Merge commit 'ed30f24e8d07d30aa3e69d1f508f4d7bd2e8ea14' of git://git.linaro.org/landi...
[firefly-linux-kernel-4.4.55.git] / drivers / dma / ste_dma40.c
1 /*
2  * Copyright (C) Ericsson AB 2007-2008
3  * Copyright (C) ST-Ericsson SA 2008-2010
4  * Author: Per Forlin <per.forlin@stericsson.com> for ST-Ericsson
5  * Author: Jonas Aaberg <jonas.aberg@stericsson.com> for ST-Ericsson
6  * License terms: GNU General Public License (GPL) version 2
7  */
8
9 #include <linux/dma-mapping.h>
10 #include <linux/kernel.h>
11 #include <linux/slab.h>
12 #include <linux/export.h>
13 #include <linux/dmaengine.h>
14 #include <linux/platform_device.h>
15 #include <linux/clk.h>
16 #include <linux/delay.h>
17 #include <linux/pm.h>
18 #include <linux/pm_runtime.h>
19 #include <linux/err.h>
20 #include <linux/amba/bus.h>
21 #include <linux/regulator/consumer.h>
22 #include <linux/platform_data/dma-ste-dma40.h>
23
24 #include "dmaengine.h"
25 #include "ste_dma40_ll.h"
26
27 #define D40_NAME "dma40"
28
29 #define D40_PHY_CHAN -1
30
31 /* For masking out/in 2 bit channel positions */
32 #define D40_CHAN_POS(chan)  (2 * (chan / 2))
33 #define D40_CHAN_POS_MASK(chan) (0x3 << D40_CHAN_POS(chan))
34
35 /* Maximum iterations taken before giving up suspending a channel */
36 #define D40_SUSPEND_MAX_IT 500
37
38 /* Milliseconds */
39 #define DMA40_AUTOSUSPEND_DELAY 100
40
41 /* Hardware requirement on LCLA alignment */
42 #define LCLA_ALIGNMENT 0x40000
43
44 /* Max number of links per event group */
45 #define D40_LCLA_LINK_PER_EVENT_GRP 128
46 #define D40_LCLA_END D40_LCLA_LINK_PER_EVENT_GRP
47
48 /* Attempts before giving up to trying to get pages that are aligned */
49 #define MAX_LCLA_ALLOC_ATTEMPTS 256
50
51 /* Bit markings for allocation map */
52 #define D40_ALLOC_FREE          (1 << 31)
53 #define D40_ALLOC_PHY           (1 << 30)
54 #define D40_ALLOC_LOG_FREE      0
55
56 #define MAX(a, b) (((a) < (b)) ? (b) : (a))
57
58 /**
59  * enum 40_command - The different commands and/or statuses.
60  *
61  * @D40_DMA_STOP: DMA channel command STOP or status STOPPED,
62  * @D40_DMA_RUN: The DMA channel is RUNNING of the command RUN.
63  * @D40_DMA_SUSPEND_REQ: Request the DMA to SUSPEND as soon as possible.
64  * @D40_DMA_SUSPENDED: The DMA channel is SUSPENDED.
65  */
66 enum d40_command {
67         D40_DMA_STOP            = 0,
68         D40_DMA_RUN             = 1,
69         D40_DMA_SUSPEND_REQ     = 2,
70         D40_DMA_SUSPENDED       = 3
71 };
72
73 /*
74  * enum d40_events - The different Event Enables for the event lines.
75  *
76  * @D40_DEACTIVATE_EVENTLINE: De-activate Event line, stopping the logical chan.
77  * @D40_ACTIVATE_EVENTLINE: Activate the Event line, to start a logical chan.
78  * @D40_SUSPEND_REQ_EVENTLINE: Requesting for suspending a event line.
79  * @D40_ROUND_EVENTLINE: Status check for event line.
80  */
81
82 enum d40_events {
83         D40_DEACTIVATE_EVENTLINE        = 0,
84         D40_ACTIVATE_EVENTLINE          = 1,
85         D40_SUSPEND_REQ_EVENTLINE       = 2,
86         D40_ROUND_EVENTLINE             = 3
87 };
88
89 /*
90  * These are the registers that has to be saved and later restored
91  * when the DMA hw is powered off.
92  * TODO: Add save/restore of D40_DREG_GCC on dma40 v3 or later, if that works.
