof(*desc));
802 				break;
803 			}
804 		}
805 	}
806 
807 	if (!desc)
808 		desc = kmem_cache_zalloc(d40c->base->desc_slab, GFP_NOWAIT);
809 
810 	if (desc)
811 		INIT_LIST_HEAD(&desc->node);
812 
813 	return desc;
814 }
815 
816 static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
817 {
818 
819 	d40_pool_lli_free(d40c, d40d);
820 	d40_lcla_free_all(d40c, d40d);
821 	kmem_cache_free(d40c->base->desc_slab, d40d);
822 }
823 
824 static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
825 {
826 	list_add_tail(&desc->node, &d40c->active);
827 }
828 
829 static void d40_phy_lli_load(struct d40_chan *chan, struct d40_desc *desc)
830 {
831 	struct d40_phy_lli *lli_dst = desc->lli_phy.dst;
832 	struct d40_phy_lli *lli_src = desc->lli_phy.src;
833 	void __iomem *base = chan_base(chan);
834 
835 	writel(lli_src->reg_cfg, base + D40_CHAN_REG_SSCFG);
836 	writel(lli_src->reg_elt, base + D40_CHAN_REG_SSELT);
837 	writel(lli_src->reg_ptr, base + D40_CHAN_REG_SSPTR);
838 	writel(lli_src->reg_lnk, base + D40_CHAN_REG_SSLNK);
839 
840 	writel(lli_dst->reg_cfg, base + D40_CHAN_REG_SDCFG);
841 	writel(lli_dst->reg_elt, base + D40_CHAN_REG_SDELT);
842 	writel(lli_dst->reg_ptr, base + D40_CHAN_REG_SDPTR);
843 	writel(lli_dst->reg_lnk, base + D40_CHAN_REG_SDLNK);
844 }
845 
846 static void d40_desc_done(struct d40_chan *d40c, struct d40_desc *desc)
847 {
848 	list_add_tail(&desc->node, &d40c->done);
849 }
850 
851 static void d40_log_lli_to_lcxa(struct d40_chan *chan, struct d40_desc *desc)
852 {
853 	struct d40_lcla_pool *pool = &chan->base->lcla_pool;
854 	struct d40_log_lli_bidir *lli = &desc->lli_log;
855 	int lli_current = desc->lli_current;
856 	int lli_len = desc->lli_len;
857 	bool cyclic = desc->cyclic;
858 	int curr_lcla = -EINVAL;
859 	int first_lcla = 0;
860 	bool use_esram_lcla = chan->base->plat_data->use_esram_lcla;
861 	bool linkback;
862 
863 	/*
864 	 * We may have partially running cyclic transfers, in case we did't get
865 	 * enough LCLA entries.
866 	 */
867 	linkback = cyclic && lli_current == 0;
868 
869 	/*
870 	 * For linkback, we need one LCLA even with only one link, because we
871 	 * can't link back to the one in LCPA space
872 	 */
873 	if (linkback || (lli_len - lli_current > 1)) {
874 		/*
875 		 * If the channel is expected to use only soft_lli don't
876 		 * allocate a lcla. This is to avoid a HW issue that exists
877 		 * in some controller during a peripheral to memory transfer
878 		 * that uses linked lists.
879 		 */
880 		if (!(chan->phy_chan->use_soft_lli &&
881 			chan->dma_cfg.dir == DMA_DEV_TO_MEM))
882 			curr_lcla = d40_lcla_alloc_one(chan, desc);
883 
884 		first_lcla = curr_lcla;
885 	}
886 
887 	/*
888 	 * For linkback, we normally load the LCPA in the loop since we need to
889 	 * link it to the second LCLA and not the first.  However, if we
890 	 * couldn't even get a first LCLA, then we have to run in LCPA and
891 	 * reload manually.
892 	 */
893 	if (!linkback || curr_lcla == -EINVAL) {
894 		unsigned int flags = 0;
895 
896 		if (curr_lcla == -EINVAL)
897 			flags |= LLI_TERM_INT;
898 
899 		d40_log_lli_lcpa_write(chan->lcpa,
900 				       &lli->dst[lli_current],
901 				       &lli->src[lli_current],
902 				       curr_lcla,
903 				       flags);
904 		lli_current++;
905 	}
906 
907 	if (curr_lcla < 0)
908 		goto set_current;
909 
910 	for (; lli_current < lli_len; lli_current++) {
911 		unsigned int lcla_offset = chan->phy_chan->num * 1024 +
912 					   8 * curr_lcla * 2;
913 		struct d40_log_lli *lcla = pool->base + lcla_offset;
914 		unsigned int flags = 0;
915 		int next_lcla;
916 
917 		if (lli_current + 1 < lli_len)
918 			next_lcla = d40_lcla_alloc_one(chan, desc);
919 		else
920 			next_lcla = linkback ? first_lcla : -EINVAL;
921 
922 		if (cyclic || next_lcla == -EINVAL)
923 			flags |= LLI_TERM_INT;
924 
925 		if (linkback && curr_lcla == first_lcla) {
926 			/* First link goes in both LCPA and LCLA */
927 			d40_log_lli_lcpa_write(chan->lcpa,
928 					       &lli->dst[lli_current],
929 					       &lli->src[lli_current],
930 					       next_lcla, flags);
931 		}
932 
933 		/*
934 		 * One unused LCLA in the cyclic case if the very first
935 		 * next_lcla fails...
936 		 */
937 		d40_log_lli_lcla_write(lcla,
938 				       &lli->dst[lli_current],
939 				       &lli->src[lli_current],
940 				       next_lcla, flags);
941 
942 		/*
943 		 * Cache maintenance is not needed if lcla is
944 		 * mapped in esram
945 		 */
946 		if (!use_esram_lcla) {
947 			dma_sync_single_range_for_device(chan->base->dev,
948 						pool->dma_addr, lcla_offset,
949 						2 * sizeof(struct d40_log_lli),
950 						DMA_TO_DEVICE);
951 		}
952 		curr_lcla = next_lcla;
953 
954 		if (curr_lcla == -EINVAL || curr_lcla == first_lcla) {
955 			lli_current++;
956 			break;
957 		}
958 	}
959  set_current:
960 	desc->lli_current = lli_current;
961 }
962 
963 static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
964 {
965 	if (chan_is_physical(d40c)) {
966 		d40_phy_lli_load(d40c, d40d);
967 		d40d->lli_current = d40d->lli_len;
968 	} else
969 		d40_log_lli_to_lcxa(d40c, d40d);
970 }
971 
972 static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
973 {
974 	return list_first_entry_or_null(&d40c->active, struct d40_desc, node);
975 }
976 
977 /* remove desc from current queue and add it to the pending_queue */
978 static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
979 {
980 	d40_desc_remove(desc);
981 	desc->is_in_client_list = false;
982 	list_add_tail(&desc->node, &d40c->pending_queue);
983 }
984 
985 static struct d40_desc *d40_first_pending(struct d40_chan *d40c)
986 {
987 	return list_first_entry_or_null(&d40c->pending_queue, struct d40_desc,
988 					node);
989 }
990 
991 static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
992 {
993 	return list_first_entry_or_null(&d40c->queue, struct d40_desc, node);
994 }
995 
996 static struct d40_desc *d40_first_done(struct d40_chan *d40c)
997 {
998 	return list_first_entry_or_null(&d40c->done, struct d40_desc, node);
999 }
1000 
1001 static int d40_psize_2_burst_size(bool is_log, int psize)
1002 {
1003 	if (is_log) {
1004 		if (psize == STEDMA40_PSIZE_LOG_1)
1005 			return 1;
1006 	} else {
1007 		if (psize == STEDMA40_PSIZE_PHY_1)
1008 			return 1;
1009 	}
1010 
1011 	return 2 << psize;
1012 }
1013 
1014 /*
1015  * The dma only supports transmitting packages up to
1016  * STEDMA40_MAX_SEG_SIZE * data_width, where data_width is stored in Bytes.
1017  *
1018  * Calculate the total number of dma elements required to send the entire sg list.
1019  */
1020 static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2)
1021 {
1022 	int dmalen;
1023 	u32 max_w = max(data_width1, data_width2);
1024 	u32 min_w = min(data_width1, data_width2);
1025 	u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE * min_w, max_w);
1026 
1027 	if (seg_max > STEDMA40_MAX_SEG_SIZE)
1028 		seg_max -= max_w;
1029 
1030 	if (!IS_ALIGNED(size, max_w))
1031 		return -EINVAL;
1032 
1033 	if (size <= seg_max)
1034 		dmalen = 1;
1035 	else {
1036 		dmalen = size / seg_max;
1037 		if (dmalen * seg_max < size)
1038 			dmalen++;
1039 	}
1040 	return dmalen;
1041 }
1042 
1043 static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len,
1044 			   u32 data_width1, u32 data_width2)
1045 {
1046 	struct scatterlist *sg;
1047 	int i;
1048 	int len = 0;
1049 	int ret;
1050 
1051 	for_each_sg(sgl, sg, sg_len, i) {
1052 		ret = d40_size_2_dmalen(sg_dma_len(sg),
1053 					data_width1, data_width2);
1054 		if (ret < 0)
1055 			return ret;
1056 		len += ret;
1057 	}
1058 	return len;
1059 }
1060 
1061 static int __d40_execute_command_phy(struct d40_chan *d40c,
1062 				     enum d40_command command)
1063 {
1064 	u32 status;
1065 	int i;
1066 	void __iomem *active_reg;
1067 	int ret = 0;
1068 	unsigned long flags;
1069 	u32 wmask;
1070 
1071 	if (command == D40_DMA_STOP) {
1072 		ret = __d40_execute_command_phy(d40c, D40_DMA_SUSPEND_REQ);
1073 		if (ret)
1074 			return ret;
1075 	}
1076 
1077 	spin_lock_irqsave(&d40c->base->execmd_lock, flags);
1078 
1079 	if (d40c->phy_chan->num % 2 == 0)
1080 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1081 	else
1082 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1083 
1084 	if (command == D40_DMA_SUSPEND_REQ) {
1085 		status = (readl(active_reg) &
1086 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1087 			D40_CHAN_POS(d40c->phy_chan->num);
1088 
1089 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1090 			goto unlock;
1091 	}
1092 
1093 	wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
1094 	writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
1095 	       active_reg);
1096 
1097 	if (command == D40_DMA_SUSPEND_REQ) {
1098 
1099 		for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
1100 			status = (readl(active_reg) &
1101 				  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1102 				D40_CHAN_POS(d40c->phy_chan->num);
1103 
1104 			cpu_relax();
1105 			/*
1106 			 * Reduce the number of bus accesses while
1107 			 * waiting for the DMA to suspend.
1108 			 */
1109 			udelay(3);
1110 
1111 			if (status == D40_DMA_STOP ||
1112 			    status == D40_DMA_SUSPENDED)
1113 				break;
1114 		}
1115 
1116 		if (i == D40_SUSPEND_MAX_IT) {
1117 			chan_err(d40c,
1118 				"unable to suspend the chl %d (log: %d) status %x\n",
1119 				d40c->phy_chan->num, d40c->log_num,
1120 				status);
1121 			dump_stack();
1122 			ret = -EBUSY;
1123 		}
1124 
1125 	}
1126  unlock:
1127 	spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
1128 	return ret;
1129 }
1130 
1131 static void d40_term_all(struct d40_chan *d40c)
1132 {
1133 	struct d40_desc *d40d;
1134 	struct d40_desc *_d;
1135 
1136 	/* Release completed descriptors */
1137 	while ((d40d = d40_first_done(d40c))) {
1138 		d40_desc_remove(d40d);
1139 		d40_desc_free(d40c, d40d);
1140 	}
1141 
1142 	/* Release active descriptors */
1143 	while ((d40d = d40_first_active_get(d40c))) {
1144 		d40_desc_remove(d40d);
1145 		d40_desc_free(d40c, d40d);
1146 	}
1147 
1148 	/* Release queued descriptors waiting for transfer */
1149 	while ((d40d = d40_first_queued(d40c))) {
1150 		d40_desc_remove(d40d);
1151 		d40_desc_free(d40c, d40d);
1152 	}
1153 
1154 	/* Release pending descriptors */
1155 	while ((d40d = d40_first_pending(d40c))) {
1156 		d40_desc_remove(d40d);
1157 		d40_desc_free(d40c, d40d);
1158 	}
1159 
1160 	/* Release client owned descriptors */
1161 	if (!list_empty(&d40c->client))
1162 		list_for_each_entry_safe(d40d, _d, &d40c->client, node) {
1163 			d40_desc_remove(d40d);
1164 			d40_desc_free(d40c, d40d);
1165 		}
1166 
1167 	/* Release descriptors in prepare queue */
1168 	if (!list_empty(&d40c->prepare_queue))
1169 		list_for_each_entry_safe(d40d, _d,
1170 					 &d40c->prepare_queue, node) {
1171 			d40_desc_remove(d40d);
1172 			d40_desc_free(d40c, d40d);
1173 		}
1174 
1175 	d40c->pending_tx = 0;
1176 }
1177 
1178 static void __d40_config_set_event(struct d40_chan *d40c,
1179 				   enum d40_events event_type, u32 event,
1180 				   int reg)
1181 {
1182 	void __iomem *addr = chan_base(d40c) + reg;
1183 	int tries;
1184 	u32 status;
1185 
1186 	switch (event_type) {
1187 
1188 	case D40_DEACTIVATE_EVENTLINE:
1189 
1190 		writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
1191 		       | ~D40_EVENTLINE_MASK(event), addr);
1192 		break;
1193 
1194 	case D40_SUSPEND_REQ_EVENTLINE:
1195 		status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1196 			  D40_EVENTLINE_POS(event);
1197 
1198 		if (status == D40_DEACTIVATE_EVENTLINE ||
1199 		    status == D40_SUSPEND_REQ_EVENTLINE)
1200 			break;
1201 
1202 		writel((D40_SUSPEND_REQ_EVENTLINE << D40_EVENTLINE_POS(event))
1203 		       | ~D40_EVENTLINE_MASK(event), addr);
1204 
1205 		for (tries = 0 ; tries < D40_SUSPEND_MAX_IT; tries++) {
1206 
1207 			status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1208 				  D40_EVENTLINE_POS(event);
1209 
1210 			cpu_relax();
1211 			/*
1212 			 * Reduce the number of bus accesses while
1213 			 * waiting for the DMA to suspend.
