xref: /linux/drivers/dma/ste_dma40.c (revision 69050f8d6d075dc01af7a5f2f550a8067510366f)
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_put_autosuspend(d40c->base->dev);
1456 	spin_unlock_irqrestore(&d40c->lock, flags);
1457 	return res;
1458 }
1459 
1460 static int d40_resume(struct dma_chan *chan)
1461 {
1462 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
1463 	int res = 0;
1464 	unsigned long flags;
1465 
1466 	if (d40c->phy_chan == NULL) {
1467 		chan_err(d40c, "Channel is not allocated!\n");
1468 		return -EINVAL;
1469 	}
1470 
1471 	if (!d40c->busy)
1472 		return 0;
1473 
1474 	spin_lock_irqsave(&d40c->lock, flags);
1475 	pm_runtime_get_sync(d40c->base->dev);
1476 
1477 	/* If bytes left to transfer or linked tx resume job */
1478 	if (d40_residue(d40c) || d40_tx_is_linked(d40c))
1479 		res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1480 
1481 	pm_runtime_put_autosuspend(d40c->base->dev);
1482 	spin_unlock_irqrestore(&d40c->lock, flags);
1483 	return res;
1484 }
1485 
1486 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
1487 {
1488 	struct d40_chan *d40c = container_of(tx->chan,
1489 					     struct d40_chan,
1490 					     chan);
1491 	struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
1492 	unsigned long flags;
1493 	dma_cookie_t cookie;
1494 
1495 	spin_lock_irqsave(&d40c->lock, flags);
1496 	cookie = dma_cookie_assign(tx);
1497 	d40_desc_queue(d40c, d40d);
1498 	spin_unlock_irqrestore(&d40c->lock, flags);
1499 
1500 	return cookie;
1501 }
1502 
1503 static int d40_start(struct d40_chan *d40c)
1504 {
1505 	return d40_channel_execute_command(d40c, D40_DMA_RUN);
1506 }
1507 
1508 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
1509 {
1510 	struct d40_desc *d40d;
1511 	int err;
1512 
1513 	/* Start queued jobs, if any */
1514 	d40d = d40_first_queued(d40c);
1515 
1516 	if (d40d != NULL) {
1517 		if (!d40c->busy) {
1518 			d40c->busy = true;
1519 			pm_runtime_get_sync(d40c->base->dev);
1520 		}
1521 
1522 		/* Remove from queue */
1523 		d40_desc_remove(d40d);
1524 
1525 		/* Add to active queue */
1526 		d40_desc_submit(d40c, d40d);
1527 
1528 		/* Initiate DMA job */
1529 		d40_desc_load(d40c, d40d);
1530 
1531 		/* Start dma job */
1532 		err = d40_start(d40c);
1533 
1534 		if (err)
1535 			return NULL;
1536 	}
1537 
1538 	return d40d;
1539 }
1540 
1541 /* called from interrupt context */
1542 static void dma_tc_handle(struct d40_chan *d40c)
1543 {
1544 	struct d40_desc *d40d;
1545 
1546 	/* Get first active entry from list */
1547 	d40d = d40_first_active_get(d40c);
1548 
1549 	if (d40d == NULL)
1550 		return;
1551 
1552 	if (d40d->cyclic) {
1553 		/*
1554 		 * If this was a paritially loaded list, we need to reloaded
1555 		 * it, and only when the list is completed.  We need to check
1556 		 * for done because the interrupt will hit for every link, and
1557 		 * not just the last one.
1558 		 */
1559 		if (d40d->lli_current < d40d->lli_len
1560 		    && !d40_tx_is_linked(d40c)
1561 		    && !d40_residue(d40c)) {
1562 			d40_lcla_free_all(d40c, d40d);
1563 			d40_desc_load(d40c, d40d);
1564 			(void) d40_start(d40c);
1565 
1566 			if (d40d->lli_current == d40d->lli_len)
1567 				d40d->lli_current = 0;
1568 		}
1569 	} else {
1570 		d40_lcla_free_all(d40c, d40d);
1571 
1572 		if (d40d->lli_current < d40d->lli_len) {
1573 			d40_desc_load(d40c, d40d);
1574 			/* Start dma job */
1575 			(void) d40_start(d40c);
1576 			return;
1577 		}
1578 
1579 		if (d40_queue_start(d40c) == NULL) {
1580 			d40c->busy = false;
1581 
1582 			pm_runtime_put_autosuspend(d40c->base->dev);
1583 		}
1584 
1585 		d40_desc_remove(d40d);
1586 		d40_desc_done(d40c, d40d);
1587 	}
1588 
1589 	d40c->pending_tx++;
1590 	tasklet_schedule(&d40c->tasklet);
1591 
1592 }
1593 
1594 static void dma_tasklet(struct tasklet_struct *t)
1595 {
1596 	struct d40_chan *d40c = from_tasklet(d40c, t, tasklet);
1597 	struct d40_desc *d40d;
1598 	unsigned long flags;
1599 	bool callback_active;
1600 	struct dmaengine_desc_callback cb;
1601 
1602 	spin_lock_irqsave(&d40c->lock, flags);
1603 
1604 	/* Get first entry from the done list */
1605 	d40d = d40_first_done(d40c);
1606 	if (d40d == NULL) {
1607 		/* Check if we have reached here for cyclic job */
1608 		d40d = d40_first_active_get(d40c);
1609 		if (d40d == NULL || !d40d->cyclic)
1610 			goto check_pending_tx;
1611 	}
1612 
1613 	if (!d40d->cyclic)
1614 		dma_cookie_complete(&d40d->txd);
1615 
1616 	/*
1617 	 * If terminating a channel pending_tx is set to zero.
1618 	 * This prevents any finished active jobs to return to the client.
1619 	 */
1620 	if (d40c->pending_tx == 0) {
1621 		spin_unlock_irqrestore(&d40c->lock, flags);
1622 		return;
1623 	}
1624 
1625 	/* Callback to client */
1626 	callback_active = !!(d40d->txd.flags & DMA_PREP_INTERRUPT);
1627 	dmaengine_desc_get_callback(&d40d->txd, &cb);
1628 
1629 	if (!d40d->cyclic) {
1630 		if (async_tx_test_ack(&d40d->txd)) {
1631 			d40_desc_remove(d40d);
1632 			d40_desc_free(d40c, d40d);
1633 		} else if (!d40d->is_in_client_list) {
1634 			d40_desc_remove(d40d);
1635 			d40_lcla_free_all(d40c, d40d);
1636 			list_add_tail(&d40d->node, &d40c->client);
1637 			d40d->is_in_client_list = true;
1638 		}
1639 	}
1640 
1641 	d40c->pending_tx--;
1642 
1643 	if (d40c->pending_tx)
1644 		tasklet_schedule(&d40c->tasklet);
1645 
1646 	spin_unlock_irqrestore(&d40c->lock, flags);
1647 
1648 	if (callback_active)
1649 		dmaengine_desc_callback_invoke(&cb, NULL);
1650 
1651 	return;
1652  check_pending_tx:
1653 	/* Rescue maneuver if receiving double interrupts */
1654 	if (d40c->pending_tx > 0)
1655 		d40c->pending_tx--;
1656 	spin_unlock_irqrestore(&d40c->lock, flags);
1657 }
1658 
1659 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1660 {
1661 	int i;
1662 	u32 idx;
1663 	u32 row;
1664 	long chan = -1;
1665 	struct d40_chan *d40c;
1666 	struct d40_base *base = data;
1667 	u32 *regs = base->regs_interrupt;
1668 	struct d40_interrupt_lookup *il = base->gen_dmac.il;
1669 	u32 il_size = base->gen_dmac.il_size;
1670 
1671 	spin_lock(&base->interrupt_lock);
1672 
1673 	/* Read interrupt status of both logical and physical channels */
1674 	for (i = 0; i < il_size; i++)
1675 		regs[i] = readl(base->virtbase + il[i].src);
1676 
1677 	for (;;) {
1678 
1679 		chan = find_next_bit((unsigned long *)regs,
1680 				     BITS_PER_LONG * il_size, chan + 1);
1681 
1682 		/* No more set bits found? */
1683 		if (chan == BITS_PER_LONG * il_size)
1684 			break;
1685 
1686 		row = chan / BITS_PER_LONG;
1687 		idx = chan & (BITS_PER_LONG - 1);
1688 
1689 		if (il[row].offset == D40_PHY_CHAN)
1690 			d40c = base->lookup_phy_chans[idx];
1691 		else
1692 			d40c = base->lookup_log_chans[il[row].offset + idx];
1693 
1694 		if (!d40c) {
1695 			/*
1696 			 * No error because this can happen if something else
1697 			 * in the system is using the channel.
1698 			 */
1699 			continue;
1700 		}
1701 
1702 		/* ACK interrupt */
1703 		writel(BIT(idx), base->virtbase + il[row].clr);
1704 
1705 		spin_lock(&d40c->lock);
1706 
1707 		if (!il[row].is_error)
1708 			dma_tc_handle(d40c);
1709 		else
1710 			d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1711 				chan, il[row].offset, idx);
1712 
1713 		spin_unlock(&d40c->lock);
1714 	}
1715 
1716 	spin_unlock(&base->interrupt_lock);
1717 
1718 	return IRQ_HANDLED;
1719 }
1720 
1721 static int d40_validate_conf(struct d40_chan *d40c,
1722 			     struct stedma40_chan_cfg *conf)
1723 {
1724 	int res = 0;
1725 	bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1726 
1727 	if (!conf->dir) {
1728 		chan_err(d40c, "Invalid direction.\n");
1729 		res = -EINVAL;
1730 	}
1731 
1732 	if ((is_log && conf->dev_type > d40c->base->num_log_chans)  ||
1733 	    (!is_log && conf->dev_type > d40c->base->num_phy_chans) ||
1734 	    (conf->dev_type < 0)) {
1735 		chan_err(d40c, "Invalid device type (%d)\n", conf->dev_type);
1736 		res = -EINVAL;
1737 	}
1738 
1739 	if (conf->dir == DMA_DEV_TO_DEV) {
1740 		/*
1741 		 * DMAC HW supports it. Will be added to this driver,
1742 		 * in case any dma client requires it.