93  */
94 static u32 d40_backup_regs[] = {
95         D40_DREG_LCPA,
96         D40_DREG_LCLA,
97         D40_DREG_PRMSE,
98         D40_DREG_PRMSO,
99         D40_DREG_PRMOE,
100         D40_DREG_PRMOO,
101 };
102
103 #define BACKUP_REGS_SZ ARRAY_SIZE(d40_backup_regs)
104
105 /*
106  * since 9540 and 8540 has the same HW revision
107  * use v4a for 9540 or ealier
108  * use v4b for 8540 or later
109  * HW revision:
110  * DB8500ed has revision 0
111  * DB8500v1 has revision 2
112  * DB8500v2 has revision 3
113  * AP9540v1 has revision 4
114  * DB8540v1 has revision 4
115  * TODO: Check if all these registers have to be saved/restored on dma40 v4a
116  */
117 static u32 d40_backup_regs_v4a[] = {
118         D40_DREG_PSEG1,
119         D40_DREG_PSEG2,
120         D40_DREG_PSEG3,
121         D40_DREG_PSEG4,
122         D40_DREG_PCEG1,
123         D40_DREG_PCEG2,
124         D40_DREG_PCEG3,
125         D40_DREG_PCEG4,
126         D40_DREG_RSEG1,
127         D40_DREG_RSEG2,
128         D40_DREG_RSEG3,
129         D40_DREG_RSEG4,
130         D40_DREG_RCEG1,
131         D40_DREG_RCEG2,
132         D40_DREG_RCEG3,
133         D40_DREG_RCEG4,
134 };
135
136 #define BACKUP_REGS_SZ_V4A ARRAY_SIZE(d40_backup_regs_v4a)
137
138 static u32 d40_backup_regs_v4b[] = {
139         D40_DREG_CPSEG1,
140         D40_DREG_CPSEG2,
141         D40_DREG_CPSEG3,
142         D40_DREG_CPSEG4,
143         D40_DREG_CPSEG5,
144         D40_DREG_CPCEG1,
145         D40_DREG_CPCEG2,
146         D40_DREG_CPCEG3,
147         D40_DREG_CPCEG4,
148         D40_DREG_CPCEG5,
149         D40_DREG_CRSEG1,
150         D40_DREG_CRSEG2,
151         D40_DREG_CRSEG3,
152         D40_DREG_CRSEG4,
153         D40_DREG_CRSEG5,
154         D40_DREG_CRCEG1,
155         D40_DREG_CRCEG2,
156         D40_DREG_CRCEG3,
157         D40_DREG_CRCEG4,
158         D40_DREG_CRCEG5,
159 };
160
161 #define BACKUP_REGS_SZ_V4B ARRAY_SIZE(d40_backup_regs_v4b)
162
163 static u32 d40_backup_regs_chan[] = {
164         D40_CHAN_REG_SSCFG,
165         D40_CHAN_REG_SSELT,
166         D40_CHAN_REG_SSPTR,
167         D40_CHAN_REG_SSLNK,
168         D40_CHAN_REG_SDCFG,
169         D40_CHAN_REG_SDELT,
170         D40_CHAN_REG_SDPTR,
171         D40_CHAN_REG_SDLNK,
172 };
173
174 /**
175  * struct d40_interrupt_lookup - lookup table for interrupt handler
176  *
177  * @src: Interrupt mask register.
178  * @clr: Interrupt clear register.
179  * @is_error: true if this is an error interrupt.
180  * @offset: start delta in the lookup_log_chans in d40_base. If equals to
181  * D40_PHY_CHAN, the lookup_phy_chans shall be used instead.
182  */
183 struct d40_interrupt_lookup {
184         u32 src;
185         u32 clr;
186         bool is_error;
187         int offset;
188 };
189
190
191 static struct d40_interrupt_lookup il_v4a[] = {
192         {D40_DREG_LCTIS0, D40_DREG_LCICR0, false,  0},
193         {D40_DREG_LCTIS1, D40_DREG_LCICR1, false, 32},
194         {D40_DREG_LCTIS2, D40_DREG_LCICR2, false, 64},
195         {D40_DREG_LCTIS3, D40_DREG_LCICR3, false, 96},
196         {D40_DREG_LCEIS0, D40_DREG_LCICR0, true,   0},
197         {D40_DREG_LCEIS1, D40_DREG_LCICR1, true,  32},
198         {D40_DREG_LCEIS2, D40_DREG_LCICR2, true,  64},
199         {D40_DREG_LCEIS3, D40_DREG_LCICR3, true,  96},
200         {D40_DREG_PCTIS,  D40_DREG_PCICR,  false, D40_PHY_CHAN},
201         {D40_DREG_PCEIS,  D40_DREG_PCICR,  true,  D40_PHY_CHAN},
202 };
203
204 static struct d40_interrupt_lookup il_v4b[] = {
205         {D40_DREG_CLCTIS1, D40_DREG_CLCICR1, false,  0},
206         {D40_DREG_CLCTIS2, D40_DREG_CLCICR2, false, 32},
207         {D40_DREG_CLCTIS3, D40_DREG_CLCICR3, false, 64},
208         {D40_DREG_CLCTIS4, D40_DREG_CLCICR4, false, 96},
209         {D40_DREG_CLCTIS5, D40_DREG_CLCICR5, false, 128},
210         {D40_DREG_CLCEIS1, D40_DREG_CLCICR1, true,   0},
211         {D40_DREG_CLCEIS2, D40_DREG_CLCICR2, true,  32},
212         {D40_DREG_CLCEIS3, D40_DREG_CLCICR3, true,  64},
213         {D40_DREG_CLCEIS4, D40_DREG_CLCICR4, true,  96},
214         {D40_DREG_CLCEIS5, D40_DREG_CLCICR5, true,  128},
215         {D40_DREG_CPCTIS,  D40_DREG_CPCICR,  false, D40_PHY_CHAN},
216         {D40_DREG_CPCEIS,  D40_DREG_CPCICR,  true,  D40_PHY_CHAN},
217 };
218
219 /**
220  * struct d40_reg_val - simple lookup struct
221  *
222  * @reg: The register.