1214 			 */
1215 			udelay(3);
1216 
1217 			if (status == D40_DEACTIVATE_EVENTLINE)
1218 				break;
1219 		}
1220 
1221 		if (tries == D40_SUSPEND_MAX_IT) {
1222 			chan_err(d40c,
1223 				"unable to stop the event_line chl %d (log: %d)"
1224 				"status %x\n", d40c->phy_chan->num,
1225 				 d40c->log_num, status);
1226 		}
1227 		break;
1228 
1229 	case D40_ACTIVATE_EVENTLINE:
1230 	/*
1231 	 * The hardware sometimes doesn't register the enable when src and dst
1232 	 * event lines are active on the same logical channel.  Retry to ensure
1233 	 * it does.  Usually only one retry is sufficient.
1234 	 */
1235 		tries = 100;
1236 		while (--tries) {
1237 			writel((D40_ACTIVATE_EVENTLINE <<
1238 				D40_EVENTLINE_POS(event)) |
1239 				~D40_EVENTLINE_MASK(event), addr);
1240 
1241 			if (readl(addr) & D40_EVENTLINE_MASK(event))
1242 				break;
1243 		}
1244 
1245 		if (tries != 99)
1246 			dev_dbg(chan2dev(d40c),
1247 				"[%s] workaround enable S%cLNK (%d tries)\n",
1248 				__func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D',
1249 				100 - tries);
1250 
1251 		WARN_ON(!tries);
1252 		break;
1253 
1254 	case D40_ROUND_EVENTLINE:
1255 		BUG();
1256 		break;
1257 
1258 	}
1259 }
1260 
1261 static void d40_config_set_event(struct d40_chan *d40c,
1262 				 enum d40_events event_type)
1263 {
1264 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
1265 
1266 	/* Enable event line connected to device (or memcpy) */
1267 	if ((d40c->dma_cfg.dir == DMA_DEV_TO_MEM) ||
1268 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
1269 		__d40_config_set_event(d40c, event_type, event,
1270 				       D40_CHAN_REG_SSLNK);
1271 
1272 	if (d40c->dma_cfg.dir !=  DMA_DEV_TO_MEM)
1273 		__d40_config_set_event(d40c, event_type, event,
1274 				       D40_CHAN_REG_SDLNK);
1275 }
1276 
1277 static u32 d40_chan_has_events(struct d40_chan *d40c)
1278 {
1279 	void __iomem *chanbase = chan_base(d40c);
1280 	u32 val;
1281 
1282 	val = readl(chanbase + D40_CHAN_REG_SSLNK);
1283 	val |= readl(chanbase + D40_CHAN_REG_SDLNK);
1284 
1285 	return val;
1286 }
1287 
1288 static int
1289 __d40_execute_command_log(struct d40_chan *d40c, enum d40_command command)
1290 {
1291 	unsigned long flags;
1292 	int ret = 0;
1293 	u32 active_status;
1294 	void __iomem *active_reg;
1295 
1296 	if (d40c->phy_chan->num % 2 == 0)
1297 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1298 	else
1299 		active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1300 
1301 
1302 	spin_lock_irqsave(&d40c->phy_chan->lock, flags);
1303 
1304 	switch (command) {
1305 	case D40_DMA_STOP:
1306 	case D40_DMA_SUSPEND_REQ:
1307 
1308 		active_status = (readl(active_reg) &
1309 				 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1310 				 D40_CHAN_POS(d40c->phy_chan->num);
1311 
1312 		if (active_status == D40_DMA_RUN)
1313 			d40_config_set_event(d40c, D40_SUSPEND_REQ_EVENTLINE);
1314 		else
1315 			d40_config_set_event(d40c, D40_DEACTIVATE_EVENTLINE);
1316 
1317 		if (!d40_chan_has_events(d40c) && (command == D40_DMA_STOP))
1318 			ret = __d40_execute_command_phy(d40c, command);
1319 
1320 		break;
1321 
1322 	case D40_DMA_RUN:
1323 
1324 		d40_config_set_event(d40c, D40_ACTIVATE_EVENTLINE);
1325 		ret = __d40_execute_command_phy(d40c, command);
1326 		break;
1327 
1328 	case D40_DMA_SUSPENDED:
1329 		BUG();
1330 		break;
1331 	}
1332 
1333 	spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
1334 	return ret;
1335 }
1336 
1337 static int d40_channel_execute_command(struct d40_chan *d40c,
1338 				       enum d40_command command)
1339 {
1340 	if (chan_is_logical(d40c))
1341 		return __d40_execute_command_log(d40c, command);
1342 	else
1343 		return __d40_execute_command_phy(d40c, command);
1344 }
1345 
1346 static u32 d40_get_prmo(struct d40_chan *d40c)
1347 {
1348 	static const unsigned int phy_map[] = {
1349 		[STEDMA40_PCHAN_BASIC_MODE]
1350 			= D40_DREG_PRMO_PCHAN_BASIC,
1351 		[STEDMA40_PCHAN_MODULO_MODE]
1352 			= D40_DREG_PRMO_PCHAN_MODULO,
1353 		[STEDMA40_PCHAN_DOUBLE_DST_MODE]
1354 			= D40_DREG_PRMO_PCHAN_DOUBLE_DST,
1355 	};
1356 	static const unsigned int log_map[] = {
1357 		[STEDMA40_LCHAN_SRC_PHY_DST_LOG]
1358 			= D40_DREG_PRMO_LCHAN_SRC_PHY_DST_LOG,
1359 		[STEDMA40_LCHAN_SRC_LOG_DST_PHY]
1360 			= D40_DREG_PRMO_LCHAN_SRC_LOG_DST_PHY,
1361 		[STEDMA40_LCHAN_SRC_LOG_DST_LOG]
1362 			= D40_DREG_PRMO_LCHAN_SRC_LOG_DST_LOG,
1363 	};
1364 
1365 	if (chan_is_physical(d40c))
1366 		return phy_map[d40c->dma_cfg.mode_opt];
1367 	else
1368 		return log_map[d40c->dma_cfg.mode_opt];
1369 }
1370 
1371 static void d40_config_write(struct d40_chan *d40c)
1372 {
1373 	u32 addr_base;
1374 	u32 var;
1375 
1376 	/* Odd addresses are even addresses + 4 */
1377 	addr_base = (d40c->phy_chan->num % 2) * 4;
1378 	/* Setup channel mode to logical or physical */
1379 	var = ((u32)(chan_is_logical(d40c)) + 1) <<
1380 		D40_CHAN_POS(d40c->phy_chan->num);
1381 	writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
1382 
1383 	/* Setup operational mode option register */
1384 	var = d40_get_prmo(d40c) << D40_CHAN_POS(d40c->phy_chan->num);
1385 
1386 	writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
1387 
1388 	if (chan_is_logical(d40c)) {
1389 		int lidx = (d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS)
1390 			   & D40_SREG_ELEM_LOG_LIDX_MASK;
1391 		void __iomem *chanbase = chan_base(d40c);
1392 
1393 		/* Set default config for CFG reg */
1394 		writel(d40c->src_def_cfg, chanbase + D40_CHAN_REG_SSCFG);
1395 		writel(d40c->dst_def_cfg, chanbase + D40_CHAN_REG_SDCFG);
1396 
1397 		/* Set LIDX for lcla */
1398 		writel(lidx, chanbase + D40_CHAN_REG_SSELT);
1399 		writel(lidx, chanbase + D40_CHAN_REG_SDELT);
1400 
1401 		/* Clear LNK which will be used by d40_chan_has_events() */
1402 		writel(0, chanbase + D40_CHAN_REG_SSLNK);
1403 		writel(0, chanbase + D40_CHAN_REG_SDLNK);
1404 	}
1405 }
1406 
1407 static u32 d40_residue(struct d40_chan *d40c)
1408 {
1409 	u32 num_elt;
1410 
1411 	if (chan_is_logical(d40c))
1412 		num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
1413 			>> D40_MEM_LCSP2_ECNT_POS;
1414 	else {
1415 		u32 val = readl(chan_base(d40c) + D40_CHAN_REG_SDELT);
1416 		num_elt = (val & D40_SREG_ELEM_PHY_ECNT_MASK)
1417 			  >> D40_SREG_ELEM_PHY_ECNT_POS;
1418 	}
1419 
1420 	return num_elt * d40c->dma_cfg.dst_info.data_width;
1421 }
1422 
1423 static bool d40_tx_is_linked(struct d40_chan *d40c)
1424 {
1425 	bool is_link;
1426 
1427 	if (chan_is_logical(d40c))
1428 		is_link = readl(&d40c->lcpa->lcsp3) &  D40_MEM_LCSP3_DLOS_MASK;
1429 	else
1430 		is_link = readl(chan_base(d40c) + D40_CHAN_REG_SDLNK)
1431 			  & D40_SREG_LNK_PHYS_LNK_MASK;
1432 
1433 	return is_link;
1434 }
1435 
1436 static int d40_pause(struct dma_chan *chan)
1437 {
1438 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
1439 	int res = 0;
1440 	unsigned long flags;
1441 
1442 	if (d40c->phy_chan == NULL) {
1443 		chan_err(d40c, "Channel is not allocated!\n");
1444 		return -EINVAL;
1445 	}
1446 
1447 	if (!d40c->busy)
1448 		return 0;
1449 
1450 	spin_lock_irqsave(&d40c->lock, flags);
1451 	pm_runtime_get_sync(d40c->base->dev);
1452 
1453 	res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1454 
1455 	pm_runtime_mark_last_busy(d40c->base->dev);
1456 	pm_runtime_put_autosuspend(d40c->base->dev);
1457 	spin_unlock_irqrestore(&d40c->lock, flags);
1458 	return res;
1459 }
1460 
1461 static int d40_resume(struct dma_chan *chan)
1462 {
1463 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
1464 	int res = 0;
1465 	unsigned long flags;
1466 
1467 	if (d40c->phy_chan == NULL) {
1468 		chan_err(d40c, "Channel is not allocated!\n");
1469 		return -EINVAL;
1470 	}
1471 
1472 	if (!d40c->busy)
1473 		return 0;
1474 
1475 	spin_lock_irqsave(&d40c->lock, flags);
1476 	pm_runtime_get_sync(d40c->base->dev);
1477 
1478 	/* If bytes left to transfer or linked tx resume job */
1479 	if (d40_residue(d40c) || d40_tx_is_linked(d40c))
1480 		res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1481 
1482 	pm_runtime_mark_last_busy(d40c->base->dev);
1483 	pm_runtime_put_autosuspend(d40c->base->dev);
1484 	spin_unlock_irqrestore(&d40c->lock, flags);
1485 	return res;
1486 }
1487 
1488 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
1489 {
1490 	struct d40_chan *d40c = container_of(tx->chan,
1491 					     struct d40_chan,
1492 					     chan);
1493 	struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
1494 	unsigned long flags;
1495 	dma_cookie_t cookie;
1496 
1497 	spin_lock_irqsave(&d40c->lock, flags);
1498 	cookie = dma_cookie_assign(tx);
1499 	d40_desc_queue(d40c, d40d);
1500 	spin_unlock_irqrestore(&d40c->lock, flags);
1501 
1502 	return cookie;
1503 }
1504 
1505 static int d40_start(struct d40_chan *d40c)
1506 {
1507 	return d40_channel_execute_command(d40c, D40_DMA_RUN);
1508 }
1509 
1510 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
1511 {
1512 	struct d40_desc *d40d;
1513 	int err;
1514 
1515 	/* Start queued jobs, if any */
1516 	d40d = d40_first_queued(d40c);
1517 
1518 	if (d40d != NULL) {
1519 		if (!d40c->busy) {
1520 			d40c->busy = true;
1521 			pm_runtime_get_sync(d40c->base->dev);
1522 		}
1523 
1524 		/* Remove from queue */
1525 		d40_desc_remove(d40d);
1526 
1527 		/* Add to active queue */
1528 		d40_desc_submit(d40c, d40d);
1529 
1530 		/* Initiate DMA job */
1531 		d40_desc_load(d40c, d40d);
1532 
1533 		/* Start dma job */
1534 		err = d40_start(d40c);
1535 
1536 		if (err)
1537 			return NULL;
1538 	}
1539 
1540 	return d40d;
1541 }
1542 
1543 /* called from interrupt context */
1544 static void dma_tc_handle(struct d40_chan *d40c)
1545 {
1546 	struct d40_desc *d40d;
1547 
1548 	/* Get first active entry from list */
1549 	d40d = d40_first_active_get(d40c);
1550 
1551 	if (d40d == NULL)
1552 		return;
1553 
1554 	if (d40d->cyclic) {
1555 		/*
1556 		 * If this was a paritially loaded list, we need to reloaded
1557 		 * it, and only when the list is completed.  We need to check
1558 		 * for done because the interrupt will hit for every link, and
1559 		 * not just the last one.