1743 		 */
1744 		chan_err(d40c, "periph to periph not supported\n");
1745 		res = -EINVAL;
1746 	}
1747 
1748 	if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1749 	    conf->src_info.data_width !=
1750 	    d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1751 	    conf->dst_info.data_width) {
1752 		/*
1753 		 * The DMAC hardware only supports
1754 		 * src (burst x width) == dst (burst x width)
1755 		 */
1756 
1757 		chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1758 		res = -EINVAL;
1759 	}
1760 
1761 	return res;
1762 }
1763 
1764 static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1765 			       bool is_src, int log_event_line, bool is_log,
1766 			       bool *first_user)
1767 {
1768 	unsigned long flags;
1769 	spin_lock_irqsave(&phy->lock, flags);
1770 
1771 	*first_user = ((phy->allocated_src | phy->allocated_dst)
1772 			== D40_ALLOC_FREE);
1773 
1774 	if (!is_log) {
1775 		/* Physical interrupts are masked per physical full channel */
1776 		if (phy->allocated_src == D40_ALLOC_FREE &&
1777 		    phy->allocated_dst == D40_ALLOC_FREE) {
1778 			phy->allocated_dst = D40_ALLOC_PHY;
1779 			phy->allocated_src = D40_ALLOC_PHY;
1780 			goto found_unlock;
1781 		} else
1782 			goto not_found_unlock;
1783 	}
1784 
1785 	/* Logical channel */
1786 	if (is_src) {
1787 		if (phy->allocated_src == D40_ALLOC_PHY)
1788 			goto not_found_unlock;
1789 
1790 		if (phy->allocated_src == D40_ALLOC_FREE)
1791 			phy->allocated_src = D40_ALLOC_LOG_FREE;
1792 
1793 		if (!(phy->allocated_src & BIT(log_event_line))) {
1794 			phy->allocated_src |= BIT(log_event_line);
1795 			goto found_unlock;
1796 		} else
1797 			goto not_found_unlock;
1798 	} else {
1799 		if (phy->allocated_dst == D40_ALLOC_PHY)
1800 			goto not_found_unlock;
1801 
1802 		if (phy->allocated_dst == D40_ALLOC_FREE)
1803 			phy->allocated_dst = D40_ALLOC_LOG_FREE;
1804 
1805 		if (!(phy->allocated_dst & BIT(log_event_line))) {
1806 			phy->allocated_dst |= BIT(log_event_line);
1807 			goto found_unlock;
1808 		}
1809 	}
1810  not_found_unlock:
1811 	spin_unlock_irqrestore(&phy->lock, flags);
1812 	return false;
1813  found_unlock:
1814 	spin_unlock_irqrestore(&phy->lock, flags);
1815 	return true;
1816 }
1817 
1818 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1819 			       int log_event_line)
1820 {
1821 	unsigned long flags;
1822 	bool is_free = false;
1823 
1824 	spin_lock_irqsave(&phy->lock, flags);
1825 	if (!log_event_line) {
1826 		phy->allocated_dst = D40_ALLOC_FREE;
1827 		phy->allocated_src = D40_ALLOC_FREE;
1828 		is_free = true;
1829 		goto unlock;
1830 	}
1831 
1832 	/* Logical channel */
1833 	if (is_src) {
1834 		phy->allocated_src &= ~BIT(log_event_line);
1835 		if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1836 			phy->allocated_src = D40_ALLOC_FREE;
1837 	} else {
1838 		phy->allocated_dst &= ~BIT(log_event_line);
1839 		if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1840 			phy->allocated_dst = D40_ALLOC_FREE;
1841 	}
1842 
1843 	is_free = ((phy->allocated_src | phy->allocated_dst) ==
1844 		   D40_ALLOC_FREE);
1845  unlock:
1846 	spin_unlock_irqrestore(&phy->lock, flags);
1847 
1848 	return is_free;
1849 }
1850 
1851 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1852 {
1853 	int dev_type = d40c->dma_cfg.dev_type;
1854 	int event_group;
1855 	int event_line;
1856 	struct d40_phy_res *phys;
1857 	int i;
1858 	int j;
1859 	int log_num;
1860 	int num_phy_chans;
1861 	bool is_src;
1862 	bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1863 
1864 	phys = d40c->base->phy_res;
1865 	num_phy_chans = d40c->base->num_phy_chans;
1866 
1867 	if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
1868 		log_num = 2 * dev_type;
1869 		is_src = true;
1870 	} else if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
1871 		   d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1872 		/* dst event lines are used for logical memcpy */
1873 		log_num = 2 * dev_type + 1;
1874 		is_src = false;
1875 	} else
1876 		return -EINVAL;
1877 
1878 	event_group = D40_TYPE_TO_GROUP(dev_type);
1879 	event_line = D40_TYPE_TO_EVENT(dev_type);
1880 
1881 	if (!is_log) {
1882 		if (d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1883 			/* Find physical half channel */
1884 			if (d40c->dma_cfg.use_fixed_channel) {
1885 				i = d40c->dma_cfg.phy_channel;
1886 				if (d40_alloc_mask_set(&phys[i], is_src,
1887 						       0, is_log,
1888 						       first_phy_user))
1889 					goto found_phy;
1890 			} else {
1891 				for (i = 0; i < num_phy_chans; i++) {
1892 					if (d40_alloc_mask_set(&phys[i], is_src,
1893 						       0, is_log,
1894 						       first_phy_user))
1895 						goto found_phy;
1896 				}
1897 			}
1898 		} else
1899 			for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1900 				int phy_num = j  + event_group * 2;
1901 				for (i = phy_num; i < phy_num + 2; i++) {
1902 					if (d40_alloc_mask_set(&phys[i],
1903 							       is_src,
1904 							       0,
1905 							       is_log,
1906 							       first_phy_user))
1907 						goto found_phy;
1908 				}
1909 			}
1910 		return -EINVAL;
1911 found_phy:
1912 		d40c->phy_chan = &phys[i];
1913 		d40c->log_num = D40_PHY_CHAN;
1914 		goto out;
1915 	}
1916 	if (dev_type == -1)
1917 		return -EINVAL;
1918 
1919 	/* Find logical channel */
1920 	for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1921 		int phy_num = j + event_group * 2;
1922 
1923 		if (d40c->dma_cfg.use_fixed_channel) {
1924 			i = d40c->dma_cfg.phy_channel;
1925 
1926 			if ((i != phy_num) && (i != phy_num + 1)) {
1927 				dev_err(chan2dev(d40c),
1928 					"invalid fixed phy channel %d\n", i);
1929 				return -EINVAL;
1930 			}
1931 
1932 			if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1933 					       is_log, first_phy_user))
1934 				goto found_log;
1935 
1936 			dev_err(chan2dev(d40c),
1937 				"could not allocate fixed phy channel %d\n", i);
1938 			return -EINVAL;
1939 		}
1940 
1941 		/*
1942 		 * Spread logical channels across all available physical rather
1943 		 * than pack every logical channel at the first available phy
1944 		 * channels.