223  * @val: The value that belongs to the register in reg.
224  */
225 struct d40_reg_val {
226         unsigned int reg;
227         unsigned int val;
228 };
229
230 static __initdata struct d40_reg_val dma_init_reg_v4a[] = {
231         /* Clock every part of the DMA block from start */
232         { .reg = D40_DREG_GCC,    .val = D40_DREG_GCC_ENABLE_ALL},
233
234         /* Interrupts on all logical channels */
235         { .reg = D40_DREG_LCMIS0, .val = 0xFFFFFFFF},
236         { .reg = D40_DREG_LCMIS1, .val = 0xFFFFFFFF},
237         { .reg = D40_DREG_LCMIS2, .val = 0xFFFFFFFF},
238         { .reg = D40_DREG_LCMIS3, .val = 0xFFFFFFFF},
239         { .reg = D40_DREG_LCICR0, .val = 0xFFFFFFFF},
240         { .reg = D40_DREG_LCICR1, .val = 0xFFFFFFFF},
241         { .reg = D40_DREG_LCICR2, .val = 0xFFFFFFFF},
242         { .reg = D40_DREG_LCICR3, .val = 0xFFFFFFFF},
243         { .reg = D40_DREG_LCTIS0, .val = 0xFFFFFFFF},
244         { .reg = D40_DREG_LCTIS1, .val = 0xFFFFFFFF},
245         { .reg = D40_DREG_LCTIS2, .val = 0xFFFFFFFF},
246         { .reg = D40_DREG_LCTIS3, .val = 0xFFFFFFFF}
247 };
248 static __initdata struct d40_reg_val dma_init_reg_v4b[] = {
249         /* Clock every part of the DMA block from start */
250         { .reg = D40_DREG_GCC,    .val = D40_DREG_GCC_ENABLE_ALL},
251
252         /* Interrupts on all logical channels */
253         { .reg = D40_DREG_CLCMIS1, .val = 0xFFFFFFFF},
254         { .reg = D40_DREG_CLCMIS2, .val = 0xFFFFFFFF},
255         { .reg = D40_DREG_CLCMIS3, .val = 0xFFFFFFFF},
256         { .reg = D40_DREG_CLCMIS4, .val = 0xFFFFFFFF},
257         { .reg = D40_DREG_CLCMIS5, .val = 0xFFFFFFFF},
258         { .reg = D40_DREG_CLCICR1, .val = 0xFFFFFFFF},
259         { .reg = D40_DREG_CLCICR2, .val = 0xFFFFFFFF},
260         { .reg = D40_DREG_CLCICR3, .val = 0xFFFFFFFF},
261         { .reg = D40_DREG_CLCICR4, .val = 0xFFFFFFFF},
262         { .reg = D40_DREG_CLCICR5, .val = 0xFFFFFFFF},
263         { .reg = D40_DREG_CLCTIS1, .val = 0xFFFFFFFF},
264         { .reg = D40_DREG_CLCTIS2, .val = 0xFFFFFFFF},
265         { .reg = D40_DREG_CLCTIS3, .val = 0xFFFFFFFF},
266         { .reg = D40_DREG_CLCTIS4, .val = 0xFFFFFFFF},
267         { .reg = D40_DREG_CLCTIS5, .val = 0xFFFFFFFF}
268 };
269
270 /**
271  * struct d40_lli_pool - Structure for keeping LLIs in memory
272  *
273  * @base: Pointer to memory area when the pre_alloc_lli's are not large
274  * enough, IE bigger than the most common case, 1 dst and 1 src. NULL if
275  * pre_alloc_lli is used.
276  * @dma_addr: DMA address, if mapped
277  * @size: The size in bytes of the memory at base or the size of pre_alloc_lli.
278  * @pre_alloc_lli: Pre allocated area for the most common case of transfers,
279  * one buffer to one buffer.
280  */
281 struct d40_lli_pool {
282         void    *base;
283         int      size;
284         dma_addr_t      dma_addr;
285         /* Space for dst and src, plus an extra for padding */
286         u8       pre_alloc_lli[3 * sizeof(struct d40_phy_lli)];
287 };
288
289 /**
290  * struct d40_desc - A descriptor is one DMA job.
291  *
292  * @lli_phy: LLI settings for physical channel. Both src and dst=
293  * points into the lli_pool, to base if lli_len > 1 or to pre_alloc_lli if
294  * lli_len equals one.
295  * @lli_log: Same as above but for logical channels.