1560 		 */
1561 		if (d40d->lli_current < d40d->lli_len
1562 		    && !d40_tx_is_linked(d40c)
1563 		    && !d40_residue(d40c)) {
1564 			d40_lcla_free_all(d40c, d40d);
1565 			d40_desc_load(d40c, d40d);
1566 			(void) d40_start(d40c);
1567 
1568 			if (d40d->lli_current == d40d->lli_len)
1569 				d40d->lli_current = 0;
1570 		}
1571 	} else {
1572 		d40_lcla_free_all(d40c, d40d);
1573 
1574 		if (d40d->lli_current < d40d->lli_len) {
1575 			d40_desc_load(d40c, d40d);
1576 			/* Start dma job */
1577 			(void) d40_start(d40c);
1578 			return;
1579 		}
1580 
1581 		if (d40_queue_start(d40c) == NULL) {
1582 			d40c->busy = false;
1583 
1584 			pm_runtime_mark_last_busy(d40c->base->dev);
1585 			pm_runtime_put_autosuspend(d40c->base->dev);
1586 		}
1587 
1588 		d40_desc_remove(d40d);
1589 		d40_desc_done(d40c, d40d);
1590 	}
1591 
1592 	d40c->pending_tx++;
1593 	tasklet_schedule(&d40c->tasklet);
1594 
1595 }
1596 
1597 static void dma_tasklet(struct tasklet_struct *t)
1598 {
1599 	struct d40_chan *d40c = from_tasklet(d40c, t, tasklet);
1600 	struct d40_desc *d40d;
1601 	unsigned long flags;
1602 	bool callback_active;
1603 	struct dmaengine_desc_callback cb;
1604 
1605 	spin_lock_irqsave(&d40c->lock, flags);
1606 
1607 	/* Get first entry from the done list */
1608 	d40d = d40_first_done(d40c);
1609 	if (d40d == NULL) {
1610 		/* Check if we have reached here for cyclic job */
1611 		d40d = d40_first_active_get(d40c);
1612 		if (d40d == NULL || !d40d->cyclic)
1613 			goto check_pending_tx;
1614 	}
1615 
1616 	if (!d40d->cyclic)
1617 		dma_cookie_complete(&d40d->txd);
1618 
1619 	/*
1620 	 * If terminating a channel pending_tx is set to zero.
1621 	 * This prevents any finished active jobs to return to the client.
1622 	 */
1623 	if (d40c->pending_tx == 0) {
1624 		spin_unlock_irqrestore(&d40c->lock, flags);
1625 		return;
1626 	}
1627 
1628 	/* Callback to client */
1629 	callback_active = !!(d40d->txd.flags & DMA_PREP_INTERRUPT);
1630 	dmaengine_desc_get_callback(&d40d->txd, &cb);
1631 
1632 	if (!d40d->cyclic) {
1633 		if (async_tx_test_ack(&d40d->txd)) {
1634 			d40_desc_remove(d40d);
1635 			d40_desc_free(d40c, d40d);
1636 		} else if (!d40d->is_in_client_list) {
1637 			d40_desc_remove(d40d);
1638 			d40_lcla_free_all(d40c, d40d);
1639 			list_add_tail(&d40d->node, &d40c->client);
1640 			d40d->is_in_client_list = true;
1641 		}
1642 	}
1643 
1644 	d40c->pending_tx--;
1645 
1646 	if (d40c->pending_tx)
1647 		tasklet_schedule(&d40c->tasklet);
1648 
1649 	spin_unlock_irqrestore(&d40c->lock, flags);
1650 
1651 	if (callback_active)
1652 		dmaengine_desc_callback_invoke(&cb, NULL);
1653 
1654 	return;
1655  check_pending_tx:
1656 	/* Rescue maneuver if receiving double interrupts */
1657 	if (d40c->pending_tx > 0)
1658 		d40c->pending_tx--;
1659 	spin_unlock_irqrestore(&d40c->lock, flags);
1660 }
1661 
1662 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1663 {
1664 	int i;
1665 	u32 idx;
1666 	u32 row;
1667 	long chan = -1;
1668 	struct d40_chan *d40c;
1669 	struct d40_base *base = data;
1670 	u32 *regs = base->regs_interrupt;
1671 	struct d40_interrupt_lookup *il = base->gen_dmac.il;
1672 	u32 il_size = base->gen_dmac.il_size;
1673 
1674 	spin_lock(&base->interrupt_lock);
1675 
1676 	/* Read interrupt status of both logical and physical channels */
1677 	for (i = 0; i < il_size; i++)
1678 		regs[i] = readl(base->virtbase + il[i].src);
1679 
1680 	for (;;) {
1681 
1682 		chan = find_next_bit((unsigned long *)regs,
1683 				     BITS_PER_LONG * il_size, chan + 1);
1684 
1685 		/* No more set bits found? */
1686 		if (chan == BITS_PER_LONG * il_size)
1687 			break;
1688 
1689 		row = chan / BITS_PER_LONG;
1690 		idx = chan & (BITS_PER_LONG - 1);
1691 
1692 		if (il[row].offset == D40_PHY_CHAN)
1693 			d40c = base->lookup_phy_chans[idx];
1694 		else
1695 			d40c = base->lookup_log_chans[il[row].offset + idx];
1696 
1697 		if (!d40c) {
1698 			/*
1699 			 * No error because this can happen if something else
1700 			 * in the system is using the channel.
1701 			 */
1702 			continue;
1703 		}
1704 
1705 		/* ACK interrupt */
1706 		writel(BIT(idx), base->virtbase + il[row].clr);
1707 
1708 		spin_lock(&d40c->lock);
1709 
1710 		if (!il[row].is_error)
1711 			dma_tc_handle(d40c);
1712 		else
1713 			d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1714 				chan, il[row].offset, idx);
1715 
1716 		spin_unlock(&d40c->lock);
1717 	}
1718 
1719 	spin_unlock(&base->interrupt_lock);
1720 
1721 	return IRQ_HANDLED;
1722 }
1723 
1724 static int d40_validate_conf(struct d40_chan *d40c,
1725 			     struct stedma40_chan_cfg *conf)
1726 {
1727 	int res = 0;
1728 	bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1729 
1730 	if (!conf->dir) {
1731 		chan_err(d40c, "Invalid direction.\n");
1732 		res = -EINVAL;
1733 	}
1734 
1735 	if ((is_log && conf->dev_type > d40c->base->num_log_chans)  ||
1736 	    (!is_log && conf->dev_type > d40c->base->num_phy_chans) ||
1737 	    (conf->dev_type < 0)) {
1738 		chan_err(d40c, "Invalid device type (%d)\n", conf->dev_type);
1739 		res = -EINVAL;
1740 	}
1741 
1742 	if (conf->dir == DMA_DEV_TO_DEV) {
1743 		/*
1744 		 * DMAC HW supports it. Will be added to this driver,
1745 		 * in case any dma client requires it.
1746 		 */
1747 		chan_err(d40c, "periph to periph not supported\n");
1748 		res = -EINVAL;
1749 	}
1750 
1751 	if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1752 	    conf->src_info.data_width !=
1753 	    d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1754 	    conf->dst_info.data_width) {
1755 		/*
1756 		 * The DMAC hardware only supports
1757 		 * src (burst x width) == dst (burst x width)
1758 		 */
1759 
1760 		chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1761 		res = -EINVAL;
1762 	}
1763 
1764 	return res;
1765 }
1766 
1767 static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1768 			       bool is_src, int log_event_line, bool is_log,
1769 			       bool *first_user)
1770 {
1771 	unsigned long flags;
1772 	spin_lock_irqsave(&phy->lock, flags);
1773 
1774 	*first_user = ((phy->allocated_src | phy->allocated_dst)
1775 			== D40_ALLOC_FREE);
1776 
1777 	if (!is_log) {
1778 		/* Physical interrupts are masked per physical full channel */
1779 		if (phy->allocated_src == D40_ALLOC_FREE &&
1780 		    phy->allocated_dst == D40_ALLOC_FREE) {
1781 			phy->allocated_dst = D40_ALLOC_PHY;
1782 			phy->allocated_src = D40_ALLOC_PHY;
1783 			goto found_unlock;
1784 		} else
1785 			goto not_found_unlock;
1786 	}
1787 
1788 	/* Logical channel */
1789 	if (is_src) {
1790 		if (phy->allocated_src == D40_ALLOC_PHY)
1791 			goto not_found_unlock;
1792 
1793 		if (phy->allocated_src == D40_ALLOC_FREE)
1794 			phy->allocated_src = D40_ALLOC_LOG_FREE;
1795 
1796 		if (!(phy->allocated_src & BIT(log_event_line))) {
1797 			phy->allocated_src |= BIT(log_event_line);
1798 			goto found_unlock;
1799 		} else
1800 			goto not_found_unlock;
1801 	} else {
1802 		if (phy->allocated_dst == D40_ALLOC_PHY)
1803 			goto not_found_unlock;
1804 
1805 		if (phy->allocated_dst == D40_ALLOC_FREE)
1806 			phy->allocated_dst = D40_ALLOC_LOG_FREE;
1807 
1808 		if (!(phy->allocated_dst & BIT(log_event_line))) {
1809 			phy->allocated_dst |= BIT(log_event_line);
1810 			goto found_unlock;
1811 		}
1812 	}
1813  not_found_unlock:
1814 	spin_unlock_irqrestore(&phy->lock, flags);
1815 	return false;
1816  found_unlock:
1817 	spin_unlock_irqrestore(&phy->lock, flags);
1818 	return true;
1819 }
1820 
1821 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1822 			       int log_event_line)
1823 {
1824 	unsigned long flags;
1825 	bool is_free = false;
1826 
1827 	spin_lock_irqsave(&phy->lock, flags);
1828 	if (!log_event_line) {
1829 		phy->allocated_dst = D40_ALLOC_FREE;
1830 		phy->allocated_src = D40_ALLOC_FREE;
1831 		is_free = true;
1832 		goto unlock;
1833 	}
1834 
1835 	/* Logical channel */
1836 	if (is_src) {
1837 		phy->allocated_src &= ~BIT(log_event_line);
1838 		if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1839 			phy->allocated_src = D40_ALLOC_FREE;
1840 	} else {
1841 		phy->allocated_dst &= ~BIT(log_event_line);
1842 		if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1843 			phy->allocated_dst = D40_ALLOC_FREE;
1844 	}
1845 
1846 	is_free = ((phy->allocated_src | phy->allocated_dst) ==
1847 		   D40_ALLOC_FREE);
1848  unlock:
1849 	spin_unlock_irqrestore(&phy->lock, flags);
1850 
1851 	return is_free;
1852 }
1853 
1854 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1855 {
1856 	int dev_type = d40c->dma_cfg.dev_type;
1857 	int event_group;
1858 	int event_line;
1859 	struct d40_phy_res *phys;
1860 	int i;
1861 	int j;
1862 	int log_num;
1863 	int num_phy_chans;
1864 	bool is_src;
1865 	bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1866 
1867 	phys = d40c->base->phy_res;
1868 	num_phy_chans = d40c->base->num_phy_chans;
1869 
1870 	if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
1871 		log_num = 2 * dev_type;
1872 		is_src = true;
1873 	} else if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
1874 		   d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1875 		/* dst event lines are used for logical memcpy */
1876 		log_num = 2 * dev_type + 1;
1877 		is_src = false;
1878 	} else
1879 		return -EINVAL;
1880 
1881 	event_group = D40_TYPE_TO_GROUP(dev_type);
1882 	event_line = D40_TYPE_TO_EVENT(dev_type);
1883 
1884 	if (!is_log) {
1885 		if (d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1886 			/* Find physical half channel */
1887 			if (d40c->dma_cfg.use_fixed_channel) {
1888 				i = d40c->dma_cfg.phy_channel;
1889 				if (d40_alloc_mask_set(&phys[i], is_src,
1890 						       0, is_log,
1891 						       first_phy_user))
1892 					goto found_phy;
1893 			} else {
1894 				for (i = 0; i < num_phy_chans; i++) {
1895 					if (d40_alloc_mask_set(&phys[i], is_src,
1896 						       0, is_log,
1897 						       first_phy_user))
1898 						goto found_phy;
1899 				}
1900 			}
1901 		} else
1902 			for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1903 				int phy_num = j  + event_group * 2;
1904 				for (i = phy_num; i < phy_num + 2; i++) {
1905 					if (d40_alloc_mask_set(&phys[i],
1906 							       is_src,
1907 							       0,
1908 							       is_log,
1909 							       first_phy_user))
1910 						goto found_phy;
1911 				}
1912 			}
1913 		return -EINVAL;
1914 found_phy:
1915 		d40c->phy_chan = &phys[i];
1916 		d40c->log_num = D40_PHY_CHAN;
1917 		goto out;
1918 	}
1919 	if (dev_type == -1)
1920 		return -EINVAL;
1921 
1922 	/* Find logical channel */
1923 	for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1924 		int phy_num = j + event_group * 2;
1925 
1926 		if (d40c->dma_cfg.use_fixed_channel) {
1927 			i = d40c->dma_cfg.phy_channel;
1928 
1929 			if ((i != phy_num) && (i != phy_num + 1)) {
1930 				dev_err(chan2dev(d40c),
1931 					"invalid fixed phy channel %d\n", i);
1932 				return -EINVAL;
1933 			}
1934 
1935 			if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1936 					       is_log, first_phy_user))
1937 				goto found_log;
1938 
1939 			dev_err(chan2dev(d40c),
1940 				"could not allocate fixed phy channel %d\n", i);
1941 			return -EINVAL;
1942 		}
1943 
1944 		/*
1945 		 * Spread logical channels across all available physical rather
1946 		 * than pack every logical channel at the first available phy
1947 		 * channels.