1945 		 */
1946 		if (is_src) {
1947 			for (i = phy_num; i < phy_num + 2; i++) {
1948 				if (d40_alloc_mask_set(&phys[i], is_src,
1949 						       event_line, is_log,
1950 						       first_phy_user))
1951 					goto found_log;
1952 			}
1953 		} else {
1954 			for (i = phy_num + 1; i >= phy_num; i--) {
1955 				if (d40_alloc_mask_set(&phys[i], is_src,
1956 						       event_line, is_log,
1957 						       first_phy_user))
1958 					goto found_log;
1959 			}
1960 		}
1961 	}
1962 	return -EINVAL;
1963 
1964 found_log:
1965 	d40c->phy_chan = &phys[i];
1966 	d40c->log_num = log_num;
1967 out:
1968 
1969 	if (is_log)
1970 		d40c->base->lookup_log_chans[d40c->log_num] = d40c;
1971 	else
1972 		d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
1973 
1974 	return 0;
1975 
1976 }
1977 
1978 static int d40_config_memcpy(struct d40_chan *d40c)
1979 {
1980 	dma_cap_mask_t cap = d40c->chan.device->cap_mask;
1981 
1982 	if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
1983 		d40c->dma_cfg = dma40_memcpy_conf_log;
1984 		d40c->dma_cfg.dev_type = dma40_memcpy_channels[d40c->chan.chan_id];
1985 
1986 		d40_log_cfg(&d40c->dma_cfg,
1987 			    &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
1988 
1989 	} else if (dma_has_cap(DMA_MEMCPY, cap) &&
1990 		   dma_has_cap(DMA_SLAVE, cap)) {
1991 		d40c->dma_cfg = dma40_memcpy_conf_phy;
1992 
1993 		/* Generate interrupt at end of transfer or relink. */
1994 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_TIM_POS);
1995 
1996 		/* Generate interrupt on error. */
1997 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
1998 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
1999 
2000 	} else {
2001 		chan_err(d40c, "No memcpy\n");
2002 		return -EINVAL;
2003 	}
2004 
2005 	return 0;
2006 }
2007 
2008 static int d40_free_dma(struct d40_chan *d40c)
2009 {
2010 
2011 	int res = 0;
2012 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2013 	struct d40_phy_res *phy = d40c->phy_chan;
2014 	bool is_src;
2015 
2016 	/* Terminate all queued and active transfers */
2017 	d40_term_all(d40c);
2018 
2019 	if (phy == NULL) {
2020 		chan_err(d40c, "phy == null\n");
2021 		return -EINVAL;
2022 	}
2023 
2024 	if (phy->allocated_src == D40_ALLOC_FREE &&
2025 	    phy->allocated_dst == D40_ALLOC_FREE) {
2026 		chan_err(d40c, "channel already free\n");
2027 		return -EINVAL;
2028 	}
2029 
2030 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2031 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM)
2032 		is_src = false;
2033 	else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2034 		is_src = true;
2035 	else {
2036 		chan_err(d40c, "Unknown direction\n");
2037 		return -EINVAL;
2038 	}
2039 
2040 	pm_runtime_get_sync(d40c->base->dev);
2041 	res = d40_channel_execute_command(d40c, D40_DMA_STOP);
2042 	if (res) {
2043 		chan_err(d40c, "stop failed\n");
2044 		goto mark_last_busy;
2045 	}
2046 
2047 	d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
2048 
2049 	if (chan_is_logical(d40c))
2050 		d40c->base->lookup_log_chans[d40c->log_num] = NULL;
2051 	else
2052 		d40c->base->lookup_phy_chans[phy->num] = NULL;
2053 
2054 	if (d40c->busy)
2055 		pm_runtime_put_autosuspend(d40c->base->dev);
2056 
2057 	d40c->busy = false;
2058 	d40c->phy_chan = NULL;
2059 	d40c->configured = false;
2060  mark_last_busy:
2061 	pm_runtime_put_autosuspend(d40c->base->dev);
2062 	return res;
2063 }
2064 
2065 static bool d40_is_paused(struct d40_chan *d40c)
2066 {
2067 	void __iomem *chanbase = chan_base(d40c);
2068 	bool is_paused = false;
2069 	unsigned long flags;
2070 	void __iomem *active_reg;
2071 	u32 status;
2072 	u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2073 
2074 	spin_lock_irqsave(&d40c->lock, flags);
2075 
2076 	if (chan_is_physical(d40c)) {
2077 		if (d40c->phy_chan->num % 2 == 0)
2078 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
2079 		else
2080 			active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
2081 
2082 		status = (readl(active_reg) &
2083 			  D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
2084 			D40_CHAN_POS(d40c->phy_chan->num);
2085 		if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
2086 			is_paused = true;
2087 		goto unlock;
2088 	}
2089 
2090 	if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2091 	    d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
2092 		status = readl(chanbase + D40_CHAN_REG_SDLNK);
2093 	} else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
2094 		status = readl(chanbase + D40_CHAN_REG_SSLNK);
2095 	} else {
2096 		chan_err(d40c, "Unknown direction\n");
2097 		goto unlock;
2098 	}
2099 
2100 	status = (status & D40_EVENTLINE_MASK(event)) >>
2101 		D40_EVENTLINE_POS(event);
2102 
2103 	if (status != D40_DMA_RUN)
2104 		is_paused = true;
2105  unlock:
2106 	spin_unlock_irqrestore(&d40c->lock, flags);
2107 	return is_paused;
2108 
2109 }
2110 
2111 static u32 stedma40_residue(struct dma_chan *chan)
2112 {
2113 	struct d40_chan *d40c =
2114 		container_of(chan, struct d40_chan, chan);
2115 	u32 bytes_left;
2116 	unsigned long flags;
2117 
2118 	spin_lock_irqsave(&d40c->lock, flags);
2119 	bytes_left = d40_residue(d40c);
2120 	spin_unlock_irqrestore(&d40c->lock, flags);
2121 
2122 	return bytes_left;
2123 }
2124 
2125 static int
2126 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
2127 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2128 		unsigned int sg_len, dma_addr_t src_dev_addr,
2129 		dma_addr_t dst_dev_addr)
2130 {
2131 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2132 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2133 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2134 	int ret;
2135 
2136 	ret = d40_log_sg_to_lli(sg_src, sg_len,
2137 				src_dev_addr,
2138 				desc->lli_log.src,
2139 				chan->log_def.lcsp1,
2140 				src_info->data_width,
2141 				dst_info->data_width);
2142 
2143 	ret = d40_log_sg_to_lli(sg_dst, sg_len,
2144 				dst_dev_addr,
2145 				desc->lli_log.dst,
2146 				chan->log_def.lcsp3,
2147 				dst_info->data_width,
2148 				src_info->data_width);
2149 
2150 	return ret < 0 ? ret : 0;
2151 }
2152 
2153 static int
2154 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2155 		struct scatterlist *sg_src, struct scatterlist *sg_dst,
2156 		unsigned int sg_len, dma_addr_t src_dev_addr,
2157 		dma_addr_t dst_dev_addr)
2158 {
2159 	struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2160 	struct stedma40_half_channel_info *src_info = &cfg->src_info;
2161 	struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2162 	unsigned long flags = 0;
2163 	int ret;
2164 
2165 	if (desc->cyclic)
2166 		flags |= LLI_CYCLIC | LLI_TERM_INT;
2167 
2168 	ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2169 				desc->lli_phy.src,
2170 				virt_to_phys(desc->lli_phy.src),
2171 				chan->src_def_cfg,
2172 				src_info, dst_info, flags);
2173 
2174 	ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2175 				desc->lli_phy.dst,
2176 				virt_to_phys(desc->lli_phy.dst),
2177 				chan->dst_def_cfg,
2178 				dst_info, src_info, flags);
2179 
2180 	dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2181 				   desc->lli_pool.size, DMA_TO_DEVICE);
2182 
2183 	return ret < 0 ? ret : 0;
2184 }
2185 
2186 static struct d40_desc *
2187 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2188 	      unsigned int sg_len, unsigned long dma_flags)
2189 {
2190 	struct stedma40_chan_cfg *cfg;
2191 	struct d40_desc *desc;
2192 	int ret;
2193 
2194 	desc = d40_desc_get(chan);
2195 	if (!desc)
2196 		return NULL;
2197 
2198 	cfg = &chan->dma_cfg;
2199 	desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2200 					cfg->dst_info.data_width);
2201 	if (desc->lli_len < 0) {
2202 		chan_err(chan, "Unaligned size\n");
2203 		goto free_desc;
2204 	}
2205 
2206 	ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2207 	if (ret < 0) {
2208 		chan_err(chan, "Could not allocate lli\n");
2209 		goto free_desc;
2210 	}
2211 
2212 	desc->lli_current = 0;
2213 	desc->txd.flags = dma_flags;
2214 	desc->txd.tx_submit = d40_tx_submit;
2215 
2216 	dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2217 
2218 	return desc;
2219  free_desc:
2220 	d40_desc_free(chan, desc);
2221 	return NULL;
2222 }
2223 
2224 static struct dma_async_tx_descriptor *
2225 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2226 	    struct scatterlist *sg_dst, unsigned int sg_len,
2227 	    enum dma_transfer_direction direction, unsigned long dma_flags)
2228 {
2229 	struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2230 	dma_addr_t src_dev_addr;
2231 	dma_addr_t dst_dev_addr;
2232 	struct d40_desc *desc;
2233 	unsigned long flags;
2234 	int ret;
2235 
2236 	if (!chan->phy_chan) {
2237 		chan_err(chan, "Cannot prepare unallocated channel\n");
2238 		return NULL;
2239 	}
2240 
2241 	d40_set_runtime_config_write(dchan, &chan->slave_config, direction);
2242 
2243 	spin_lock_irqsave(&chan->lock, flags);
2244 
2245 	desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2246 	if (desc == NULL)
2247 		goto unlock;
2248 
2249 	if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2250 		desc->cyclic = true;
2251 
2252 	src_dev_addr = 0;
2253 	dst_dev_addr = 0;
2254 	if (direction == DMA_DEV_TO_MEM)
2255 		src_dev_addr = chan->runtime_addr;
2256 	else if (direction == DMA_MEM_TO_DEV)
2257 		dst_dev_addr = chan->runtime_addr;
2258 
2259 	if (chan_is_logical(chan))
2260 		ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2261 				      sg_len, src_dev_addr, dst_dev_addr);
2262 	else
2263 		ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2264 				      sg_len, src_dev_addr, dst_dev_addr);
2265 
2266 	if (ret) {
2267 		chan_err(chan, "Failed to prepare %s sg job: %d\n",
2268 			 chan_is_logical(chan) ? "log" : "phy", ret);
2269 		goto free_desc;
2270 	}
2271 
2272 	/*
2273 	 * add descriptor to the prepare queue in order to be able
2274 	 * to free them later in terminate_all
2275 	 */
2276 	list_add_tail(&desc->node, &chan->prepare_queue);
2277 
2278 	spin_unlock_irqrestore(&chan->lock, flags);
2279 
2280 	return &desc->txd;
2281  free_desc:
2282 	d40_desc_free(chan, desc);
2283  unlock:
2284 	spin_unlock_irqrestore(&chan->lock, flags);
2285 	return NULL;
2286 }
2287 
2288 static bool stedma40_filter(struct dma_chan *chan, void *data)
2289 {
2290 	struct stedma40_chan_cfg *info = data;
2291 	struct d40_chan *d40c =
2292 		container_of(chan, struct d40_chan, chan);
2293 	int err;
2294 
2295 	if (data) {
2296 		err = d40_validate_conf(d40c, info);
2297 		if (!err)
2298 			d40c->dma_cfg = *info;
2299 	} else
2300 		err = d40_config_memcpy(d40c);
2301 
2302 	if (!err)
2303 		d40c->configured = true;
2304 
2305 	return err == 0;
2306 }
2307 
2308 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2309 {
2310 	bool realtime = d40c->dma_cfg.realtime;
2311 	bool highprio = d40c->dma_cfg.high_priority;
2312 	u32 rtreg;
2313 	u32 event = D40_TYPE_TO_EVENT(dev_type);
2314 	u32 group = D40_TYPE_TO_GROUP(dev_type);
2315 	u32 bit = BIT(event);
2316 	u32 prioreg;
2317 	struct d40_gen_dmac *dmac = &d40c->base->gen_dmac;
2318 
2319 	rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear;
2320 	/*
2321 	 * Due to a hardware bug, in some cases a logical channel triggered by
2322 	 * a high priority destination event line can generate extra packet
2323 	 * transactions.
2324 	 *
2325 	 * The workaround is to not set the high priority level for the
2326 	 * destination event lines that trigger logical channels.