296  * @lli_pool: The pool with two entries pre-allocated.
297  * @lli_len: Number of llis of current descriptor.
298  * @lli_current: Number of transferred llis.
299  * @lcla_alloc: Number of LCLA entries allocated.
300  * @txd: DMA engine struct. Used for among other things for communication
301  * during a transfer.
302  * @node: List entry.
303  * @is_in_client_list: true if the client owns this descriptor.
304  * @cyclic: true if this is a cyclic job
305  *
306  * This descriptor is used for both logical and physical transfers.
307  */
308 struct d40_desc {
309         /* LLI physical */
310         struct d40_phy_lli_bidir         lli_phy;
311         /* LLI logical */
312         struct d40_log_lli_bidir         lli_log;
313
314         struct d40_lli_pool              lli_pool;
315         int                              lli_len;
316         int                              lli_current;
317         int                              lcla_alloc;
318
319         struct dma_async_tx_descriptor   txd;
320         struct list_head                 node;
321
322         bool                             is_in_client_list;
323         bool                             cyclic;
324 };
325
326 /**
327  * struct d40_lcla_pool - LCLA pool settings and data.
328  *
329  * @base: The virtual address of LCLA. 18 bit aligned.
330  * @base_unaligned: The orignal kmalloc pointer, if kmalloc is used.
331  * This pointer is only there for clean-up on error.
332  * @pages: The number of pages needed for all physical channels.
333  * Only used later for clean-up on error
334  * @lock: Lock to protect the content in this struct.
335  * @alloc_map: big map over which LCLA entry is own by which job.
336  */
337 struct d40_lcla_pool {
338         void            *base;
339         dma_addr_t      dma_addr;
340         void            *base_unaligned;
341         int              pages;
342         spinlock_t       lock;
343         struct d40_desc **alloc_map;
344 };
345
346 /**
347  * struct d40_phy_res - struct for handling eventlines mapped to physical
348  * channels.
349  *
350  * @lock: A lock protection this entity.
351  * @reserved: True if used by secure world or otherwise.
352  * @num: The physical channel number of this entity.
353  * @allocated_src: Bit mapped to show which src event line's are mapped to
354  * this physical channel. Can also be free or physically allocated.
355  * @allocated_dst: Same as for src but is dst.
356  * allocated_dst and allocated_src uses the D40_ALLOC* defines as well as
357  * event line number.
358  * @use_soft_lli: To mark if the linked lists of channel are managed by SW.
359  */
360 struct d40_phy_res {
361         spinlock_t lock;
362         bool       reserved;
363         int        num;
364         u32        allocated_src;
365         u32        allocated_dst;
366         bool       use_soft_lli;
367 };
368
369 struct d40_base;
370
371 /**
372  * struct d40_chan - Struct that describes a channel.
373  *
374  * @lock: A spinlock to protect this struct.
375  * @log_num: The logical number, if any of this channel.
376  * @pending_tx: The number of pending transfers. Used between interrupt handler
377  * and tasklet.
378  * @busy: Set to true when transfer is ongoing on this channel.
379  * @phy_chan: Pointer to physical channel which this instance runs on. If this
380  * point is NULL, then the channel is not allocated.
381  * @chan: DMA engine handle.
382  * @tasklet: Tasklet that gets scheduled from interrupt context to complete a
383  * transfer and call client callback.
384  * @client: Cliented owned descriptor list.
385  * @pending_queue: Submitted jobs, to be issued by issue_pending()
386  * @active: Active descriptor.
387  * @done: Completed jobs
388  * @queue: Queued jobs.
389  * @prepare_queue: Prepared jobs.
390  * @dma_cfg: The client configuration of this dma channel.
391  * @configured: whether the dma_cfg configuration is valid
392  * @base: Pointer to the device instance struct.
393  * @src_def_cfg: Default cfg register setting for src.
394  * @dst_def_cfg: Default cfg register setting for dst.
395  * @log_def: Default logical channel settings.
396  * @lcpa: Pointer to dst and src lcpa settings.
397  * @runtime_addr: runtime configured address.
398  * @runtime_direction: runtime configured direction.
399  *
400  * This struct can either "be" a logical or a physical channel.