1948 		 */
1949 		if (is_src) {
1950 			for (i = phy_num; i < phy_num + 2; i++) {
1951 				if (d40_alloc_mask_set(&phys[i], is_src,
1952 						       event_line, is_log,
1953 						       first_phy_user))
1954 					goto found_log;
1955 			}
1956 		} else {
1957 			for (i = phy_num + 1; i >= phy_num; i--) {
1958 				if (d40_alloc_mask_set(&phys[i], is_src,
1959 						       event_line, is_log,
1960 						       first_phy_user))
1961 					goto found_log;
1962 			}
1963 		}
1964 	}
1965 	return -EINVAL;
1966 
1967 found_log:
1968 	d40c->phy_chan = &phys[i];
1969 	d40c->log_num = log_num;
1970 out:
1971 
1972 	if (is_log)
1973 		d40c->base->lookup_log_chans[d40c->log_num] = d40c;
1974 	else
1975 		d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
1976 
1977 	return 0;
1978 
1979 }
1980 
1981 static int d40_config_memcpy(struct d40_chan *d40c)
1982 {
1983 	dma_cap_mask_t cap = d40c->chan.device->cap_mask;
1984 
1985 	if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
1986 		d40c->dma_cfg = dma40_memcpy_conf_log;
1987 		d40c->dma_cfg.dev_type = dma40_memcpy_channels[d40c->chan.chan_id];
1988 
1989 		d40_log_cfg(&d40c->dma_cfg,
1990 			    &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
1991 
1992 	} else if (dma_has_cap(DMA_MEMCPY, cap) &&
1993 		   dma_has_cap(DMA_SLAVE, cap)) {
1994 		d40c->dma_cfg = dma40_memcpy_conf_phy;
1995 
1996 		/* Generate interrupt at end of transfer or relink. */
1997 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_TIM_POS);
1998 
1999 		/* Generate interrupt on error. */
2000 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
2001 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
2002 
2003 	} else {
2004 		chan_err(d40c, "No memcpy\n");
2005 		return -EINVAL;
2006 	}
2007 
2008 	return 0;
2009 }
2010 
2011 static int d40_free_dma(struct d40_chan *d40c)
2012 {
2013 
2014 	int res = 0;
2015 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2016 	struct d40_phy_res *phy = d40c->phy_chan;
2017 	bool is_src;
2018 
2019 	/* Terminate all queued and active transfers */
2020 	d40_term_all(d40c);
2021 
2022 	if (phy == NULL) {
2023 		chan_err(d40c, "phy == null\n");
2024 		return -EINVAL;
2025 	}
2026 
2027 	if (phy->allocated_src == D40_ALLOC_FREE &&
2028 	    phy->allocated_dst == D40_ALLOC_FREE) {
2029 		chan_err(d40c, "channel already free\n");
2030 		return -EINVAL;
2031 	}
2032 
2033 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2034 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM)
2035 		is_src = false;
2036 	else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2037 		is_src = true;
2038 	else {
2039 		chan_err(d40c, "Unknown direction\n");
2040 		return -EINVAL;
2041 	}
2042 
2043 	pm_runtime_get_sync(d40c->base->dev);
2044 	res = d40_channel_execute_command(d40c, D40_DMA_STOP);
2045 	if (res) {
2046 		chan_err(d40c, "stop failed\n");
2047 		goto mark_last_busy;
2048 	}
2049 
2050 	d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
2051 
2052 	if (chan_is_logical(d40c))
2053 		d40c->base->lookup_log_chans[d40c->log_num] = NULL;
2054 	else
2055 		d40c->base->lookup_phy_chans[phy->num] = NULL;
2056 
2057 	if (d40c->busy) {
2058 		pm_runtime_mark_last_busy(d40c->base->dev);
2059 		pm_runtime_put_autosuspend(d40c->base->dev);
2060 	}
2061 
2062 	d40c->busy = false;
2063 	d40c->phy_chan = NULL;
2064 	d40c->configured = false;
2065  mark_last_busy:
2066 	pm_runtime_mark_last_busy(d40c->base->dev);
2067 	pm_runtime_put_autosuspend(d40c->base->dev);
2068 	return res;
2069 }
2070 
2071 static bool d40_is_paused(struct d40_chan *d40c)
2072 {
2073 	void __iomem *chanbase = chan_base(d40c);
2074 	bool is_paused = false;
2075 	unsigned long flags;
2076 	void __iomem *active_reg;
2077 	u32 status;
2078 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2079 
2080 	spin_lock_irqsave(&d40c->lock, flags);
2081 
2082 	if (chan_is_physical(d40c)) {
2083 		if (d40c->phy_chan->num % 2 == 0)
2084 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
2085 		else
2086 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
2087 
2088 		status = (readl(active_reg) &
2089 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
2090 			D40_CHAN_POS(d40c->phy_chan->num);
2091 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
2092 			is_paused = true;
2093 		goto unlock;
2094 	}
2095 
2096 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2097 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
2098 		status = readl(chanbase + D40_CHAN_REG_SDLNK);
2099 	} else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
2100 		status = readl(chanbase + D40_CHAN_REG_SSLNK);
2101 	} else {
2102 		chan_err(d40c, "Unknown direction\n");
2103 		goto unlock;
2104 	}
2105 
2106 	status = (status & D40_EVENTLINE_MASK(event)) >>
2107 		D40_EVENTLINE_POS(event);
2108 
2109 	if (status != D40_DMA_RUN)
2110 		is_paused = true;
2111  unlock:
2112 	spin_unlock_irqrestore(&d40c->lock, flags);
2113 	return is_paused;
2114 
2115 }
2116 
2117 static u32 stedma40_residue(struct dma_chan *chan)
2118 {
2119 	struct d40_chan *d40c =
2120 		container_of(chan, struct d40_chan, chan);
2121 	u32 bytes_left;
2122 	unsigned long flags;
2123 
2124 	spin_lock_irqsave(&d40c->lock, flags);
2125 	bytes_left = d40_residue(d40c);
2126 	spin_unlock_irqrestore(&d40c->lock, flags);
2127 
2128 	return bytes_left;
2129 }
2130 
2131 static int
2132 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
2133 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2134 		unsigned int sg_len, dma_addr_t src_dev_addr,
2135 		dma_addr_t dst_dev_addr)
2136 {
2137 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2138 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2139 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2140 	int ret;
2141 
2142 	ret = d40_log_sg_to_lli(sg_src, sg_len,
2143 				src_dev_addr,
2144 				desc->lli_log.src,
2145 				chan->log_def.lcsp1,
2146 				src_info->data_width,
2147 				dst_info->data_width);
2148 
2149 	ret = d40_log_sg_to_lli(sg_dst, sg_len,
2150 				dst_dev_addr,
2151 				desc->lli_log.dst,
2152 				chan->log_def.lcsp3,
2153 				dst_info->data_width,
2154 				src_info->data_width);
2155 
2156 	return ret < 0 ? ret : 0;
2157 }
2158 
2159 static int
2160 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2161 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2162 		unsigned int sg_len, dma_addr_t src_dev_addr,
2163 		dma_addr_t dst_dev_addr)
2164 {
2165 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2166 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2167 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2168 	unsigned long flags = 0;
2169 	int ret;
2170 
2171 	if (desc->cyclic)
2172 		flags |= LLI_CYCLIC | LLI_TERM_INT;
2173 
2174 	ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2175 				desc->lli_phy.src,
2176 				virt_to_phys(desc->lli_phy.src),
2177 				chan->src_def_cfg,
2178 				src_info, dst_info, flags);
2179 
2180 	ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2181 				desc->lli_phy.dst,
2182 				virt_to_phys(desc->lli_phy.dst),
2183 				chan->dst_def_cfg,
2184 				dst_info, src_info, flags);
2185 
2186 	dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2187 				   desc->lli_pool.size, DMA_TO_DEVICE);
2188 
2189 	return ret < 0 ? ret : 0;
2190 }
2191 
2192 static struct d40_desc *
2193 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2194 	      unsigned int sg_len, unsigned long dma_flags)
2195 {
2196 	struct stedma40_chan_cfg *cfg;
2197 	struct d40_desc *desc;
2198 	int ret;
2199 
2200 	desc = d40_desc_get(chan);
2201 	if (!desc)
2202 		return NULL;
2203 
2204 	cfg = &chan->dma_cfg;
2205 	desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2206 					cfg->dst_info.data_width);
2207 	if (desc->lli_len < 0) {
2208 		chan_err(chan, "Unaligned size\n");
2209 		goto free_desc;
2210 	}
2211 
2212 	ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2213 	if (ret < 0) {
2214 		chan_err(chan, "Could not allocate lli\n");
2215 		goto free_desc;
2216 	}
2217 
2218 	desc->lli_current = 0;
2219 	desc->txd.flags = dma_flags;
2220 	desc->txd.tx_submit = d40_tx_submit;
2221 
2222 	dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2223 
2224 	return desc;
2225  free_desc:
2226 	d40_desc_free(chan, desc);
2227 	return NULL;
2228 }
2229 
2230 static struct dma_async_tx_descriptor *
2231 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2232 	    struct scatterlist *sg_dst, unsigned int sg_len,
2233 	    enum dma_transfer_direction direction, unsigned long dma_flags)
2234 {
2235 	struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2236 	dma_addr_t src_dev_addr;
2237 	dma_addr_t dst_dev_addr;
2238 	struct d40_desc *desc;
2239 	unsigned long flags;
2240 	int ret;
2241 
2242 	if (!chan->phy_chan) {
2243 		chan_err(chan, "Cannot prepare unallocated channel\n");
2244 		return NULL;
2245 	}
2246 
2247 	d40_set_runtime_config_write(dchan, &chan->slave_config, direction);
2248 
2249 	spin_lock_irqsave(&chan->lock, flags);
2250 
2251 	desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2252 	if (desc == NULL)
2253 		goto unlock;
2254 
2255 	if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2256 		desc->cyclic = true;
2257 
2258 	src_dev_addr = 0;
2259 	dst_dev_addr = 0;
2260 	if (direction == DMA_DEV_TO_MEM)
2261 		src_dev_addr = chan->runtime_addr;
2262 	else if (direction == DMA_MEM_TO_DEV)
2263 		dst_dev_addr = chan->runtime_addr;
2264 
2265 	if (chan_is_logical(chan))
2266 		ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2267 				      sg_len, src_dev_addr, dst_dev_addr);
2268 	else
2269 		ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2270 				      sg_len, src_dev_addr, dst_dev_addr);
2271 
2272 	if (ret) {
2273 		chan_err(chan, "Failed to prepare %s sg job: %d\n",
2274 			 chan_is_logical(chan) ? "log" : "phy", ret);
2275 		goto free_desc;
2276 	}
2277 
2278 	/*
2279 	 * add descriptor to the prepare queue in order to be able
2280 	 * to free them later in terminate_all
2281 	 */
2282 	list_add_tail(&desc->node, &chan->prepare_queue);
2283 
2284 	spin_unlock_irqrestore(&chan->lock, flags);
2285 
2286 	return &desc->txd;
2287  free_desc:
2288 	d40_desc_free(chan, desc);
2289  unlock:
2290 	spin_unlock_irqrestore(&chan->lock, flags);
2291 	return NULL;
2292 }
2293 
2294 static bool stedma40_filter(struct dma_chan *chan, void *data)
2295 {
2296 	struct stedma40_chan_cfg *info = data;
2297 	struct d40_chan *d40c =
2298 		container_of(chan, struct d40_chan, chan);
2299 	int err;
2300 
2301 	if (data) {
2302 		err = d40_validate_conf(d40c, info);
2303 		if (!err)
2304 			d40c->dma_cfg = *info;
2305 	} else
2306 		err = d40_config_memcpy(d40c);
2307 
2308 	if (!err)
2309 		d40c->configured = true;
2310 
2311 	return err == 0;
2312 }
2313 
2314 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2315 {
2316 	bool realtime = d40c->dma_cfg.realtime;
2317 	bool highprio = d40c->dma_cfg.high_priority;
2318 	u32 rtreg;
2319 	u32 event = D40_TYPE_TO_EVENT(dev_type);
2320 	u32 group = D40_TYPE_TO_GROUP(dev_type);
2321 	u32 bit = BIT(event);
2322 	u32 prioreg;
2323 	struct d40_gen_dmac *dmac = &d40c->base->gen_dmac;
2324 
2325 	rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear;
2326 	/*
2327 	 * Due to a hardware bug, in some cases a logical channel triggered by
2328 	 * a high priority destination event line can generate extra packet
2329 	 * transactions.