2327 	 */
2328 	if (!src && chan_is_logical(d40c))
2329 		highprio = false;
2330 
2331 	prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear;
2332 
2333 	/* Destination event lines are stored in the upper halfword */
2334 	if (!src)
2335 		bit <<= 16;
2336 
2337 	writel(bit, d40c->base->virtbase + prioreg + group * 4);
2338 	writel(bit, d40c->base->virtbase + rtreg + group * 4);
2339 }
2340 
2341 static void d40_set_prio_realtime(struct d40_chan *d40c)
2342 {
2343 	if (d40c->base->rev < 3)
2344 		return;
2345 
2346 	if ((d40c->dma_cfg.dir ==  DMA_DEV_TO_MEM) ||
2347 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2348 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, true);
2349 
2350 	if ((d40c->dma_cfg.dir ==  DMA_MEM_TO_DEV) ||
2351 	    (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2352 		__d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, false);
2353 }
2354 
2355 #define D40_DT_FLAGS_MODE(flags)       ((flags >> 0) & 0x1)
2356 #define D40_DT_FLAGS_DIR(flags)        ((flags >> 1) & 0x1)
2357 #define D40_DT_FLAGS_BIG_ENDIAN(flags) ((flags >> 2) & 0x1)
2358 #define D40_DT_FLAGS_FIXED_CHAN(flags) ((flags >> 3) & 0x1)
2359 #define D40_DT_FLAGS_HIGH_PRIO(flags)  ((flags >> 4) & 0x1)
2360 
2361 static struct dma_chan *d40_xlate(struct of_phandle_args *dma_spec,
2362 				  struct of_dma *ofdma)
2363 {
2364 	struct stedma40_chan_cfg cfg;
2365 	dma_cap_mask_t cap;
2366 	u32 flags;
2367 
2368 	memset(&cfg, 0, sizeof(struct stedma40_chan_cfg));
2369 
2370 	dma_cap_zero(cap);
2371 	dma_cap_set(DMA_SLAVE, cap);
2372 
2373 	cfg.dev_type = dma_spec->args[0];
2374 	flags = dma_spec->args[2];
2375 
2376 	switch (D40_DT_FLAGS_MODE(flags)) {
2377 	case 0: cfg.mode = STEDMA40_MODE_LOGICAL; break;
2378 	case 1: cfg.mode = STEDMA40_MODE_PHYSICAL; break;
2379 	}
2380 
2381 	switch (D40_DT_FLAGS_DIR(flags)) {
2382 	case 0:
2383 		cfg.dir = DMA_MEM_TO_DEV;
2384 		cfg.dst_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2385 		break;
2386 	case 1:
2387 		cfg.dir = DMA_DEV_TO_MEM;
2388 		cfg.src_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2389 		break;
2390 	}
2391 
2392 	if (D40_DT_FLAGS_FIXED_CHAN(flags)) {
2393 		cfg.phy_channel = dma_spec->args[1];
2394 		cfg.use_fixed_channel = true;
2395 	}
2396 
2397 	if (D40_DT_FLAGS_HIGH_PRIO(flags))
2398 		cfg.high_priority = true;
2399 
2400 	return dma_request_channel(cap, stedma40_filter, &cfg);
2401 }
2402 
2403 /* DMA ENGINE functions */
2404 static int d40_alloc_chan_resources(struct dma_chan *chan)
2405 {
2406 	int err;
2407 	unsigned long flags;
2408 	struct d40_chan *d40c =
2409 		container_of(chan, struct d40_chan, chan);
2410 	bool is_free_phy;
2411 	spin_lock_irqsave(&d40c->lock, flags);
2412 
2413 	dma_cookie_init(chan);
2414 
2415 	/* If no dma configuration is set use default configuration (memcpy) */
2416 	if (!d40c->configured) {
2417 		err = d40_config_memcpy(d40c);
2418 		if (err) {
2419 			chan_err(d40c, "Failed to configure memcpy channel\n");
2420 			goto mark_last_busy;
2421 		}
2422 	}
2423 
2424 	err = d40_allocate_channel(d40c, &is_free_phy);
2425 	if (err) {
2426 		chan_err(d40c, "Failed to allocate channel\n");
2427 		d40c->configured = false;
2428 		goto mark_last_busy;
2429 	}
2430 
2431 	pm_runtime_get_sync(d40c->base->dev);
2432 
2433 	d40_set_prio_realtime(d40c);
2434 
2435 	if (chan_is_logical(d40c)) {
2436 		if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2437 			d40c->lcpa = d40c->base->lcpa_base +
2438 				d40c->dma_cfg.dev_type * D40_LCPA_CHAN_SIZE;
2439 		else
2440 			d40c->lcpa = d40c->base->lcpa_base +
2441 				d40c->dma_cfg.dev_type *
2442 				D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2443 
2444 		/* Unmask the Global Interrupt Mask. */
2445 		d40c->src_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2446 		d40c->dst_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2447 	}
2448 
2449 	dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2450 		 chan_is_logical(d40c) ? "logical" : "physical",
2451 		 d40c->phy_chan->num,
2452 		 d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2453 
2454 
2455 	/*
2456 	 * Only write channel configuration to the DMA if the physical
2457 	 * resource is free. In case of multiple logical channels
2458 	 * on the same physical resource, only the first write is necessary.
2459 	 */
2460 	if (is_free_phy)
2461 		d40_config_write(d40c);
2462  mark_last_busy:
2463 	pm_runtime_put_autosuspend(d40c->base->dev);
2464 	spin_unlock_irqrestore(&d40c->lock, flags);
2465 	return err;
2466 }
2467 
2468 static void d40_free_chan_resources(struct dma_chan *chan)
2469 {
2470 	struct d40_chan *d40c =
2471 		container_of(chan, struct d40_chan, chan);
2472 	int err;
2473 	unsigned long flags;
2474 
2475 	if (d40c->phy_chan == NULL) {
2476 		chan_err(d40c, "Cannot free unallocated channel\n");
2477 		return;
2478 	}
2479 
2480 	spin_lock_irqsave(&d40c->lock, flags);
2481 
2482 	err = d40_free_dma(d40c);
2483 
2484 	if (err)
2485 		chan_err(d40c, "Failed to free channel\n");
2486 	spin_unlock_irqrestore(&d40c->lock, flags);
2487 }
2488 
2489 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2490 						       dma_addr_t dst,
2491 						       dma_addr_t src,
2492 						       size_t size,
2493 						       unsigned long dma_flags)
2494 {
2495 	struct scatterlist dst_sg;
2496 	struct scatterlist src_sg;
2497 
2498 	sg_init_table(&dst_sg, 1);
2499 	sg_init_table(&src_sg, 1);
2500 
2501 	sg_dma_address(&dst_sg) = dst;
2502 	sg_dma_address(&src_sg) = src;
2503 
2504 	sg_dma_len(&dst_sg) = size;
2505 	sg_dma_len(&src_sg) = size;
2506 
2507 	return d40_prep_sg(chan, &src_sg, &dst_sg, 1,
2508 			   DMA_MEM_TO_MEM, dma_flags);
2509 }
2510 
2511 static struct dma_async_tx_descriptor *
2512 d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2513 		  unsigned int sg_len, enum dma_transfer_direction direction,
2514 		  unsigned long dma_flags, void *context)
2515 {
2516 	if (!is_slave_direction(direction))
2517 		return NULL;
2518 
2519 	return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2520 }
2521 
2522 static struct dma_async_tx_descriptor *
2523 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2524 		     size_t buf_len, size_t period_len,
2525 		     enum dma_transfer_direction direction, unsigned long flags)
2526 {
2527 	unsigned int periods = buf_len / period_len;
2528 	struct dma_async_tx_descriptor *txd;
2529 	struct scatterlist *sg;
2530 	int i;
2531 
2532 	sg = kzalloc_objs(struct scatterlist, periods + 1, GFP_NOWAIT);
2533 	if (!sg)
2534 		return NULL;
2535 
2536 	for (i = 0; i < periods; i++) {
2537 		sg_dma_address(&sg[i]) = dma_addr;
2538 		sg_dma_len(&sg[i]) = period_len;
2539 		dma_addr += period_len;
2540 	}
2541 
2542 	sg_chain(sg, periods + 1, sg);
2543 
2544 	txd = d40_prep_sg(chan, sg, sg, periods, direction,
2545 			  DMA_PREP_INTERRUPT);
2546 
2547 	kfree(sg);
2548 
2549 	return txd;
2550 }
2551 
2552 static enum dma_status d40_tx_status(struct dma_chan *chan,
2553 				     dma_cookie_t cookie,
2554 				     struct dma_tx_state *txstate)
2555 {
2556 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2557 	enum dma_status ret;
2558 
2559 	if (d40c->phy_chan == NULL) {
2560 		chan_err(d40c, "Cannot read status of unallocated channel\n");
2561 		return -EINVAL;
2562 	}
2563 
2564 	ret = dma_cookie_status(chan, cookie, txstate);
2565 	if (ret != DMA_COMPLETE && txstate)
2566 		dma_set_residue(txstate, stedma40_residue(chan));
2567 
2568 	if (d40_is_paused(d40c))
2569 		ret = DMA_PAUSED;
2570 
2571 	return ret;
2572 }
2573 
2574 static void d40_issue_pending(struct dma_chan *chan)
2575 {
2576 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2577 	unsigned long flags;
2578 
2579 	if (d40c->phy_chan == NULL) {
2580 		chan_err(d40c, "Channel is not allocated!\n");
2581 		return;
2582 	}
2583 
2584 	spin_lock_irqsave(&d40c->lock, flags);
2585 
2586 	list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2587 
2588 	/* Busy means that queued jobs are already being processed */
2589 	if (!d40c->busy)
2590 		(void) d40_queue_start(d40c);