401  */
402 struct d40_chan {
403         spinlock_t                       lock;
404         int                              log_num;
405         int                              pending_tx;
406         bool                             busy;
407         struct d40_phy_res              *phy_chan;
408         struct dma_chan                  chan;
409         struct tasklet_struct            tasklet;
410         struct list_head                 client;
411         struct list_head                 pending_queue;
412         struct list_head                 active;
413         struct list_head                 done;
414         struct list_head                 queue;
415         struct list_head                 prepare_queue;
416         struct stedma40_chan_cfg         dma_cfg;
417         bool                             configured;
418         struct d40_base                 *base;
419         /* Default register configurations */
420         u32                              src_def_cfg;
421         u32                              dst_def_cfg;
422         struct d40_def_lcsp              log_def;
423         struct d40_log_lli_full         *lcpa;
424         /* Runtime reconfiguration */
425         dma_addr_t                      runtime_addr;
426         enum dma_transfer_direction     runtime_direction;
427 };
428
429 /**
430  * struct d40_gen_dmac - generic values to represent u8500/u8540 DMA
431  * controller
432  *
433  * @backup: the pointer to the registers address array for backup
434  * @backup_size: the size of the registers address array for backup
435  * @realtime_en: the realtime enable register
436  * @realtime_clear: the realtime clear register
437  * @high_prio_en: the high priority enable register
438  * @high_prio_clear: the high priority clear register
439  * @interrupt_en: the interrupt enable register
440  * @interrupt_clear: the interrupt clear register
441  * @il: the pointer to struct d40_interrupt_lookup
442  * @il_size: the size of d40_interrupt_lookup array
443  * @init_reg: the pointer to the struct d40_reg_val
444  * @init_reg_size: the size of d40_reg_val array
445  */
446 struct d40_gen_dmac {
447         u32                             *backup;
448         u32                              backup_size;
449         u32                              realtime_en;
450         u32                              realtime_clear;
451         u32                              high_prio_en;
452         u32                              high_prio_clear;
453         u32                              interrupt_en;
454         u32                              interrupt_clear;
455         struct d40_interrupt_lookup     *il;
456         u32                              il_size;
457         struct d40_reg_val              *init_reg;
458         u32                              init_reg_size;
459 };
460
461 /**
462  * struct d40_base - The big global struct, one for each probe'd instance.
463  *
464  * @interrupt_lock: Lock used to make sure one interrupt is handle a time.
465  * @execmd_lock: Lock for execute command usage since several channels share
466  * the same physical register.
467  * @dev: The device structure.
468  * @virtbase: The virtual base address of the DMA's register.
469  * @rev: silicon revision detected.
470  * @clk: Pointer to the DMA clock structure.
471  * @phy_start: Physical memory start of the DMA registers.
472  * @phy_size: Size of the DMA register map.
473  * @irq: The IRQ number.
474  * @num_phy_chans: The number of physical channels. Read from HW. This
475  * is the number of available channels for this driver, not counting "Secure
476  * mode" allocated physical channels.
477  * @num_log_chans: The number of logical channels. Calculated from
478  * num_phy_chans.
479  * @dma_both: dma_device channels that can do both memcpy and slave transfers.
480  * @dma_slave: dma_device channels that can do only do slave transfers.
481  * @dma_memcpy: dma_device channels that can do only do memcpy transfers.
482  * @phy_chans: Room for all possible physical channels in system.
483  * @log_chans: Room for all possible logical channels in system.
484  * @lookup_log_chans: Used to map interrupt number to logical channel. Points
485  * to log_chans entries.
486  * @lookup_phy_chans: Used to map interrupt number to physical channel. Points
487  * to phy_chans entries.
488  * @plat_data: Pointer to provided platform_data which is the driver
489  * configuration.
490  * @lcpa_regulator: Pointer to hold the regulator for the esram bank for lcla.
491  * @phy_res: Vector containing all physical channels.
492  * @lcla_pool: lcla pool settings and data.
493  * @lcpa_base: The virtual mapped address of LCPA.
494  * @phy_lcpa: The physical address of the LCPA.
495  * @lcpa_size: The size of the LCPA area.
496  * @desc_slab: cache for descriptors.
497  * @reg_val_backup: Here the values of some hardware registers are stored
498  * before the DMA is powered off. They are restored when the power is back on.
499  * @reg_val_backup_v4: Backup of registers that only exits on dma40 v3 and
500  * later
501  * @reg_val_backup_chan: Backup data for standard channel parameter registers.
502  * @gcc_pwr_off_mask: Mask to maintain the channels that can be turned off.