2330 	 *
2331 	 * The workaround is to not set the high priority level for the
2332 	 * destination event lines that trigger logical channels.
2333 	 */
2334 	if (!src && chan_is_logical(d40c))
2335 		highprio = false;
2336 
2337 	prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear;
2338 
2339 	/* Destination event lines are stored in the upper halfword */
2340 	if (!src)
2341 		bit <<= 16;
2342 
2343 	writel(bit, d40c->base->virtbase + prioreg + group * 4);
2344 	writel(bit, d40c->base->virtbase + rtreg + group * 4);
2345 }
2346 
2347 static void d40_set_prio_realtime(struct d40_chan *d40c)
2348 {
2349 	if (d40c->base->rev < 3)
2350 		return;
2351 
2352 	if ((d40c->dma_cfg.dir ==  DMA_DEV_TO_MEM) ||
2353 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2354 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, true);
2355 
2356 	if ((d40c->dma_cfg.dir ==  DMA_MEM_TO_DEV) ||
2357 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2358 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, false);
2359 }
2360 
2361 #define D40_DT_FLAGS_MODE(flags)       ((flags >> 0) & 0x1)
2362 #define D40_DT_FLAGS_DIR(flags)        ((flags >> 1) & 0x1)
2363 #define D40_DT_FLAGS_BIG_ENDIAN(flags) ((flags >> 2) & 0x1)
2364 #define D40_DT_FLAGS_FIXED_CHAN(flags) ((flags >> 3) & 0x1)
2365 #define D40_DT_FLAGS_HIGH_PRIO(flags)  ((flags >> 4) & 0x1)
2366 
2367 static struct dma_chan *d40_xlate(struct of_phandle_args *dma_spec,
2368 				  struct of_dma *ofdma)
2369 {
2370 	struct stedma40_chan_cfg cfg;
2371 	dma_cap_mask_t cap;
2372 	u32 flags;
2373 
2374 	memset(&cfg, 0, sizeof(struct stedma40_chan_cfg));
2375 
2376 	dma_cap_zero(cap);
2377 	dma_cap_set(DMA_SLAVE, cap);
2378 
2379 	cfg.dev_type = dma_spec->args[0];
2380 	flags = dma_spec->args[2];
2381 
2382 	switch (D40_DT_FLAGS_MODE(flags)) {
2383 	case 0: cfg.mode = STEDMA40_MODE_LOGICAL; break;
2384 	case 1: cfg.mode = STEDMA40_MODE_PHYSICAL; break;
2385 	}
2386 
2387 	switch (D40_DT_FLAGS_DIR(flags)) {
2388 	case 0:
2389 		cfg.dir = DMA_MEM_TO_DEV;
2390 		cfg.dst_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2391 		break;
2392 	case 1:
2393 		cfg.dir = DMA_DEV_TO_MEM;
2394 		cfg.src_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2395 		break;
2396 	}
2397 
2398 	if (D40_DT_FLAGS_FIXED_CHAN(flags)) {
2399 		cfg.phy_channel = dma_spec->args[1];
2400 		cfg.use_fixed_channel = true;
2401 	}
2402 
2403 	if (D40_DT_FLAGS_HIGH_PRIO(flags))
2404 		cfg.high_priority = true;
2405 
2406 	return dma_request_channel(cap, stedma40_filter, &cfg);
2407 }
2408 
2409 /* DMA ENGINE functions */
2410 static int d40_alloc_chan_resources(struct dma_chan *chan)
2411 {
2412 	int err;
2413 	unsigned long flags;
2414 	struct d40_chan *d40c =
2415 		container_of(chan, struct d40_chan, chan);
2416 	bool is_free_phy;
2417 	spin_lock_irqsave(&d40c->lock, flags);
2418 
2419 	dma_cookie_init(chan);
2420 
2421 	/* If no dma configuration is set use default configuration (memcpy) */
2422 	if (!d40c->configured) {
2423 		err = d40_config_memcpy(d40c);
2424 		if (err) {
2425 			chan_err(d40c, "Failed to configure memcpy channel\n");
2426 			goto mark_last_busy;
2427 		}
2428 	}
2429 
2430 	err = d40_allocate_channel(d40c, &is_free_phy);
2431 	if (err) {
2432 		chan_err(d40c, "Failed to allocate channel\n");
2433 		d40c->configured = false;
2434 		goto mark_last_busy;
2435 	}
2436 
2437 	pm_runtime_get_sync(d40c->base->dev);
2438 
2439 	d40_set_prio_realtime(d40c);
2440 
2441 	if (chan_is_logical(d40c)) {
2442 		if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2443 			d40c->lcpa = d40c->base->lcpa_base +
2444 				d40c->dma_cfg.dev_type * D40_LCPA_CHAN_SIZE;
2445 		else
2446 			d40c->lcpa = d40c->base->lcpa_base +
2447 				d40c->dma_cfg.dev_type *
2448 				D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2449 
2450 		/* Unmask the Global Interrupt Mask. */
2451 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2452 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2453 	}
2454 
2455 	dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2456 		 chan_is_logical(d40c) ? "logical" : "physical",
2457 		 d40c->phy_chan->num,
2458 		 d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2459 
2460 
2461 	/*
2462 	 * Only write channel configuration to the DMA if the physical
2463 	 * resource is free. In case of multiple logical channels
2464 	 * on the same physical resource, only the first write is necessary.
2465 	 */
2466 	if (is_free_phy)
2467 		d40_config_write(d40c);
2468  mark_last_busy:
2469 	pm_runtime_mark_last_busy(d40c->base->dev);
2470 	pm_runtime_put_autosuspend(d40c->base->dev);
2471 	spin_unlock_irqrestore(&d40c->lock, flags);
2472 	return err;
2473 }
2474 
2475 static void d40_free_chan_resources(struct dma_chan *chan)
2476 {
2477 	struct d40_chan *d40c =
2478 		container_of(chan, struct d40_chan, chan);
2479 	int err;
2480 	unsigned long flags;
2481 
2482 	if (d40c->phy_chan == NULL) {
2483 		chan_err(d40c, "Cannot free unallocated channel\n");
2484 		return;
2485 	}
2486 
2487 	spin_lock_irqsave(&d40c->lock, flags);
2488 
2489 	err = d40_free_dma(d40c);
2490 
2491 	if (err)
2492 		chan_err(d40c, "Failed to free channel\n");
2493 	spin_unlock_irqrestore(&d40c->lock, flags);
2494 }
2495 
2496 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2497 						       dma_addr_t dst,
2498 						       dma_addr_t src,
2499 						       size_t size,
2500 						       unsigned long dma_flags)
2501 {
2502 	struct scatterlist dst_sg;
2503 	struct scatterlist src_sg;
2504 
2505 	sg_init_table(&dst_sg, 1);
2506 	sg_init_table(&src_sg, 1);
2507 
2508 	sg_dma_address(&dst_sg) = dst;
2509 	sg_dma_address(&src_sg) = src;
2510 
2511 	sg_dma_len(&dst_sg) = size;
2512 	sg_dma_len(&src_sg) = size;
2513 
2514 	return d40_prep_sg(chan, &src_sg, &dst_sg, 1,
2515 			   DMA_MEM_TO_MEM, dma_flags);
2516 }
2517 
2518 static struct dma_async_tx_descriptor *
2519 d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2520 		  unsigned int sg_len, enum dma_transfer_direction direction,
2521 		  unsigned long dma_flags, void *context)
2522 {
2523 	if (!is_slave_direction(direction))
2524 		return NULL;
2525 
2526 	return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2527 }
2528 
2529 static struct dma_async_tx_descriptor *
2530 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2531 		     size_t buf_len, size_t period_len,
2532 		     enum dma_transfer_direction direction, unsigned long flags)
2533 {
2534 	unsigned int periods = buf_len / period_len;
2535 	struct dma_async_tx_descriptor *txd;
2536 	struct scatterlist *sg;
2537 	int i;
2538 
2539 	sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2540 	if (!sg)
2541 		return NULL;
2542 
2543 	for (i = 0; i < periods; i++) {
2544 		sg_dma_address(&sg[i]) = dma_addr;
2545 		sg_dma_len(&sg[i]) = period_len;
2546 		dma_addr += period_len;
2547 	}
2548 
2549 	sg_chain(sg, periods + 1, sg);
2550 
2551 	txd = d40_prep_sg(chan, sg, sg, periods, direction,
2552 			  DMA_PREP_INTERRUPT);
2553 
2554 	kfree(sg);
2555 
2556 	return txd;
2557 }
2558 
2559 static enum dma_status d40_tx_status(struct dma_chan *chan,
2560 				     dma_cookie_t cookie,
2561 				     struct dma_tx_state *txstate)
2562 {
2563 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2564 	enum dma_status ret;
2565 
2566 	if (d40c->phy_chan == NULL) {
2567 		chan_err(d40c, "Cannot read status of unallocated channel\n");
2568 		return -EINVAL;
2569 	}
2570 
2571 	ret = dma_cookie_status(chan, cookie, txstate);
2572 	if (ret != DMA_COMPLETE && txstate)
2573 		dma_set_residue(txstate, stedma40_residue(chan));
2574 
2575 	if (d40_is_paused(d40c))
2576 		ret = DMA_PAUSED;
2577 
2578 	return ret;
2579 }
2580 
2581 static void d40_issue_pending(struct dma_chan *chan)
2582 {
2583 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2584 	unsigned long flags;
2585 
2586 	if (d40c->phy_chan == NULL) {
2587 		chan_err(d40c, "Channel is not allocated!\n");
2588 		return;
2589 	}
2590 
2591 	spin_lock_irqsave(&d40c->lock, flags);
2592 
2593 	list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2594 
2595 	/* Busy means that queued jobs are already being processed */
2596 	if (!d40c->busy)
2597 		(void) d40_queue_start(d40c);
2598 
2599 	spin_unlock_irqrestore(&d40c->lock, flags);
2600 }
2601 
2602 static int d40_terminate_all(struct dma_chan *chan)
2603 {
2604 	unsigned long flags;
2605 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2606 	int ret;
2607 
2608 	if (d40c->phy_chan == NULL) {
2609 		chan_err(d40c, "Channel is not allocated!\n");
2610 		return -EINVAL;
2611 	}
2612 
2613 	spin_lock_irqsave(&d40c->lock, flags);
2614 
2615 	pm_runtime_get_sync(d40c->base->dev);
2616 	ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2617 	if (ret)
2618 		chan_err(d40c, "Failed to stop channel\n");
2619 
2620 	d40_term_all(d40c);
2621 	pm_runtime_mark_last_busy(d40c->base->dev);
2622 	pm_runtime_put_autosuspend(d40c->base->dev);
2623 	if (d40c->busy) {
2624 		pm_runtime_mark_last_busy(d40c->base->dev);
2625 		pm_runtime_put_autosuspend(d40c->base->dev);
2626 	}
2627 	d40c->busy = false;
2628 
2629 	spin_unlock_irqrestore(&d40c->lock, flags);
2630 	return 0;
2631 }
2632 
2633 static int
2634 dma40_config_to_halfchannel(struct d40_chan *d40c,
2635 			    struct stedma40_half_channel_info *info,
2636 			    u32 maxburst)
2637 {
2638 	int psize;
2639 
2640 	if (chan_is_logical(d40c)) {
2641 		if (maxburst >= 16)
2642 			psize = STEDMA40_PSIZE_LOG_16;
2643 		else if (maxburst >= 8)
2644 			psize = STEDMA40_PSIZE_LOG_8;
2645 		else if (maxburst >= 4)
2646 			psize = STEDMA40_PSIZE_LOG_4;
2647 		else
2648 			psize = STEDMA40_PSIZE_LOG_1;
2649 	} else {
2650 		if (maxburst >= 16)
2651 			psize = STEDMA40_PSIZE_PHY_16;
2652 		else if (maxburst >= 8)
2653 			psize = STEDMA40_PSIZE_PHY_8;
2654 		else if (maxburst >= 4)
2655 			psize = STEDMA40_PSIZE_PHY_4;
2656 		else
2657 			psize = STEDMA40_PSIZE_PHY_1;
2658 	}
2659 
2660 	info->psize = psize;
2661 	info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2662 
2663 	return 0;
2664 }
2665 
2666 static int d40_set_runtime_config(struct dma_chan *chan,
2667 				  struct dma_slave_config *config)
2668 {
2669 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2670 
2671 	memcpy(&d40c->slave_config, config, sizeof(*config));
2672 
2673 	return 0;
2674 }
2675 
2676 /* Runtime reconfiguration extension */
2677 static int d40_set_runtime_config_write(struct dma_chan *chan,
2678 				  struct dma_slave_config *config,
2679 				  enum dma_transfer_direction direction)
2680 {
2681 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2682 	struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2683 	enum dma_slave_buswidth src_addr_width, dst_addr_width;
2684 	dma_addr_t config_addr;
2685 	u32 src_maxburst, dst_maxburst;
2686 	int ret;
2687 
2688 	if (d40c->phy_chan == NULL) {
2689 		chan_err(d40c, "Channel is not allocated!\n");
2690 		return -EINVAL;
2691 	}
2692 
2693 	src_addr_width = config->src_addr_width;
2694 	src_maxburst = config->src_maxburst;
2695 	dst_addr_width = config->dst_addr_width;
2696 	dst_maxburst = config->dst_maxburst;
2697 
2698 	if (direction == DMA_DEV_TO_MEM) {
2699 		config_addr = config->src_addr;
2700 
2701 		if (cfg->dir != DMA_DEV_TO_MEM)
2702 			dev_dbg(d40c->base->dev,
2703 				"channel was not configured for peripheral "
2704 				"to memory transfer (%d) overriding\n",
2705 				cfg->dir);
2706 		cfg->dir = DMA_DEV_TO_MEM;
2707 
2708 		/* Configure the memory side */
2709 		if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2710 			dst_addr_width = src_addr_width;
2711 		if (dst_maxburst == 0)
2712 			dst_maxburst = src_maxburst;
2713 
2714 	} else if (direction == DMA_MEM_TO_DEV) {
2715 		config_addr = config->dst_addr;
2716 
2717 		if (cfg->dir != DMA_MEM_TO_DEV)
2718 			dev_dbg(d40c->base->dev,
2719 				"channel was not configured for memory "
2720 				"to peripheral transfer (%d) overriding\n",
2721 				cfg->dir);
2722 		cfg->dir = DMA_MEM_TO_DEV;
2723 
2724 		/* Configure the memory side */
2725 		if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2726 			src_addr_width = dst_addr_width;