2591 
2592 	spin_unlock_irqrestore(&d40c->lock, flags);
2593 }
2594 
2595 static int d40_terminate_all(struct dma_chan *chan)
2596 {
2597 	unsigned long flags;
2598 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2599 	int ret;
2600 
2601 	if (d40c->phy_chan == NULL) {
2602 		chan_err(d40c, "Channel is not allocated!\n");
2603 		return -EINVAL;
2604 	}
2605 
2606 	spin_lock_irqsave(&d40c->lock, flags);
2607 
2608 	pm_runtime_get_sync(d40c->base->dev);
2609 	ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2610 	if (ret)
2611 		chan_err(d40c, "Failed to stop channel\n");
2612 
2613 	d40_term_all(d40c);
2614 	pm_runtime_put_autosuspend(d40c->base->dev);
2615 	if (d40c->busy)
2616 		pm_runtime_put_autosuspend(d40c->base->dev);
2617 	d40c->busy = false;
2618 
2619 	spin_unlock_irqrestore(&d40c->lock, flags);
2620 	return 0;
2621 }
2622 
2623 static int
2624 dma40_config_to_halfchannel(struct d40_chan *d40c,
2625 			    struct stedma40_half_channel_info *info,
2626 			    u32 maxburst)
2627 {
2628 	int psize;
2629 
2630 	if (chan_is_logical(d40c)) {
2631 		if (maxburst >= 16)
2632 			psize = STEDMA40_PSIZE_LOG_16;
2633 		else if (maxburst >= 8)
2634 			psize = STEDMA40_PSIZE_LOG_8;
2635 		else if (maxburst >= 4)
2636 			psize = STEDMA40_PSIZE_LOG_4;
2637 		else
2638 			psize = STEDMA40_PSIZE_LOG_1;
2639 	} else {
2640 		if (maxburst >= 16)
2641 			psize = STEDMA40_PSIZE_PHY_16;
2642 		else if (maxburst >= 8)
2643 			psize = STEDMA40_PSIZE_PHY_8;
2644 		else if (maxburst >= 4)
2645 			psize = STEDMA40_PSIZE_PHY_4;
2646 		else
2647 			psize = STEDMA40_PSIZE_PHY_1;
2648 	}
2649 
2650 	info->psize = psize;
2651 	info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2652 
2653 	return 0;
2654 }
2655 
2656 static int d40_set_runtime_config(struct dma_chan *chan,
2657 				  struct dma_slave_config *config)
2658 {
2659 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2660 
2661 	memcpy(&d40c->slave_config, config, sizeof(*config));
2662 
2663 	return 0;
2664 }
2665 
2666 /* Runtime reconfiguration extension */
2667 static int d40_set_runtime_config_write(struct dma_chan *chan,
2668 				  struct dma_slave_config *config,
2669 				  enum dma_transfer_direction direction)
2670 {
2671 	struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2672 	struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2673 	enum dma_slave_buswidth src_addr_width, dst_addr_width;
2674 	dma_addr_t config_addr;
2675 	u32 src_maxburst, dst_maxburst;
2676 	int ret;
2677 
2678 	if (d40c->phy_chan == NULL) {
2679 		chan_err(d40c, "Channel is not allocated!\n");
2680 		return -EINVAL;
2681 	}
2682 
2683 	src_addr_width = config->src_addr_width;
2684 	src_maxburst = config->src_maxburst;
2685 	dst_addr_width = config->dst_addr_width;
2686 	dst_maxburst = config->dst_maxburst;
2687 
2688 	if (direction == DMA_DEV_TO_MEM) {
2689 		config_addr = config->src_addr;
2690 
2691 		if (cfg->dir != DMA_DEV_TO_MEM)
2692 			dev_dbg(d40c->base->dev,
2693 				"channel was not configured for peripheral "
2694 				"to memory transfer (%d) overriding\n",
2695 				cfg->dir);
2696 		cfg->dir = DMA_DEV_TO_MEM;
2697 
2698 		/* Configure the memory side */
2699 		if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2700 			dst_addr_width = src_addr_width;
2701 		if (dst_maxburst == 0)
2702 			dst_maxburst = src_maxburst;
2703 
2704 	} else if (direction == DMA_MEM_TO_DEV) {
2705 		config_addr = config->dst_addr;
2706 
2707 		if (cfg->dir != DMA_MEM_TO_DEV)
2708 			dev_dbg(d40c->base->dev,
2709 				"channel was not configured for memory "
2710 				"to peripheral transfer (%d) overriding\n",
2711 				cfg->dir);
2712 		cfg->dir = DMA_MEM_TO_DEV;
2713 
2714 		/* Configure the memory side */
2715 		if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2716 			src_addr_width = dst_addr_width;
2717 		if (src_maxburst == 0)
2718 			src_maxburst = dst_maxburst;
2719 	} else {
2720 		dev_err(d40c->base->dev,
2721 			"unrecognized channel direction %d\n",
2722 			direction);
2723 		return -EINVAL;
2724 	}
2725 
2726 	if (config_addr <= 0) {
2727 		dev_err(d40c->base->dev, "no address supplied\n");
2728 		return -EINVAL;
2729 	}
2730 
2731 	if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2732 		dev_err(d40c->base->dev,
2733 			"src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2734 			src_maxburst,
2735 			src_addr_width,
2736 			dst_maxburst,
2737 			dst_addr_width);
2738 		return -EINVAL;
2739 	}
2740 
2741 	if (src_maxburst > 16) {
2742 		src_maxburst = 16;
2743 		dst_maxburst = src_maxburst * src_addr_width / dst_addr_width;
2744 	} else if (dst_maxburst > 16) {
2745 		dst_maxburst = 16;
2746 		src_maxburst = dst_maxburst * dst_addr_width / src_addr_width;
2747 	}
2748 
2749 	/* Only valid widths are; 1, 2, 4 and 8. */
2750 	if (src_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2751 	    src_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2752 	    dst_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2753 	    dst_addr_width >  DMA_SLAVE_BUSWIDTH_8_BYTES   ||
2754 	    !is_power_of_2(src_addr_width) ||
2755 	    !is_power_of_2(dst_addr_width))
2756 		return -EINVAL;
2757 
2758 	cfg->src_info.data_width = src_addr_width;
2759 	cfg->dst_info.data_width = dst_addr_width;
2760 
2761 	ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2762 					  src_maxburst);
2763 	if (ret)
2764 		return ret;
2765 
2766 	ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2767 					  dst_maxburst);
2768 	if (ret)
2769 		return ret;
2770 
2771 	/* Fill in register values */
2772 	if (chan_is_logical(d40c))
2773 		d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2774 	else
2775 		d40_phy_cfg(cfg, &d40c->src_def_cfg, &d40c->dst_def_cfg);
2776 
2777 	/* These settings will take precedence later */
2778 	d40c->runtime_addr = config_addr;
2779 	d40c->runtime_direction = direction;
2780 	dev_dbg(d40c->base->dev,
2781 		"configured channel %s for %s, data width %d/%d, "
2782 		"maxburst %d/%d elements, LE, no flow control\n",
2783 		dma_chan_name(chan),
2784 		(direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2785 		src_addr_width, dst_addr_width,
2786 		src_maxburst, dst_maxburst);
2787 
2788 	return 0;
2789 }
2790 
2791 /* Initialization functions */
2792 
2793 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2794 				 struct d40_chan *chans, int offset,
2795 				 int num_chans)
2796 {
2797 	int i = 0;
2798 	struct d40_chan *d40c;
2799 
2800 	INIT_LIST_HEAD(&dma->channels);
2801 
2802 	for (i = offset; i < offset + num_chans; i++) {
2803 		d40c = &chans[i];
2804 		d40c->base = base;
2805 		d40c->chan.device = dma;
2806 
2807 		spin_lock_init(&d40c->lock);
2808 
2809 		d40c->log_num = D40_PHY_CHAN;
2810 
2811 		INIT_LIST_HEAD(&d40c->done);
2812 		INIT_LIST_HEAD(&d40c->active);
2813 		INIT_LIST_HEAD(&d40c->queue);
2814 		INIT_LIST_HEAD(&d40c->pending_queue);
2815 		INIT_LIST_HEAD(&d40c->client);
2816 		INIT_LIST_HEAD(&d40c->prepare_queue);
2817 
2818 		tasklet_setup(&d40c->tasklet, dma_tasklet);
2819 
2820 		list_add_tail(&d40c->chan.device_node,
2821 			      &dma->channels);
2822 	}
2823 }
2824 
2825 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2826 {
2827 	if (dma_has_cap(DMA_SLAVE, dev->cap_mask)) {
2828 		dev->device_prep_slave_sg = d40_prep_slave_sg;
2829 		dev->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
2830 	}
2831 
2832 	if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2833 		dev->device_prep_dma_memcpy = d40_prep_memcpy;
2834 		dev->directions = BIT(DMA_MEM_TO_MEM);
2835 		/*
2836 		 * This controller can only access address at even
2837 		 * 32bit boundaries, i.e. 2^2
2838 		 */
2839 		dev->copy_align = DMAENGINE_ALIGN_4_BYTES;
2840 	}
2841 
2842 	if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2843 		dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2844 
2845 	dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2846 	dev->device_free_chan_resources = d40_free_chan_resources;
2847 	dev->device_issue_pending = d40_issue_pending;
2848 	dev->device_tx_status = d40_tx_status;