503  * @initialized: true if the dma has been initialized
504  * @gen_dmac: the struct for generic registers values to represent u8500/8540
505  * DMA controller
506  */
507 struct d40_base {
508         spinlock_t                       interrupt_lock;
509         spinlock_t                       execmd_lock;
510         struct device                    *dev;
511         void __iomem                     *virtbase;
512         u8                                rev:4;
513         struct clk                       *clk;
514         phys_addr_t                       phy_start;
515         resource_size_t                   phy_size;
516         int                               irq;
517         int                               num_phy_chans;
518         int                               num_log_chans;
519         struct device_dma_parameters      dma_parms;
520         struct dma_device                 dma_both;
521         struct dma_device                 dma_slave;
522         struct dma_device                 dma_memcpy;
523         struct d40_chan                  *phy_chans;
524         struct d40_chan                  *log_chans;
525         struct d40_chan                 **lookup_log_chans;
526         struct d40_chan                 **lookup_phy_chans;
527         struct stedma40_platform_data    *plat_data;
528         struct regulator                 *lcpa_regulator;
529         /* Physical half channels */
530         struct d40_phy_res               *phy_res;
531         struct d40_lcla_pool              lcla_pool;
532         void                             *lcpa_base;
533         dma_addr_t                        phy_lcpa;
534         resource_size_t                   lcpa_size;
535         struct kmem_cache                *desc_slab;
536         u32                               reg_val_backup[BACKUP_REGS_SZ];
537         u32                               reg_val_backup_v4[MAX(BACKUP_REGS_SZ_V4A, BACKUP_REGS_SZ_V4B)];
538         u32                              *reg_val_backup_chan;
539         u16                               gcc_pwr_off_mask;
540         bool                              initialized;
541         struct d40_gen_dmac               gen_dmac;
542 };
543
544 static struct device *chan2dev(struct d40_chan *d40c)
545 {
546         return &d40c->chan.dev->device;
547 }
548
549 static bool chan_is_physical(struct d40_chan *chan)
550 {
551         return chan->log_num == D40_PHY_CHAN;
552 }
553
554 static bool chan_is_logical(struct d40_chan *chan)
555 {
556         return !chan_is_physical(chan);
557 }
558
559 static void __iomem *chan_base(struct d40_chan *chan)
560 {
561         return chan->base->virtbase + D40_DREG_PCBASE +
562                chan->phy_chan->num * D40_DREG_PCDELTA;
563 }
564
565 #define d40_err(dev, format, arg...)            \
566         dev_err(dev, "[%s] " format, __func__, ## arg)
567
568 #define chan_err(d40c, format, arg...)          \
569         d40_err(chan2dev(d40c), format, ## arg)
570
571 static int d40_pool_lli_alloc(struct d40_chan *d40c, struct d40_desc *d40d,
572                               int lli_len)
573 {
574         bool is_log = chan_is_logical(d40c);
575         u32 align;
576         void *base;
577
578         if (is_log)
579                 align = sizeof(struct d40_log_lli);
580         else
581                 align = sizeof(struct d40_phy_lli);
582
583         if (lli_len == 1) {
584                 base = d40d->lli_pool.pre_alloc_lli;
585                 d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
586                 d40d->lli_pool.base = NULL;
587         } else {
588                 d40d->lli_pool.size = lli_len * 2 * align;
589
590                 base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
591                 d40d->lli_pool.base = base;
592
593                 if (d40d->lli_pool.base == NULL)
594                         return -ENOMEM;
595         }
596
597         if (is_log) {
598                 d40d->lli_log.src = PTR_ALIGN(base, align);
599                 d40d->lli_log.dst = d40d->lli_log.src + lli_len;
600
601                 d40d->lli_pool.dma_addr = 0;
602         } else {
603                 d40d->lli_phy.src = PTR_ALIGN(base, align);
604                 d40d->lli_phy.dst = d40d->lli_phy.src + lli_len;
605
606                 d40d->lli_pool.dma_addr = dma_map_single(d40c->base->dev,
607                                                          d40d->lli_phy.src,
608                                                          d40d->lli_pool.size,
609                                                          DMA_TO_DEVICE);
610
611                 if (dma_mapping_error(d40c->base->dev,
612                                       d40d->lli_pool.dma_addr)) {
613                         kfree(d40d->lli_pool.base);
614                         d40d->lli_pool.base = NULL;
615                         d40d->lli_pool.dma_addr = 0;
616                         return -ENOMEM;
617                 }
618         }
619
620         return 0;
621 }
622
623 static void d40_pool_lli_free(struct d40_chan *d40c, struct d40_desc *d40d)
624 {
625         if (d40d->lli_pool.dma_addr)
626                 dma_unmap_single(d40c->base->dev, d40d->lli_pool.dma_addr,
627                                  d40d->lli_pool.size, DMA_TO_DEVICE);
628
629         kfree(d40d->lli_pool.base);
630         d40d->lli_pool.base = NULL;
631         d40d->lli_pool.size = 0;
632         d40d->lli_log.src = NULL;
633         d40d->lli_log.dst = NULL;
634         d40d->lli_phy.src = NULL;
635         d40d->lli_phy.dst = NULL;
636 }
637
638 static int d40_lcla_alloc_one(struct d40_chan *d40c,
639                               struct d40_desc *d40d)
640 {
641         unsigned long flags;
642         int i;
643         int ret = -EINVAL;
644
645         spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
646
647         /*
648          * Allocate both src and dst at the same time, therefore the half
649          * start on 1 since 0 can't be used since zero is used as end marker.