2727 		if (src_maxburst == 0)
2728 			src_maxburst = dst_maxburst;
2729 	} else {
2730 		dev_err(d40c->base->dev,
2731 			"unrecognized channel direction %d\n",
2732 			direction);
2733 		return -EINVAL;
2734 	}
2735 
2736 	if (config_addr <= 0) {
2737 		dev_err(d40c->base->dev, "no address supplied\n");
2738 		return -EINVAL;
2739 	}
2740 
2741 	if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2742 		dev_err(d40c->base->dev,
2743 			"src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2744 			src_maxburst,
2745 			src_addr_width,
2746 			dst_maxburst,
2747 			dst_addr_width);
2748 		return -EINVAL;
2749 	}
2750 
2751 	if (src_maxburst > 16) {
2752 		src_maxburst = 16;
2753 		dst_maxburst = src_maxburst * src_addr_width / dst_addr_width;
2754 	} else if (dst_maxburst > 16) {
2755 		dst_maxburst = 16;
2756 		src_maxburst = dst_maxburst * dst_addr_width / src_addr_width;
2757 	}
2758 
2759 	/* Only valid widths are; 1, 2, 4 and 8. */
2760 	if (src_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2761 	    src_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2762 	    dst_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2763 	    dst_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2764 	    !is_power_of_2(src_addr_width) ||
2765 	    !is_power_of_2(dst_addr_width))
2766 		return -EINVAL;
2767 
2768 	cfg->src_info.data_width = src_addr_width;
2769 	cfg->dst_info.data_width = dst_addr_width;
2770 
2771 	ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2772 					  src_maxburst);
2773 	if (ret)
2774 		return ret;
2775 
2776 	ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2777 					  dst_maxburst);
2778 	if (ret)
2779 		return ret;
2780 
2781 	/* Fill in register values */
2782 	if (chan_is_logical(d40c))
2783 		d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2784 	else
2785 		d40_phy_cfg(cfg, &d40c->src_def_cfg, &d40c->dst_def_cfg);
2786 
2787 	/* These settings will take precedence later */
2788 	d40c->runtime_addr = config_addr;
2789 	d40c->runtime_direction = direction;
2790 	dev_dbg(d40c->base->dev,
2791 		"configured channel %s for %s, data width %d/%d, "
2792 		"maxburst %d/%d elements, LE, no flow control\n",
2793 		dma_chan_name(chan),
2794 		(direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2795 		src_addr_width, dst_addr_width,
2796 		src_maxburst, dst_maxburst);
2797 
2798 	return 0;
2799 }
2800 
2801 /* Initialization functions */
2802 
2803 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2804 				 struct d40_chan *chans, int offset,
2805 				 int num_chans)
2806 {
2807 	int i = 0;
2808 	struct d40_chan *d40c;
2809 
2810 	INIT_LIST_HEAD(&dma->channels);
2811 
2812 	for (i = offset; i < offset + num_chans; i++) {
2813 		d40c = &chans[i];
2814 		d40c->base = base;
2815 		d40c->chan.device = dma;
2816 
2817 		spin_lock_init(&d40c->lock);
2818 
2819 		d40c->log_num = D40_PHY_CHAN;
2820 
2821 		INIT_LIST_HEAD(&d40c->done);
2822 		INIT_LIST_HEAD(&d40c->active);
2823 		INIT_LIST_HEAD(&d40c->queue);
2824 		INIT_LIST_HEAD(&d40c->pending_queue);
2825 		INIT_LIST_HEAD(&d40c->client);
2826 		INIT_LIST_HEAD(&d40c->prepare_queue);
2827 
2828 		tasklet_setup(&d40c->tasklet, dma_tasklet);
2829 
2830 		list_add_tail(&d40c->chan.device_node,
2831 			      &dma->channels);
2832 	}
2833 }
2834 
2835 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2836 {
2837 	if (dma_has_cap(DMA_SLAVE, dev->cap_mask)) {
2838 		dev->device_prep_slave_sg = d40_prep_slave_sg;
2839 		dev->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
2840 	}
2841 
2842 	if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2843 		dev->device_prep_dma_memcpy = d40_prep_memcpy;
2844 		dev->directions = BIT(DMA_MEM_TO_MEM);
2845 		/*
2846 		 * This controller can only access address at even
2847 		 * 32bit boundaries, i.e. 2^2
2848 		 */
2849 		dev->copy_align = DMAENGINE_ALIGN_4_BYTES;
2850 	}
2851 
2852 	if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2853 		dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2854 
2855 	dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2856 	dev->device_free_chan_resources = d40_free_chan_resources;
2857 	dev->device_issue_pending = d40_issue_pending;
2858 	dev->device_tx_status = d40_tx_status;
2859 	dev->device_config = d40_set_runtime_config;
2860 	dev->device_pause = d40_pause;
2861 	dev->device_resume = d40_resume;
2862 	dev->device_terminate_all = d40_terminate_all;
2863 	dev->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
2864 	dev->dev = base->dev;
2865 }
2866 
2867 static int __init d40_dmaengine_init(struct d40_base *base,
2868 				     int num_reserved_chans)
2869 {
2870 	int err ;
2871 
2872 	d40_chan_init(base, &base->dma_slave, base->log_chans,
2873 		      0, base->num_log_chans);
2874 
2875 	dma_cap_zero(base->dma_slave.cap_mask);
2876 	dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2877 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2878 
2879 	d40_ops_init(base, &base->dma_slave);
2880 
2881 	err = dmaenginem_async_device_register(&base->dma_slave);
2882 
2883 	if (err) {
2884 		d40_err(base->dev, "Failed to register slave channels\n");
2885 		goto exit;
2886 	}
2887 
2888 	d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2889 		      base->num_log_chans, base->num_memcpy_chans);
2890 
2891 	dma_cap_zero(base->dma_memcpy.cap_mask);
2892 	dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2893 
2894 	d40_ops_init(base, &base->dma_memcpy);
2895 
2896 	err = dmaenginem_async_device_register(&base->dma_memcpy);
2897 
2898 	if (err) {
2899 		d40_err(base->dev,
2900 			"Failed to register memcpy only channels\n");
2901 		goto exit;
2902 	}
2903 
2904 	d40_chan_init(base, &base->dma_both, base->phy_chans,
2905 		      0, num_reserved_chans);
2906 
2907 	dma_cap_zero(base->dma_both.cap_mask);
2908 	dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2909 	dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2910 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2911 
2912 	d40_ops_init(base, &base->dma_both);
2913 	err = dmaenginem_async_device_register(&base->dma_both);
2914 
2915 	if (err) {
2916 		d40_err(base->dev,
2917 			"Failed to register logical and physical capable channels\n");
2918 		goto exit;
2919 	}
2920 	return 0;
2921  exit:
2922 	return err;
2923 }
2924 
2925 /* Suspend resume functionality */
2926 #ifdef CONFIG_PM_SLEEP
2927 static int dma40_suspend(struct device *dev)
2928 {
2929 	struct d40_base *base = dev_get_drvdata(dev);
2930 	int ret;
2931 
2932 	ret = pm_runtime_force_suspend(dev);
2933 	if (ret)
2934 		return ret;
2935 
2936 	if (base->lcpa_regulator)
2937 		ret = regulator_disable(base->lcpa_regulator);
2938 	return ret;
2939 }
2940 
2941 static int dma40_resume(struct device *dev)
2942 {
2943 	struct d40_base *base = dev_get_drvdata(dev);
2944 	int ret = 0;
2945 
2946 	if (base->lcpa_regulator) {
2947 		ret = regulator_enable(base->lcpa_regulator);
2948 		if (ret)
2949 			return ret;
2950 	}
2951 
2952 	return pm_runtime_force_resume(dev);
2953 }
2954 #endif
2955 
2956 #ifdef CONFIG_PM
2957 static void dma40_backup(void __iomem *baseaddr, u32 *backup,
2958 			 u32 *regaddr, int num, bool save)
2959 {
2960 	int i;
2961 
2962 	for (i = 0; i < num; i++) {
2963 		void __iomem *addr = baseaddr + regaddr[i];
2964 
2965 		if (save)
2966 			backup[i] = readl_relaxed(addr);
2967 		else
2968 			writel_relaxed(backup[i], addr);
2969 	}
2970 }
2971 
2972 static void d40_save_restore_registers(struct d40_base *base, bool save)
2973 {
2974 	int i;
2975 
2976 	/* Save/Restore channel specific registers */
2977 	for (i = 0; i < base->num_phy_chans; i++) {
2978 		void __iomem *addr;
2979 		int idx;
2980 
2981 		if (base->phy_res[i].reserved)
2982 			continue;
2983 
2984 		addr = base->virtbase + D40_DREG_PCBASE + i * D40_DREG_PCDELTA;
2985 		idx = i * ARRAY_SIZE(d40_backup_regs_chan);
2986 
2987 		dma40_backup(addr, &base->reg_val_backup_chan[idx],
2988 			     d40_backup_regs_chan,
2989 			     ARRAY_SIZE(d40_backup_regs_chan),
2990 			     save);
2991 	}
2992 
2993 	/* Save/Restore global registers */
2994 	dma40_backup(base->virtbase, base->reg_val_backup,
2995 		     d40_backup_regs, ARRAY_SIZE(d40_backup_regs),
2996 		     save);
2997 
2998 	/* Save/Restore registers only existing on dma40 v3 and later */
2999 	if (base->gen_dmac.backup)
3000 		dma40_backup(base->virtbase, base->reg_val_backup_v4,
3001 			     base->gen_dmac.backup,
3002 			base->gen_dmac.backup_size,
3003 			save);
3004 }
3005 
3006 static int dma40_runtime_suspend(struct device *dev)
3007 {
3008 	struct d40_base *base = dev_get_drvdata(dev);
3009 
3010 	d40_save_restore_registers(base, true);
3011 
3012 	/* Don't disable/enable clocks for v1 due to HW bugs */
3013 	if (base->rev != 1)
3014 		writel_relaxed(base->gcc_pwr_off_mask,
3015 			       base->virtbase + D40_DREG_GCC);
3016 
3017 	return 0;
3018 }
3019 
3020 static int dma40_runtime_resume(struct device *dev)
3021 {
3022 	struct d40_base *base = dev_get_drvdata(dev);
3023 
3024 	d40_save_restore_registers(base, false);
3025 
3026 	writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
3027 		       base->virtbase + D40_DREG_GCC);
3028 	return 0;
3029 }
3030 #endif
3031 
3032 static const struct dev_pm_ops dma40_pm_ops = {
3033 	SET_LATE_SYSTEM_SLEEP_PM_OPS(dma40_suspend, dma40_resume)
3034 	SET_RUNTIME_PM_OPS(dma40_runtime_suspend,
3035 				dma40_runtime_resume,
3036 				NULL)
3037 };
3038 
3039 /* Initialization functions. */
3040 
3041 static int __init d40_phy_res_init(struct d40_base *base)
3042 {
3043 	int i;
3044 	int num_phy_chans_avail = 0;
3045 	u32 val[2];
3046 	int odd_even_bit = -2;
3047 	int gcc = D40_DREG_GCC_ENA;
3048 
3049 	val[0] = readl(base->virtbase + D40_DREG_PRSME);
3050 	val[1] = readl(base->virtbase + D40_DREG_PRSMO);
3051 
3052 	for (i = 0; i < base->num_phy_chans; i++) {
3053 		base->phy_res[i].num = i;
3054 		odd_even_bit += 2 * ((i % 2) == 0);
3055 		if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
3056 			/* Mark security only channels as occupied */
3057 			base->phy_res[i].allocated_src = D40_ALLOC_PHY;
3058 			base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
3059 			base->phy_res[i].reserved = true;
3060 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3061 						       D40_DREG_GCC_SRC);
3062 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3063 						       D40_DREG_GCC_DST);
3064 
3065 
3066 		} else {
3067 			base->phy_res[i].allocated_src = D40_ALLOC_FREE;
3068 			base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
3069 			base->phy_res[i].reserved = false;
3070 			num_phy_chans_avail++;
3071 		}
3072 		spin_lock_init(&base->phy_res[i].lock);
3073 	}
3074 
3075 	/* Mark disabled channels as occupied */
3076 	for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
3077 		int chan = base->plat_data->disabled_channels[i];
3078 
3079 		base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
3080 		base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
3081 		base->phy_res[chan].reserved = true;
3082 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3083 					       D40_DREG_GCC_SRC);
3084 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3085 					       D40_DREG_GCC_DST);
3086 		num_phy_chans_avail--;
3087 	}
3088 
3089 	/* Mark soft_lli channels */
3090 	for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) {
3091 		int chan = base->plat_data->soft_lli_chans[i];
3092 
3093 		base->phy_res[chan].use_soft_lli = true;
3094 	}
3095 
3096 	dev_info(base->dev, "%d of %d physical DMA channels available\n",
3097 		 num_phy_chans_avail, base->num_phy_chans);
3098 
3099 	/* Verify settings extended vs standard */
3100 	val[0] = readl(base->virtbase + D40_DREG_PRTYP);
3101 
3102 	for (i = 0; i < base->num_phy_chans; i++) {
3103 
3104 		if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
3105 		    (val[0] & 0x3) != 1)
3106 			dev_info(base->dev,
3107 				 "[%s] INFO: channel %d is misconfigured (%d)\n",
3108 				 __func__, i, val[0] & 0x3);
3109 
3110 		val[0] = val[0] >> 2;
3111 	}
3112 
3113 	/*
3114 	 * To keep things simple, Enable all clocks initially.