2849 	dev->device_config = d40_set_runtime_config;
2850 	dev->device_pause = d40_pause;
2851 	dev->device_resume = d40_resume;
2852 	dev->device_terminate_all = d40_terminate_all;
2853 	dev->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
2854 	dev->dev = base->dev;
2855 }
2856 
2857 static int __init d40_dmaengine_init(struct d40_base *base,
2858 				     int num_reserved_chans)
2859 {
2860 	int err ;
2861 
2862 	d40_chan_init(base, &base->dma_slave, base->log_chans,
2863 		      0, base->num_log_chans);
2864 
2865 	dma_cap_zero(base->dma_slave.cap_mask);
2866 	dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2867 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2868 
2869 	d40_ops_init(base, &base->dma_slave);
2870 
2871 	err = dmaenginem_async_device_register(&base->dma_slave);
2872 
2873 	if (err) {
2874 		d40_err(base->dev, "Failed to register slave channels\n");
2875 		goto exit;
2876 	}
2877 
2878 	d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2879 		      base->num_log_chans, base->num_memcpy_chans);
2880 
2881 	dma_cap_zero(base->dma_memcpy.cap_mask);
2882 	dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2883 
2884 	d40_ops_init(base, &base->dma_memcpy);
2885 
2886 	err = dmaenginem_async_device_register(&base->dma_memcpy);
2887 
2888 	if (err) {
2889 		d40_err(base->dev,
2890 			"Failed to register memcpy only channels\n");
2891 		goto exit;
2892 	}
2893 
2894 	d40_chan_init(base, &base->dma_both, base->phy_chans,
2895 		      0, num_reserved_chans);
2896 
2897 	dma_cap_zero(base->dma_both.cap_mask);
2898 	dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2899 	dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2900 	dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2901 
2902 	d40_ops_init(base, &base->dma_both);
2903 	err = dmaenginem_async_device_register(&base->dma_both);
2904 
2905 	if (err) {
2906 		d40_err(base->dev,
2907 			"Failed to register logical and physical capable channels\n");
2908 		goto exit;
2909 	}
2910 	return 0;
2911  exit:
2912 	return err;
2913 }
2914 
2915 /* Suspend resume functionality */
2916 #ifdef CONFIG_PM_SLEEP
2917 static int dma40_suspend(struct device *dev)
2918 {
2919 	struct d40_base *base = dev_get_drvdata(dev);
2920 	int ret;
2921 
2922 	ret = pm_runtime_force_suspend(dev);
2923 	if (ret)
2924 		return ret;
2925 
2926 	if (base->lcpa_regulator)
2927 		ret = regulator_disable(base->lcpa_regulator);
2928 	return ret;
2929 }
2930 
2931 static int dma40_resume(struct device *dev)
2932 {
2933 	struct d40_base *base = dev_get_drvdata(dev);
2934 	int ret = 0;
2935 
2936 	if (base->lcpa_regulator) {
2937 		ret = regulator_enable(base->lcpa_regulator);
2938 		if (ret)
2939 			return ret;
2940 	}
2941 
2942 	return pm_runtime_force_resume(dev);
2943 }
2944 #endif
2945 
2946 #ifdef CONFIG_PM
2947 static void dma40_backup(void __iomem *baseaddr, u32 *backup,
2948 			 u32 *regaddr, int num, bool save)
2949 {
2950 	int i;
2951 
2952 	for (i = 0; i < num; i++) {
2953 		void __iomem *addr = baseaddr + regaddr[i];
2954 
2955 		if (save)
2956 			backup[i] = readl_relaxed(addr);
2957 		else
2958 			writel_relaxed(backup[i], addr);
2959 	}
2960 }
2961 
2962 static void d40_save_restore_registers(struct d40_base *base, bool save)
2963 {
2964 	int i;
2965 
2966 	/* Save/Restore channel specific registers */
2967 	for (i = 0; i < base->num_phy_chans; i++) {
2968 		void __iomem *addr;
2969 		int idx;
2970 
2971 		if (base->phy_res[i].reserved)
2972 			continue;
2973 
2974 		addr = base->virtbase + D40_DREG_PCBASE + i * D40_DREG_PCDELTA;
2975 		idx = i * ARRAY_SIZE(d40_backup_regs_chan);
2976 
2977 		dma40_backup(addr, &base->reg_val_backup_chan[idx],
2978 			     d40_backup_regs_chan,
2979 			     ARRAY_SIZE(d40_backup_regs_chan),
2980 			     save);
2981 	}
2982 
2983 	/* Save/Restore global registers */
2984 	dma40_backup(base->virtbase, base->reg_val_backup,
2985 		     d40_backup_regs, ARRAY_SIZE(d40_backup_regs),
2986 		     save);
2987 
2988 	/* Save/Restore registers only existing on dma40 v3 and later */
2989 	if (base->gen_dmac.backup)
2990 		dma40_backup(base->virtbase, base->reg_val_backup_v4,
2991 			     base->gen_dmac.backup,
2992 			base->gen_dmac.backup_size,
2993 			save);
2994 }
2995 
2996 static int dma40_runtime_suspend(struct device *dev)
2997 {
2998 	struct d40_base *base = dev_get_drvdata(dev);
2999 
3000 	d40_save_restore_registers(base, true);
3001 
3002 	/* Don't disable/enable clocks for v1 due to HW bugs */
3003 	if (base->rev != 1)
3004 		writel_relaxed(base->gcc_pwr_off_mask,
3005 			       base->virtbase + D40_DREG_GCC);
3006 
3007 	return 0;
3008 }
3009 
3010 static int dma40_runtime_resume(struct device *dev)
3011 {
3012 	struct d40_base *base = dev_get_drvdata(dev);
3013 
3014 	d40_save_restore_registers(base, false);
3015 
3016 	writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
3017 		       base->virtbase + D40_DREG_GCC);
3018 	return 0;
3019 }
3020 #endif
3021 
3022 static const struct dev_pm_ops dma40_pm_ops = {
3023 	SET_LATE_SYSTEM_SLEEP_PM_OPS(dma40_suspend, dma40_resume)
3024 	SET_RUNTIME_PM_OPS(dma40_runtime_suspend,
3025 				dma40_runtime_resume,
3026 				NULL)
3027 };
3028 
3029 /* Initialization functions. */
3030 
3031 static int __init d40_phy_res_init(struct d40_base *base)
3032 {
3033 	int i;
3034 	int num_phy_chans_avail = 0;
3035 	u32 val[2];
3036 	int odd_even_bit = -2;
3037 	int gcc = D40_DREG_GCC_ENA;
3038 
3039 	val[0] = readl(base->virtbase + D40_DREG_PRSME);
3040 	val[1] = readl(base->virtbase + D40_DREG_PRSMO);
3041 
3042 	for (i = 0; i < base->num_phy_chans; i++) {
3043 		base->phy_res[i].num = i;
3044 		odd_even_bit += 2 * ((i % 2) == 0);
3045 		if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
3046 			/* Mark security only channels as occupied */
3047 			base->phy_res[i].allocated_src = D40_ALLOC_PHY;
3048 			base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
3049 			base->phy_res[i].reserved = true;
3050 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3051 						       D40_DREG_GCC_SRC);
3052 			gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3053 						       D40_DREG_GCC_DST);
3054 
3055 
3056 		} else {
3057 			base->phy_res[i].allocated_src = D40_ALLOC_FREE;
3058 			base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
3059 			base->phy_res[i].reserved = false;
3060 			num_phy_chans_avail++;
3061 		}
3062 		spin_lock_init(&base->phy_res[i].lock);
3063 	}
3064 
3065 	/* Mark disabled channels as occupied */
3066 	for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
3067 		int chan = base->plat_data->disabled_channels[i];
3068 
3069 		base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
3070 		base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
3071 		base->phy_res[chan].reserved = true;
3072 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3073 					       D40_DREG_GCC_SRC);
3074 		gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3075 					       D40_DREG_GCC_DST);
3076 		num_phy_chans_avail--;
3077 	}
3078 
3079 	/* Mark soft_lli channels */
3080 	for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) {
3081 		int chan = base->plat_data->soft_lli_chans[i];
3082 
3083 		base->phy_res[chan].use_soft_lli = true;
3084 	}
3085 
3086 	dev_info(base->dev, "%d of %d physical DMA channels available\n",
3087 		 num_phy_chans_avail, base->num_phy_chans);
3088 
3089 	/* Verify settings extended vs standard */
3090 	val[0] = readl(base->virtbase + D40_DREG_PRTYP);
3091 
3092 	for (i = 0; i < base->num_phy_chans; i++) {
3093 
3094 		if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
3095 		    (val[0] & 0x3) != 1)
3096 			dev_info(base->dev,
3097 				 "[%s] INFO: channel %d is misconfigured (%d)\n",
3098 				 __func__, i, val[0] & 0x3);
3099 
3100 		val[0] = val[0] >> 2;
3101 	}
3102 
3103 	/*
3104 	 * To keep things simple, Enable all clocks initially.
3105 	 * The clocks will get managed later post channel allocation.
3106 	 * The clocks for the event lines on which reserved channels exists
3107 	 * are not managed here.