650          */
651         for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
652                 int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
653
654                 if (!d40c->base->lcla_pool.alloc_map[idx]) {
655                         d40c->base->lcla_pool.alloc_map[idx] = d40d;
656                         d40d->lcla_alloc++;
657                         ret = i;
658                         break;
659                 }
660         }
661
662         spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
663
664         return ret;
665 }
666
667 static int d40_lcla_free_all(struct d40_chan *d40c,
668                              struct d40_desc *d40d)
669 {
670         unsigned long flags;
671         int i;
672         int ret = -EINVAL;
673
674         if (chan_is_physical(d40c))
675                 return 0;
676
677         spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
678
679         for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
680                 int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
681
682                 if (d40c->base->lcla_pool.alloc_map[idx] == d40d) {
683                         d40c->base->lcla_pool.alloc_map[idx] = NULL;
684                         d40d->lcla_alloc--;
685                         if (d40d->lcla_alloc == 0) {
686                                 ret = 0;
687                                 break;
688                         }
689                 }
690         }
691
692         spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
693
694         return ret;
695
696 }
697
698 static void d40_desc_remove(struct d40_desc *d40d)
699 {
700         list_del(&d40d->node);
701 }
702
703 static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
704 {
705         struct d40_desc *desc = NULL;
706
707         if (!list_empty(&d40c->client)) {
708                 struct d40_desc *d;
709                 struct d40_desc *_d;
710
711                 list_for_each_entry_safe(d, _d, &d40c->client, node) {
712                         if (async_tx_test_ack(&d->txd)) {
713                                 d40_desc_remove(d);
714                                 desc = d;
715                                 memset(desc, 0, sizeof(*desc));
716                                 break;
717                         }
718                 }
719         }
720
721         if (!desc)
722                 desc = kmem_cache_zalloc(d40c->base->desc_slab, GFP_NOWAIT);
723
724         if (desc)
725                 INIT_LIST_HEAD(&desc->node);
726
727         return desc;
728 }
729
730 static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
731 {
732
733         d40_pool_lli_free(d40c, d40d);
734         d40_lcla_free_all(d40c, d40d);
735         kmem_cache_free(d40c->base->desc_slab, d40d);
736 }
737
738 static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
739 {
740         list_add_tail(&desc->node, &d40c->active);
741 }
742
743 static void d40_phy_lli_load(struct d40_chan *chan, struct d40_desc *desc)
744 {
745         struct d40_phy_lli *lli_dst = desc->lli_phy.dst;
746         struct d40_phy_lli *lli_src = desc->lli_phy.src;
747         void __iomem *base = chan_base(chan);
748
749         writel(lli_src->reg_cfg, base + D40_CHAN_REG_SSCFG);
750         writel(lli_src->reg_elt, base + D40_CHAN_REG_SSELT);
751         writel(lli_src->reg_ptr, base + D40_CHAN_REG_SSPTR);
752         writel(lli_src->reg_lnk, base + D40_CHAN_REG_SSLNK);
753
754         writel(lli_dst->reg_cfg, base + D40_CHAN_REG_SDCFG);
755         writel(lli_dst->reg_elt, base + D40_CHAN_REG_SDELT);
756         writel(lli_dst->reg_ptr, base + D40_CHAN_REG_SDPTR);
757         writel(lli_dst->reg_lnk, base + D40_CHAN_REG_SDLNK);
758 }
759
760 static void d40_desc_done(struct d40_chan *d40c, struct d40_desc *desc)
761 {
762         list_add_tail(&desc->node, &d40c->done);
763 }
764
765 static void d40_log_lli_to_lcxa(struct d40_chan *chan, struct d40_desc *desc)
766 {
767         struct d40_lcla_pool *pool = &chan->base->lcla_pool;
768         struct d40_log_lli_bidir *lli = &desc->lli_log;
769         int lli_current = desc->lli_current;
770         int lli_len = desc->lli_len;
771         bool cyclic = desc->cyclic;
772         int curr_lcla = -EINVAL;
773         int first_lcla = 0;
774         bool use_esram_lcla = chan->base->plat_data->use_esram_lcla;
775         bool linkback;
776
777         /*
778          * We may have partially running cyclic transfers, in case we did't get
779          * enough LCLA entries.
780          */
781         linkback = cyclic && lli_current == 0;
782
783         /*
784          * For linkback, we need one LCLA even with only one link, because we
785          * can't link back to the one in LCPA space
786          */
787         if (linkback || (lli_len - lli_current > 1)) {
788                 /*
789                  * If the channel is expected to use only soft_lli don't
790                  * allocate a lcla. This is to avoid a HW issue that exists
791                  * in some controller during a peripheral to memory transfer
792                  * that uses linked lists.
793                  */
794                 if (!(chan->phy_chan->use_soft_lli &&
795                         chan->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM))
796                         curr_lcla = d40_lcla_alloc_one(chan, desc);
797
798                 first_lcla = curr_lcla;
799         }
800
801         /*
802          * For linkback, we normally load the LCPA in the loop since we need to
803          * link it to the second LCLA and not the first.  However, if we
804          * couldn't even get a first LCLA, then we have to run in LCPA and
805          * reload manually.