3115 	 * The clocks will get managed later post channel allocation.
3116 	 * The clocks for the event lines on which reserved channels exists
3117 	 * are not managed here.
3118 	 */
3119 	writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3120 	base->gcc_pwr_off_mask = gcc;
3121 
3122 	return num_phy_chans_avail;
3123 }
3124 
3125 /* Called from the registered devm action */
3126 static void d40_drop_kmem_cache_action(void *d)
3127 {
3128 	struct kmem_cache *desc_slab = d;
3129 
3130 	kmem_cache_destroy(desc_slab);
3131 }
3132 
3133 static int __init d40_hw_detect_init(struct platform_device *pdev,
3134 				     struct d40_base **retbase)
3135 {
3136 	struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3137 	struct device *dev = &pdev->dev;
3138 	struct clk *clk;
3139 	void __iomem *virtbase;
3140 	struct d40_base *base;
3141 	int num_log_chans;
3142 	int num_phy_chans;
3143 	int num_memcpy_chans;
3144 	int i;
3145 	u32 pid;
3146 	u32 cid;
3147 	u8 rev;
3148 	int ret;
3149 
3150 	clk = devm_clk_get_enabled(dev, NULL);
3151 	if (IS_ERR(clk))
3152 		return PTR_ERR(clk);
3153 
3154 	/* Get IO for DMAC base address */
3155 	virtbase = devm_platform_ioremap_resource_byname(pdev, "base");
3156 	if (IS_ERR(virtbase))
3157 		return PTR_ERR(virtbase);
3158 
3159 	/* This is just a regular AMBA PrimeCell ID actually */
3160 	for (pid = 0, i = 0; i < 4; i++)
3161 		pid |= (readl(virtbase + SZ_4K - 0x20 + 4 * i)
3162 			& 255) << (i * 8);
3163 	for (cid = 0, i = 0; i < 4; i++)
3164 		cid |= (readl(virtbase + SZ_4K - 0x10 + 4 * i)
3165 			& 255) << (i * 8);
3166 
3167 	if (cid != AMBA_CID) {
3168 		d40_err(dev, "Unknown hardware! No PrimeCell ID\n");
3169 		return -EINVAL;
3170 	}
3171 	if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
3172 		d40_err(dev, "Unknown designer! Got %x wanted %x\n",
3173 			AMBA_MANF_BITS(pid),
3174 			AMBA_VENDOR_ST);
3175 		return -EINVAL;
3176 	}
3177 	/*
3178 	 * HW revision:
3179 	 * DB8500ed has revision 0
3180 	 * ? has revision 1
3181 	 * DB8500v1 has revision 2
3182 	 * DB8500v2 has revision 3
3183 	 * AP9540v1 has revision 4
3184 	 * DB8540v1 has revision 4
3185 	 */
3186 	rev = AMBA_REV_BITS(pid);
3187 	if (rev < 2) {
3188 		d40_err(dev, "hardware revision: %d is not supported", rev);
3189 		return -EINVAL;
3190 	}
3191 
3192 	/* The number of physical channels on this HW */
3193 	if (plat_data->num_of_phy_chans)
3194 		num_phy_chans = plat_data->num_of_phy_chans;
3195 	else
3196 		num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
3197 
3198 	/* The number of channels used for memcpy */
3199 	if (plat_data->num_of_memcpy_chans)
3200 		num_memcpy_chans = plat_data->num_of_memcpy_chans;
3201 	else
3202 		num_memcpy_chans = ARRAY_SIZE(dma40_memcpy_channels);
3203 
3204 	num_log_chans = num_phy_chans * D40_MAX_LOG_CHAN_PER_PHY;
3205 
3206 	dev_info(dev,
3207 		 "hardware rev: %d with %d physical and %d logical channels\n",
3208 		 rev, num_phy_chans, num_log_chans);
3209 
3210 	base = devm_kzalloc(dev,
3211 		ALIGN(sizeof(struct d40_base), 4) +
3212 		(num_phy_chans + num_log_chans + num_memcpy_chans) *
3213 		sizeof(struct d40_chan), GFP_KERNEL);
3214 
3215 	if (!base)
3216 		return -ENOMEM;
3217 
3218 	base->rev = rev;
3219 	base->clk = clk;
3220 	base->num_memcpy_chans = num_memcpy_chans;
3221 	base->num_phy_chans = num_phy_chans;
3222 	base->num_log_chans = num_log_chans;
3223 	base->virtbase = virtbase;
3224 	base->plat_data = plat_data;
3225 	base->dev = dev;
3226 	base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3227 	base->log_chans = &base->phy_chans[num_phy_chans];
3228 
3229 	if (base->plat_data->num_of_phy_chans == 14) {
3230 		base->gen_dmac.backup = d40_backup_regs_v4b;
3231 		base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
3232 		base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
3233 		base->gen_dmac.interrupt_clear = D40_DREG_CPCICR;
3234 		base->gen_dmac.realtime_en = D40_DREG_CRSEG1;
3235 		base->gen_dmac.realtime_clear = D40_DREG_CRCEG1;
3236 		base->gen_dmac.high_prio_en = D40_DREG_CPSEG1;
3237 		base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1;
3238 		base->gen_dmac.il = il_v4b;
3239 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4b);
3240 		base->gen_dmac.init_reg = dma_init_reg_v4b;
3241 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
3242 	} else {
3243 		if (base->rev >= 3) {
3244 			base->gen_dmac.backup = d40_backup_regs_v4a;
3245 			base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
3246 		}
3247 		base->gen_dmac.interrupt_en = D40_DREG_PCMIS;
3248 		base->gen_dmac.interrupt_clear = D40_DREG_PCICR;
3249 		base->gen_dmac.realtime_en = D40_DREG_RSEG1;
3250 		base->gen_dmac.realtime_clear = D40_DREG_RCEG1;
3251 		base->gen_dmac.high_prio_en = D40_DREG_PSEG1;
3252 		base->gen_dmac.high_prio_clear = D40_DREG_PCEG1;
3253 		base->gen_dmac.il = il_v4a;
3254 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4a);
3255 		base->gen_dmac.init_reg = dma_init_reg_v4a;
3256 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a);
3257 	}
3258 
3259 	base->phy_res = devm_kcalloc(dev, num_phy_chans,
3260 				     sizeof(*base->phy_res),
3261 				     GFP_KERNEL);
3262 	if (!base->phy_res)
3263 		return -ENOMEM;
3264 
3265 	base->lookup_phy_chans = devm_kcalloc(dev, num_phy_chans,
3266 					      sizeof(*base->lookup_phy_chans),
3267 					      GFP_KERNEL);
3268 	if (!base->lookup_phy_chans)
3269 		return -ENOMEM;
3270 
3271 	base->lookup_log_chans = devm_kcalloc(dev, num_log_chans,
3272 					      sizeof(*base->lookup_log_chans),
3273 					      GFP_KERNEL);
3274 	if (!base->lookup_log_chans)
3275 		return -ENOMEM;
3276 
3277 	base->reg_val_backup_chan = devm_kmalloc_array(dev, base->num_phy_chans,
3278 						  sizeof(d40_backup_regs_chan),
3279 						  GFP_KERNEL);
3280 	if (!base->reg_val_backup_chan)
3281 		return -ENOMEM;
3282 
3283 	base->lcla_pool.alloc_map = devm_kcalloc(dev, num_phy_chans
3284 					    * D40_LCLA_LINK_PER_EVENT_GRP,
3285 					    sizeof(*base->lcla_pool.alloc_map),
3286 					    GFP_KERNEL);
3287 	if (!base->lcla_pool.alloc_map)
3288 		return -ENOMEM;
3289 
3290 	base->regs_interrupt = devm_kmalloc_array(dev, base->gen_dmac.il_size,
3291 					     sizeof(*base->regs_interrupt),
3292 					     GFP_KERNEL);
3293 	if (!base->regs_interrupt)
3294 		return -ENOMEM;
3295 
3296 	base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3297 					    0, SLAB_HWCACHE_ALIGN,
3298 					    NULL);
3299 	if (!base->desc_slab)
3300 		return -ENOMEM;
3301 
3302 	ret = devm_add_action_or_reset(dev, d40_drop_kmem_cache_action,
3303 				       base->desc_slab);
3304 	if (ret)
3305 		return ret;
3306 
3307 	*retbase = base;
3308 
3309 	return 0;
3310 }
3311 
3312 static void __init d40_hw_init(struct d40_base *base)
3313 {
3314 
3315 	int i;
3316 	u32 prmseo[2] = {0, 0};
3317 	u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3318 	u32 pcmis = 0;
3319 	u32 pcicr = 0;
3320 	struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg;
3321 	u32 reg_size = base->gen_dmac.init_reg_size;
3322 
3323 	for (i = 0; i < reg_size; i++)
3324 		writel(dma_init_reg[i].val,
3325 		       base->virtbase + dma_init_reg[i].reg);
3326 
3327 	/* Configure all our dma channels to default settings */
3328 	for (i = 0; i < base->num_phy_chans; i++) {
3329 
3330 		activeo[i % 2] = activeo[i % 2] << 2;
3331 
3332 		if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3333 		    == D40_ALLOC_PHY) {
3334 			activeo[i % 2] |= 3;
3335 			continue;
3336 		}
3337 
3338 		/* Enable interrupt # */
3339 		pcmis = (pcmis << 1) | 1;
3340 
3341 		/* Clear interrupt # */
3342 		pcicr = (pcicr << 1) | 1;
3343 
3344 		/* Set channel to physical mode */
3345 		prmseo[i % 2] = prmseo[i % 2] << 2;
3346 		prmseo[i % 2] |= 1;
3347 
3348 	}
3349 
3350 	writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3351 	writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3352 	writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3353 	writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3354 
3355 	/* Write which interrupt to enable */
3356 	writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en);
3357 
3358 	/* Write which interrupt to clear */
3359 	writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear);
3360 
3361 	/* These are __initdata and cannot be accessed after init */
3362 	base->gen_dmac.init_reg = NULL;
3363 	base->gen_dmac.init_reg_size = 0;
3364 }
3365 
3366 static int __init d40_lcla_allocate(struct d40_base *base)
3367 {
3368 	struct d40_lcla_pool *pool = &base->lcla_pool;
3369 	unsigned long *page_list;
3370 	int i, j;
3371 	int ret;
3372 
3373 	/*
3374 	 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3375 	 * To full fill this hardware requirement without wasting 256 kb
3376 	 * we allocate pages until we get an aligned one.