3108 	 */
3109 	writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3110 	base->gcc_pwr_off_mask = gcc;
3111 
3112 	return num_phy_chans_avail;
3113 }
3114 
3115 /* Called from the registered devm action */
3116 static void d40_drop_kmem_cache_action(void *d)
3117 {
3118 	struct kmem_cache *desc_slab = d;
3119 
3120 	kmem_cache_destroy(desc_slab);
3121 }
3122 
3123 static int __init d40_hw_detect_init(struct platform_device *pdev,
3124 				     struct d40_base **retbase)
3125 {
3126 	struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3127 	struct device *dev = &pdev->dev;
3128 	struct clk *clk;
3129 	void __iomem *virtbase;
3130 	struct d40_base *base;
3131 	int num_log_chans;
3132 	int num_phy_chans;
3133 	int num_memcpy_chans;
3134 	int i;
3135 	u32 pid;
3136 	u32 cid;
3137 	u8 rev;
3138 	int ret;
3139 
3140 	clk = devm_clk_get_enabled(dev, NULL);
3141 	if (IS_ERR(clk))
3142 		return PTR_ERR(clk);
3143 
3144 	/* Get IO for DMAC base address */
3145 	virtbase = devm_platform_ioremap_resource_byname(pdev, "base");
3146 	if (IS_ERR(virtbase))
3147 		return PTR_ERR(virtbase);
3148 
3149 	/* This is just a regular AMBA PrimeCell ID actually */
3150 	for (pid = 0, i = 0; i < 4; i++)
3151 		pid |= (readl(virtbase + SZ_4K - 0x20 + 4 * i)
3152 			& 255) << (i * 8);
3153 	for (cid = 0, i = 0; i < 4; i++)
3154 		cid |= (readl(virtbase + SZ_4K - 0x10 + 4 * i)
3155 			& 255) << (i * 8);
3156 
3157 	if (cid != AMBA_CID) {
3158 		d40_err(dev, "Unknown hardware! No PrimeCell ID\n");
3159 		return -EINVAL;
3160 	}
3161 	if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
3162 		d40_err(dev, "Unknown designer! Got %x wanted %x\n",
3163 			AMBA_MANF_BITS(pid),
3164 			AMBA_VENDOR_ST);
3165 		return -EINVAL;
3166 	}
3167 	/*
3168 	 * HW revision:
3169 	 * DB8500ed has revision 0
3170 	 * ? has revision 1
3171 	 * DB8500v1 has revision 2
3172 	 * DB8500v2 has revision 3
3173 	 * AP9540v1 has revision 4
3174 	 * DB8540v1 has revision 4
3175 	 */
3176 	rev = AMBA_REV_BITS(pid);
3177 	if (rev < 2) {
3178 		d40_err(dev, "hardware revision: %d is not supported", rev);
3179 		return -EINVAL;
3180 	}
3181 
3182 	/* The number of physical channels on this HW */
3183 	if (plat_data->num_of_phy_chans)
3184 		num_phy_chans = plat_data->num_of_phy_chans;
3185 	else
3186 		num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
3187 
3188 	/* The number of channels used for memcpy */
3189 	if (plat_data->num_of_memcpy_chans)
3190 		num_memcpy_chans = plat_data->num_of_memcpy_chans;
3191 	else
3192 		num_memcpy_chans = ARRAY_SIZE(dma40_memcpy_channels);
3193 
3194 	num_log_chans = num_phy_chans * D40_MAX_LOG_CHAN_PER_PHY;
3195 
3196 	dev_info(dev,
3197 		 "hardware rev: %d with %d physical and %d logical channels\n",
3198 		 rev, num_phy_chans, num_log_chans);
3199 
3200 	base = devm_kzalloc(dev,
3201 		ALIGN(sizeof(struct d40_base), 4) +
3202 		(num_phy_chans + num_log_chans + num_memcpy_chans) *
3203 		sizeof(struct d40_chan), GFP_KERNEL);
3204 
3205 	if (!base)
3206 		return -ENOMEM;
3207 
3208 	base->rev = rev;
3209 	base->clk = clk;
3210 	base->num_memcpy_chans = num_memcpy_chans;
3211 	base->num_phy_chans = num_phy_chans;
3212 	base->num_log_chans = num_log_chans;
3213 	base->virtbase = virtbase;
3214 	base->plat_data = plat_data;
3215 	base->dev = dev;
3216 	base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3217 	base->log_chans = &base->phy_chans[num_phy_chans];
3218 
3219 	if (base->plat_data->num_of_phy_chans == 14) {
3220 		base->gen_dmac.backup = d40_backup_regs_v4b;
3221 		base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
3222 		base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
3223 		base->gen_dmac.interrupt_clear = D40_DREG_CPCICR;
3224 		base->gen_dmac.realtime_en = D40_DREG_CRSEG1;
3225 		base->gen_dmac.realtime_clear = D40_DREG_CRCEG1;
3226 		base->gen_dmac.high_prio_en = D40_DREG_CPSEG1;
3227 		base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1;
3228 		base->gen_dmac.il = il_v4b;
3229 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4b);
3230 		base->gen_dmac.init_reg = dma_init_reg_v4b;
3231 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
3232 	} else {
3233 		if (base->rev >= 3) {
3234 			base->gen_dmac.backup = d40_backup_regs_v4a;
3235 			base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
3236 		}
3237 		base->gen_dmac.interrupt_en = D40_DREG_PCMIS;
3238 		base->gen_dmac.interrupt_clear = D40_DREG_PCICR;
3239 		base->gen_dmac.realtime_en = D40_DREG_RSEG1;
3240 		base->gen_dmac.realtime_clear = D40_DREG_RCEG1;
3241 		base->gen_dmac.high_prio_en = D40_DREG_PSEG1;
3242 		base->gen_dmac.high_prio_clear = D40_DREG_PCEG1;
3243 		base->gen_dmac.il = il_v4a;
3244 		base->gen_dmac.il_size = ARRAY_SIZE(il_v4a);
3245 		base->gen_dmac.init_reg = dma_init_reg_v4a;
3246 		base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a);
3247 	}
3248 
3249 	base->phy_res = devm_kcalloc(dev, num_phy_chans,
3250 				     sizeof(*base->phy_res),
3251 				     GFP_KERNEL);
3252 	if (!base->phy_res)
3253 		return -ENOMEM;
3254 
3255 	base->lookup_phy_chans = devm_kcalloc(dev, num_phy_chans,
3256 					      sizeof(*base->lookup_phy_chans),
3257 					      GFP_KERNEL);
3258 	if (!base->lookup_phy_chans)
3259 		return -ENOMEM;
3260 
3261 	base->lookup_log_chans = devm_kcalloc(dev, num_log_chans,
3262 					      sizeof(*base->lookup_log_chans),
3263 					      GFP_KERNEL);
3264 	if (!base->lookup_log_chans)
3265 		return -ENOMEM;
3266 
3267 	base->reg_val_backup_chan = devm_kmalloc_array(dev, base->num_phy_chans,
3268 						  sizeof(d40_backup_regs_chan),
3269 						  GFP_KERNEL);
3270 	if (!base->reg_val_backup_chan)
3271 		return -ENOMEM;
3272 
3273 	base->lcla_pool.alloc_map = devm_kcalloc(dev, num_phy_chans
3274 					    * D40_LCLA_LINK_PER_EVENT_GRP,
3275 					    sizeof(*base->lcla_pool.alloc_map),
3276 					    GFP_KERNEL);
3277 	if (!base->lcla_pool.alloc_map)
3278 		return -ENOMEM;
3279 
3280 	base->regs_interrupt = devm_kmalloc_array(dev, base->gen_dmac.il_size,
3281 					     sizeof(*base->regs_interrupt),
3282 					     GFP_KERNEL);
3283 	if (!base->regs_interrupt)
3284 		return -ENOMEM;
3285 
3286 	base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3287 					    0, SLAB_HWCACHE_ALIGN,
3288 					    NULL);
3289 	if (!base->desc_slab)
3290 		return -ENOMEM;
3291 
3292 	ret = devm_add_action_or_reset(dev, d40_drop_kmem_cache_action,
3293 				       base->desc_slab);
3294 	if (ret)
3295 		return ret;
3296 
3297 	*retbase = base;
3298 
3299 	return 0;
3300 }
3301 
3302 static void __init d40_hw_init(struct d40_base *base)
3303 {
3304 
3305 	int i;
3306 	u32 prmseo[2] = {0, 0};
3307 	u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3308 	u32 pcmis = 0;
3309 	u32 pcicr = 0;
3310 	struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg;
3311 	u32 reg_size = base->gen_dmac.init_reg_size;
3312 
3313 	for (i = 0; i < reg_size; i++)
3314 		writel(dma_init_reg[i].val,
3315 		       base->virtbase + dma_init_reg[i].reg);
3316 
3317 	/* Configure all our dma channels to default settings */
3318 	for (i = 0; i < base->num_phy_chans; i++) {
3319 
3320 		activeo[i % 2] = activeo[i % 2] << 2;
3321 
3322 		if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3323 		    == D40_ALLOC_PHY) {
3324 			activeo[i % 2] |= 3;
3325 			continue;
3326 		}
3327 
3328 		/* Enable interrupt # */
3329 		pcmis = (pcmis << 1) | 1;
3330 
3331 		/* Clear interrupt # */
3332 		pcicr = (pcicr << 1) | 1;
3333 
3334 		/* Set channel to physical mode */
3335 		prmseo[i % 2] = prmseo[i % 2] << 2;
3336 		prmseo[i % 2] |= 1;
3337 
3338 	}
3339 
3340 	writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3341 	writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3342 	writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3343 	writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3344 
3345 	/* Write which interrupt to enable */
3346 	writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en);
3347 
3348 	/* Write which interrupt to clear */
3349 	writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear);
3350 
3351 	/* These are __initdata and cannot be accessed after init */
3352 	base->gen_dmac.init_reg = NULL;
3353 	base->gen_dmac.init_reg_size = 0;
3354 }
3355 
3356 static int __init d40_lcla_allocate(struct d40_base *base)
3357 {
3358 	struct d40_lcla_pool *pool = &base->lcla_pool;
3359 	unsigned long *page_list;
3360 	int i, j;
3361 	int ret;
3362 
3363 	/*
3364 	 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3365 	 * To full fill this hardware requirement without wasting 256 kb
3366 	 * we allocate pages until we get an aligned one.
3367 	 */
3368 	page_list = kmalloc_array(MAX_LCLA_ALLOC_ATTEMPTS,
3369 				  sizeof(*page_list),
3370 				  GFP_KERNEL);
3371 	if (!page_list)
3372 		return -ENOMEM;
3373 
3374 	/* Calculating how many pages that are required */
3375 	base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3376 
3377 	for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3378 		page_list[i] = __get_free_pages(GFP_KERNEL,
3379 						base->lcla_pool.pages);
3380 		if (!page_list[i]) {
3381 
3382 			d40_err(base->dev, "Failed to allocate %d pages.\n",
3383 				base->lcla_pool.pages);
3384 			ret = -ENOMEM;
3385 
3386 			for (j = 0; j < i; j++)
3387 				free_pages(page_list[j], base->lcla_pool.pages);
3388 			goto free_page_list;
3389 		}
3390 
3391 		if ((virt_to_phys((void *)page_list[i]) &
3392 		     (LCLA_ALIGNMENT - 1)) == 0)
3393 			break;
3394 	}
3395 
3396 	for (j = 0; j < i; j++)
3397 		free_pages(page_list[j], base->lcla_pool.pages);
3398 
3399 	if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3400 		base->lcla_pool.base = (void *)page_list[i];
3401 	} else {
3402 		/*
3403 		 * After many attempts and no success with finding the correct
3404 		 * alignment, try with allocating a big buffer.
3405 		 */
3406 		dev_warn(base->dev,
3407 			 "[%s] Failed to get %d pages @ 18 bit align.\n",
3408 			 __func__, base->lcla_pool.pages);
3409 		base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3410 							 base->num_phy_chans +
3411 							 LCLA_ALIGNMENT,
3412 							 GFP_KERNEL);
3413 		if (!base->lcla_pool.base_unaligned) {
3414 			ret = -ENOMEM;
3415 			goto free_page_list;
3416 		}
3417 
3418 		base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3419 						 LCLA_ALIGNMENT);
3420 	}
3421 
3422 	pool->dma_addr = dma_map_single(base->dev, pool->base,
3423 					SZ_1K * base->num_phy_chans,
3424 					DMA_TO_DEVICE);
3425 	if (dma_mapping_error(base->dev, pool->dma_addr)) {
3426 		pool->dma_addr = 0;
3427 		ret = -ENOMEM;
3428 		goto free_page_list;
3429 	}
3430 
3431 	writel(virt_to_phys(base->lcla_pool.base),
3432 	       base->virtbase + D40_DREG_LCLA);
3433 	ret = 0;
3434  free_page_list:
3435 	kfree(page_list);
3436 	return ret;
3437 }
3438 
3439 static int __init d40_of_probe(struct device *dev,
3440 			       struct device_node *np)
3441 {
3442 	struct stedma40_platform_data *pdata;
3443 	int num_phy = 0, num_memcpy = 0, num_disabled = 0;
3444 	const __be32 *list;
3445 
3446 	pdata = devm_kzalloc(dev, sizeof(*pdata), GFP_KERNEL);
3447 	if (!pdata)
3448 		return -ENOMEM;
3449 
3450 	/* If absent this value will be obtained from h/w. */
3451 	of_property_read_u32(np, "dma-channels", &num_phy);
3452 	if (num_phy > 0)
3453 		pdata->num_of_phy_chans = num_phy;
3454 
3455 	list = of_get_property(np, "memcpy-channels", &num_memcpy);
3456 	num_memcpy /= sizeof(*list);
3457 
3458 	if (num_memcpy > D40_MEMCPY_MAX_CHANS || num_memcpy <= 0) {
3459 		d40_err(dev,
3460 			"Invalid number of memcpy channels specified (%d)\n",
3461 			num_memcpy);
3462 		return -EINVAL;
3463 	}
3464 	pdata->num_of_memcpy_chans = num_memcpy;
3465 
3466 	of_property_read_u32_array(np, "memcpy-channels",
3467 				   dma40_memcpy_channels,
3468 				   num_memcpy);
3469 
3470 	list = of_get_property(np, "disabled-channels", &num_disabled);
3471 	num_disabled /= sizeof(*list);
3472 
3473 	if (num_disabled >= STEDMA40_MAX_PHYS || num_disabled < 0) {
3474 		d40_err(dev,
3475 			"Invalid number of disabled channels specified (%d)\n",
3476 			num_disabled);
3477 		return -EINVAL;
3478 	}
3479 
3480 	of_property_read_u32_array(np, "disabled-channels",
3481 				   pdata->disabled_channels,
3482 				   num_disabled);
3483 	pdata->disabled_channels[num_disabled] = -1;
3484 
3485 	dev->platform_data = pdata;
3486 
3487 	return 0;
3488 }
3489 
3490 static int __init d40_probe(struct platform_device *pdev)
3491 {
3492 	struct device *dev = &pdev->dev;
3493 	struct device_node *np = pdev->dev.of_node;
3494 	struct device_node *np_lcpa;
3495 	struct d40_base *base;
3496 	struct resource *res;
3497 	struct resource res_lcpa;
3498 	int num_reserved_chans;
3499 	u32 val;
3500 	int ret;
3501 
3502 	if (d40_of_probe(dev, np)) {
3503 		ret = -ENOMEM;
3504 		goto report_failure;
3505 	}
3506 
3507 	ret = d40_hw_detect_init(pdev, &base);
3508 	if (ret)
3509 		goto report_failure;
3510 
3511 	num_reserved_chans = d40_phy_res_init(base);
3512 
3513 	platform_set_drvdata(pdev, base);
3514 
3515 	spin_lock_init(&base->interrupt_lock);
3516 	spin_lock_init(&base->execmd_lock);
3517 
3518 	/* Get IO for logical channel parameter address (LCPA) */
3519 	np_lcpa = of_parse_phandle(np, "sram", 0);
3520 	if (!np_lcpa) {
3521 		dev_err(dev, "no LCPA SRAM node\n");
3522 		ret = -EINVAL;
3523 		goto report_failure;
3524 	}
3525 	/* This is no device so read the address directly from the node */
3526 	ret = of_address_to_resource(np_lcpa, 0, &res_lcpa);
3527 	if (ret) {
3528 		dev_err(dev, "no LCPA SRAM resource\n");
3529 		goto report_failure;
3530 	}
3531 	base->lcpa_size = resource_size(&res_lcpa);
3532 	base->phy_lcpa = res_lcpa.start;
3533 	dev_info(dev, "found LCPA SRAM at %pad, size %pa\n",
3534 		 &base->phy_lcpa, &base->lcpa_size);
3535 
3536 	/* We make use of ESRAM memory for this. */
3537 	val = readl(base->virtbase + D40_DREG_LCPA);
3538 	if (base->phy_lcpa != val && val != 0) {
3539 		dev_warn(dev,
3540 			 "[%s] Mismatch LCPA dma 0x%x, def %08x\n",
3541 			 __func__, val, (u32)base->phy_lcpa);
3542 	} else
3543 		writel(base->phy_lcpa, base->virtbase + D40_DREG_LCPA);
3544 
3545 	base->lcpa_base = devm_ioremap(dev, base->phy_lcpa, base->lcpa_size);
3546 	if (!base->lcpa_base) {
3547 		ret = -ENOMEM;
3548 		d40_err(dev, "Failed to ioremap LCPA region\n");
3549 		goto report_failure;
3550 	}
3551 	/* If lcla has to be located in ESRAM we don't need to allocate */
3552 	if (base->plat_data->use_esram_lcla) {
3553 		res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3554 							"lcla_esram");
3555 		if (!res) {
3556 			ret = -ENOENT;
3557 			d40_err(dev,
3558 				"No \"lcla_esram\" memory resource\n");
3559 			goto report_failure;
3560 		}
3561 		base->lcla_pool.base = devm_ioremap(dev, res->start,
3562 						    resource_size(res));
3563 		if (!base->lcla_pool.base) {
3564 			ret = -ENOMEM;
3565 			d40_err(dev, "Failed to ioremap LCLA region\n");
3566 			goto report_failure;
3567 		}
3568 		writel(res->start, base->virtbase + D40_DREG_LCLA);
3569 
3570 	} else {
3571 		ret = d40_lcla_allocate(base);
3572 		if (ret) {
3573 			d40_err(dev, "Failed to allocate LCLA area\n");
3574 			goto destroy_cache;
3575 		}
3576 	}
3577 
3578 	spin_lock_init(&base->lcla_pool.lock);
3579 
3580 	base->irq = platform_get_irq(pdev, 0);
3581 	if (base->irq < 0) {
3582 		ret = base->irq;
3583 		goto destroy_cache;
3584 	}
3585 
3586 	ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3587 	if (ret) {
3588 		d40_err(dev, "No IRQ defined\n");
3589 		goto destroy_cache;
3590 	}
3591 
3592 	if (base->plat_data->use_esram_lcla) {
3593 
3594 		base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3595 		if (IS_ERR(base->lcpa_regulator)) {
3596 			d40_err(dev, "Failed to get lcpa_regulator\n");
3597 			ret = PTR_ERR(base->lcpa_regulator);
3598 			base->lcpa_regulator = NULL;
3599 			goto destroy_cache;
3600 		}
3601 
3602 		ret = regulator_enable(base->lcpa_regulator);
3603 		if (ret) {
3604 			d40_err(dev,
3605 				"Failed to enable lcpa_regulator\n");
3606 			regulator_put(base->lcpa_regulator);
3607 			base->lcpa_regulator = NULL;
3608 			goto destroy_cache;
3609 		}
3610 	}
3611 
3612 	writel_relaxed(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3613 
3614 	pm_runtime_irq_safe(base->dev);
3615 	pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3616 	pm_runtime_use_autosuspend(base->dev);
3617 	pm_runtime_mark_last_busy(base->dev);
3618 	pm_runtime_set_active(base->dev);
3619 	pm_runtime_enable(base->dev);
3620 
3621 	ret = d40_dmaengine_init(base, num_reserved_chans);
3622 	if (ret)
3623 		goto destroy_cache;
3624 
3625 	dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
3626 
3627 	d40_hw_init(base);
3628 
3629 	ret = of_dma_controller_register(np, d40_xlate, NULL);
3630 	if (ret) {
3631 		dev_err(dev,
3632 			"could not register of_dma_controller\n");
3633 		goto destroy_cache;
3634 	}
3635 
3636 	dev_info(base->dev, "initialized\n");
3637 	return 0;
3638 
3639  destroy_cache:
3640 	if (base->lcla_pool.dma_addr)
3641 		dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3642 				 SZ_1K * base->num_phy_chans,
3643 				 DMA_TO_DEVICE);
3644 
3645 	if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3646 		free_pages((unsigned long)base->lcla_pool.base,
3647 			   base->lcla_pool.pages);
3648 
3649 	kfree(base->lcla_pool.base_unaligned);
3650 
3651 	if (base->lcpa_regulator) {
3652 		regulator_disable(base->lcpa_regulator);
3653 		regulator_put(base->lcpa_regulator);
3654 	}
3655 	pm_runtime_disable(base->dev);
3656 
3657  report_failure:
3658 	d40_err(dev, "probe failed\n");
3659 	return ret;
3660 }
3661 
3662 static const struct of_device_id d40_match[] = {
3663         { .compatible = "stericsson,dma40", },
3664         {}
3665 };
3666 
3667 static struct platform_driver d40_driver = {
3668 	.driver = {
3669 		.name  = D40_NAME,
3670 		.pm = &dma40_pm_ops,
3671 		.of_match_table = d40_match,
3672 	},
3673 };
3674 
3675 static int __init stedma40_init(void)
3676 {
3677 	return platform_driver_probe(&d40_driver, d40_probe);
3678 }
3679 subsys_initcall(stedma40_init);
3680