3377 	 */
3378 	page_list = kmalloc_array(MAX_LCLA_ALLOC_ATTEMPTS,
3379 				  sizeof(*page_list),
3380 				  GFP_KERNEL);
3381 	if (!page_list)
3382 		return -ENOMEM;
3383 
3384 	/* Calculating how many pages that are required */
3385 	base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3386 
3387 	for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3388 		page_list[i] = __get_free_pages(GFP_KERNEL,
3389 						base->lcla_pool.pages);
3390 		if (!page_list[i]) {
3391 
3392 			d40_err(base->dev, "Failed to allocate %d pages.\n",
3393 				base->lcla_pool.pages);
3394 			ret = -ENOMEM;
3395 
3396 			for (j = 0; j < i; j++)
3397 				free_pages(page_list[j], base->lcla_pool.pages);
3398 			goto free_page_list;
3399 		}
3400 
3401 		if ((virt_to_phys((void *)page_list[i]) &
3402 		     (LCLA_ALIGNMENT - 1)) == 0)
3403 			break;
3404 	}
3405 
3406 	for (j = 0; j < i; j++)
3407 		free_pages(page_list[j], base->lcla_pool.pages);
3408 
3409 	if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3410 		base->lcla_pool.base = (void *)page_list[i];
3411 	} else {
3412 		/*
3413 		 * After many attempts and no success with finding the correct
3414 		 * alignment, try with allocating a big buffer.
3415 		 */
3416 		dev_warn(base->dev,
3417 			 "[%s] Failed to get %d pages @ 18 bit align.\n",
3418 			 __func__, base->lcla_pool.pages);
3419 		base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3420 							 base->num_phy_chans +
3421 							 LCLA_ALIGNMENT,
3422 							 GFP_KERNEL);
3423 		if (!base->lcla_pool.base_unaligned) {
3424 			ret = -ENOMEM;
3425 			goto free_page_list;
3426 		}
3427 
3428 		base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3429 						 LCLA_ALIGNMENT);
3430 	}
3431 
3432 	pool->dma_addr = dma_map_single(base->dev, pool->base,
3433 					SZ_1K * base->num_phy_chans,
3434 					DMA_TO_DEVICE);
3435 	if (dma_mapping_error(base->dev, pool->dma_addr)) {
3436 		pool->dma_addr = 0;
3437 		ret = -ENOMEM;
3438 		goto free_page_list;
3439 	}
3440 
3441 	writel(virt_to_phys(base->lcla_pool.base),
3442 	       base->virtbase + D40_DREG_LCLA);
3443 	ret = 0;
3444  free_page_list:
3445 	kfree(page_list);
3446 	return ret;
3447 }
3448 
3449 static int __init d40_of_probe(struct device *dev,
3450 			       struct device_node *np)
3451 {
3452 	struct stedma40_platform_data *pdata;
3453 	int num_phy = 0, num_memcpy = 0, num_disabled = 0;
3454 	const __be32 *list;
3455 
3456 	pdata = devm_kzalloc(dev, sizeof(*pdata), GFP_KERNEL);
3457 	if (!pdata)
3458 		return -ENOMEM;
3459 
3460 	/* If absent this value will be obtained from h/w. */
3461 	of_property_read_u32(np, "dma-channels", &num_phy);
3462 	if (num_phy > 0)
3463 		pdata->num_of_phy_chans = num_phy;
3464 
3465 	list = of_get_property(np, "memcpy-channels", &num_memcpy);
3466 	num_memcpy /= sizeof(*list);
3467 
3468 	if (num_memcpy > D40_MEMCPY_MAX_CHANS || num_memcpy <= 0) {
3469 		d40_err(dev,
3470 			"Invalid number of memcpy channels specified (%d)\n",
3471 			num_memcpy);
3472 		return -EINVAL;
3473 	}
3474 	pdata->num_of_memcpy_chans = num_memcpy;
3475 
3476 	of_property_read_u32_array(np, "memcpy-channels",
3477 				   dma40_memcpy_channels,
3478 				   num_memcpy);
3479 
3480 	list = of_get_property(np, "disabled-channels", &num_disabled);
3481 	num_disabled /= sizeof(*list);
3482 
3483 	if (num_disabled >= STEDMA40_MAX_PHYS || num_disabled < 0) {
3484 		d40_err(dev,
3485 			"Invalid number of disabled channels specified (%d)\n",
3486 			num_disabled);
3487 		return -EINVAL;
3488 	}
3489 
3490 	of_property_read_u32_array(np, "disabled-channels",
3491 				   pdata->disabled_channels,
3492 				   num_disabled);
3493 	pdata->disabled_channels[num_disabled] = -1;
3494 
3495 	dev->platform_data = pdata;
3496 
3497 	return 0;
3498 }
3499 
3500 static int __init d40_probe(struct platform_device *pdev)
3501 {
3502 	struct device *dev = &pdev->dev;
3503 	struct device_node *np = pdev->dev.of_node;
3504 	struct device_node *np_lcpa;
3505 	struct d40_base *base;
3506 	struct resource *res;
3507 	struct resource res_lcpa;
3508 	int num_reserved_chans;
3509 	u32 val;
3510 	int ret;
3511 
3512 	if (d40_of_probe(dev, np)) {
3513 		ret = -ENOMEM;
3514 		goto report_failure;
3515 	}
3516 
3517 	ret = d40_hw_detect_init(pdev, &base);
3518 	if (ret)
3519 		goto report_failure;
3520 
3521 	num_reserved_chans = d40_phy_res_init(base);
3522 
3523 	platform_set_drvdata(pdev, base);
3524 
3525 	spin_lock_init(&base->interrupt_lock);
3526 	spin_lock_init(&base->execmd_lock);
3527 
3528 	/* Get IO for logical channel parameter address (LCPA) */
3529 	np_lcpa = of_parse_phandle(np, "sram", 0);
3530 	if (!np_lcpa) {
3531 		dev_err(dev, "no LCPA SRAM node\n");
3532 		ret = -EINVAL;
3533 		goto report_failure;
3534 	}
3535 	/* This is no device so read the address directly from the node */
3536 	ret = of_address_to_resource(np_lcpa, 0, &res_lcpa);
3537 	if (ret) {
3538 		dev_err(dev, "no LCPA SRAM resource\n");
3539 		goto report_failure;
3540 	}
3541 	base->lcpa_size = resource_size(&res_lcpa);
3542 	base->phy_lcpa = res_lcpa.start;
3543 	dev_info(dev, "found LCPA SRAM at %pad, size %pa\n",
3544 		 &base->phy_lcpa, &base->lcpa_size);
3545 
3546 	/* We make use of ESRAM memory for this. */
3547 	val = readl(base->virtbase + D40_DREG_LCPA);
3548 	if (base->phy_lcpa != val && val != 0) {
3549 		dev_warn(dev,
3550 			 "[%s] Mismatch LCPA dma 0x%x, def %08x\n",
3551 			 __func__, val, (u32)base->phy_lcpa);
3552 	} else
3553 		writel(base->phy_lcpa, base->virtbase + D40_DREG_LCPA);
3554 
3555 	base->lcpa_base = devm_ioremap(dev, base->phy_lcpa, base->lcpa_size);
3556 	if (!base->lcpa_base) {
3557 		ret = -ENOMEM;
3558 		d40_err(dev, "Failed to ioremap LCPA region\n");
3559 		goto report_failure;
3560 	}
3561 	/* If lcla has to be located in ESRAM we don't need to allocate */
3562 	if (base->plat_data->use_esram_lcla) {
3563 		res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3564 							"lcla_esram");
3565 		if (!res) {
3566 			ret = -ENOENT;
3567 			d40_err(dev,
3568 				"No \"lcla_esram\" memory resource\n");
3569 			goto report_failure;
3570 		}
3571 		base->lcla_pool.base = devm_ioremap(dev, res->start,
3572 						    resource_size(res));
3573 		if (!base->lcla_pool.base) {
3574 			ret = -ENOMEM;
3575 			d40_err(dev, "Failed to ioremap LCLA region\n");
3576 			goto report_failure;
3577 		}
3578 		writel(res->start, base->virtbase + D40_DREG_LCLA);
3579 
3580 	} else {
3581 		ret = d40_lcla_allocate(base);
3582 		if (ret) {
3583 			d40_err(dev, "Failed to allocate LCLA area\n");
3584 			goto destroy_cache;
3585 		}
3586 	}
3587 
3588 	spin_lock_init(&base->lcla_pool.lock);
3589 
3590 	base->irq = platform_get_irq(pdev, 0);
3591 	if (base->irq < 0) {
3592 		ret = base->irq;
3593 		goto destroy_cache;
3594 	}
3595 
3596 	ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3597 	if (ret) {
3598 		d40_err(dev, "No IRQ defined\n");
3599 		goto destroy_cache;
3600 	}
3601 
3602 	if (base->plat_data->use_esram_lcla) {
3603 
3604 		base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3605 		if (IS_ERR(base->lcpa_regulator)) {
3606 			d40_err(dev, "Failed to get lcpa_regulator\n");
3607 			ret = PTR_ERR(base->lcpa_regulator);
3608 			base->lcpa_regulator = NULL;
3609 			goto destroy_cache;
3610 		}
3611 
3612 		ret = regulator_enable(base->lcpa_regulator);
3613 		if (ret) {
3614 			d40_err(dev,
3615 				"Failed to enable lcpa_regulator\n");
3616 			regulator_put(base->lcpa_regulator);
3617 			base->lcpa_regulator = NULL;
3618 			goto destroy_cache;
3619 		}
3620 	}
3621 
3622 	writel_relaxed(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3623 
3624 	pm_runtime_irq_safe(base->dev);
3625 	pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3626 	pm_runtime_use_autosuspend(base->dev);
3627 	pm_runtime_mark_last_busy(base->dev);
3628 	pm_runtime_set_active(base->dev);
3629 	pm_runtime_enable(base->dev);
3630 
3631 	ret = d40_dmaengine_init(base, num_reserved_chans);
3632 	if (ret)
3633 		goto destroy_cache;
3634 
3635 	ret = dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
3636 	if (ret) {
3637 		d40_err(dev, "Failed to set dma max seg size\n");
3638 		goto destroy_cache;
3639 	}
3640 
3641 	d40_hw_init(base);
3642 
3643 	ret = of_dma_controller_register(np, d40_xlate, NULL);
3644 	if (ret) {
3645 		dev_err(dev,
3646 			"could not register of_dma_controller\n");
3647 		goto destroy_cache;
3648 	}
3649 
3650 	dev_info(base->dev, "initialized\n");
3651 	return 0;
3652 
3653  destroy_cache:
3654 	if (base->lcla_pool.dma_addr)
3655 		dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3656 				 SZ_1K * base->num_phy_chans,
3657 				 DMA_TO_DEVICE);
3658 
3659 	if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3660 		free_pages((unsigned long)base->lcla_pool.base,
3661 			   base->lcla_pool.pages);
3662 
3663 	kfree(base->lcla_pool.base_unaligned);
3664 
3665 	if (base->lcpa_regulator) {
3666 		regulator_disable(base->lcpa_regulator);
3667 		regulator_put(base->lcpa_regulator);
3668 	}
3669 	pm_runtime_disable(base->dev);
3670 
3671  report_failure:
3672 	d40_err(dev, "probe failed\n");
3673 	return ret;
3674 }
3675 
3676 static const struct of_device_id d40_match[] = {
3677         { .compatible = "stericsson,dma40", },
3678         {}
3679 };
3680 
3681 static struct platform_driver d40_driver = {
3682 	.driver = {
3683 		.name  = D40_NAME,
3684 		.pm = &dma40_pm_ops,
3685 		.of_match_table = d40_match,
3686 	},
3687 };
3688 
3689 static int __init stedma40_init(void)
3690 {
3691 	return platform_driver_probe(&d40_driver, d40_probe);
3692 }
3693 subsys_initcall(stedma40_init);
3694