806          */
807         if (!linkback || curr_lcla == -EINVAL) {
808                 unsigned int flags = 0;
809
810                 if (curr_lcla == -EINVAL)
811                         flags |= LLI_TERM_INT;
812
813                 d40_log_lli_lcpa_write(chan->lcpa,
814                                        &lli->dst[lli_current],
815                                        &lli->src[lli_current],
816                                        curr_lcla,
817                                        flags);
818                 lli_current++;
819         }
820
821         if (curr_lcla < 0)
822                 goto out;
823
824         for (; lli_current < lli_len; lli_current++) {
825                 unsigned int lcla_offset = chan->phy_chan->num * 1024 +
826                                            8 * curr_lcla * 2;
827                 struct d40_log_lli *lcla = pool->base + lcla_offset;
828                 unsigned int flags = 0;
829                 int next_lcla;
830
831                 if (lli_current + 1 < lli_len)
832                         next_lcla = d40_lcla_alloc_one(chan, desc);
833                 else
834                         next_lcla = linkback ? first_lcla : -EINVAL;
835
836                 if (cyclic || next_lcla == -EINVAL)
837                         flags |= LLI_TERM_INT;
838
839                 if (linkback && curr_lcla == first_lcla) {
840                         /* First link goes in both LCPA and LCLA */
841                         d40_log_lli_lcpa_write(chan->lcpa,
842                                                &lli->dst[lli_current],
843                                                &lli->src[lli_current],
844                                                next_lcla, flags);
845                 }
846
847                 /*
848                  * One unused LCLA in the cyclic case if the very first
849                  * next_lcla fails...
850                  */
851                 d40_log_lli_lcla_write(lcla,
852                                        &lli->dst[lli_current],
853                                        &lli->src[lli_current],
854                                        next_lcla, flags);
855
856                 /*
857                  * Cache maintenance is not needed if lcla is
858                  * mapped in esram
859                  */
860                 if (!use_esram_lcla) {
861                         dma_sync_single_range_for_device(chan->base->dev,
862                                                 pool->dma_addr, lcla_offset,
863                                                 2 * sizeof(struct d40_log_lli),
864                                                 DMA_TO_DEVICE);
865                 }
866                 curr_lcla = next_lcla;
867
868                 if (curr_lcla == -EINVAL || curr_lcla == first_lcla) {
869                         lli_current++;
870                         break;
871                 }
872         }
873
874 out:
875         desc->lli_current = lli_current;
876 }
877
878 static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
879 {
880         if (chan_is_physical(d40c)) {
881                 d40_phy_lli_load(d40c, d40d);
882                 d40d->lli_current = d40d->lli_len;
883         } else
884                 d40_log_lli_to_lcxa(d40c, d40d);
885 }
886
887 static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
888 {
889         struct d40_desc *d;
890
891         if (list_empty(&d40c->active))
892                 return NULL;
893
894         d = list_first_entry(&d40c->active,
895                              struct d40_desc,
896                              node);
897         return d;
898 }
899
900 /* remove desc from current queue and add it to the pending_queue */
901 static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
902 {
903         d40_desc_remove(desc);
904         desc->is_in_client_list = false;
905         list_add_tail(&desc->node, &d40c->pending_queue);
906 }
907
908 static struct d40_desc *d40_first_pending(struct d40_chan *d40c)
909 {
910         struct d40_desc *d;
911
912         if (list_empty(&d40c->pending_queue))
913                 return NULL;
914
915         d = list_first_entry(&d40c->pending_queue,
916                              struct d40_desc,
917                              node);
918         return d;
919 }
920
921 static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
922 {
923         struct d40_desc *d;
924
925         if (list_empty(&d40c->queue))
926                 return NULL;
927
928         d = list_first_entry(&d40c->queue,
929                              struct d40_desc,
930                              node);
931         return d;
932 }
933
934 static struct d40_desc *d40_first_done(struct d40_chan *d40c)
935 {
936         if (list_empty(&d40c->done))
937                 return NULL;
938
939         return list_first_entry(&d40c->done, struct d40_desc, node);
940 }
941
942 static int d40_psize_2_burst_size(bool is_log, int psize)
943 {
944         if (is_log) {
945                 if (psize == STEDMA40_PSIZE_LOG_1)
946                         return 1;
947         } else {
948                 if (psize == STEDMA40_PSIZE_PHY_1)
949                         return 1;
950         }
951
952         return 2 << psize;
953 }
954
955 /*
956  * The dma only supports transmitting packages up to
957  * STEDMA40_MAX_SEG_SIZE << data_width. Calculate the total number of
958  * dma elements required to send the entire sg list
959  */
960 static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2)
961 {
962         int dmalen;
963         u32 max_w = max(data_width1, data_width2);
964         u32 min_w = min(data_width1, data_width2);
965         u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE << min_w, 1 << max_w);
966
967         if (seg_max > STEDMA40_MAX_SEG_SIZE)
968                 seg_max -= (1 << max_w);
969
970         if (!IS_ALIGNED(size, 1 << max_w))
971                 return -EINVAL;
972
973         if (size <= seg_max)
974                 dmalen = 1;
975         else {
976                 dmalen = size / seg_max;
977                 if (dmalen * seg_max < size)
978                         dmalen++;
979         }
980         return dmalen;
981 }
982
983 static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len,
984                            u32 data_width1, u32 data_width2)
985 {
986         struct scatterlist *sg;
987         int i;
988         int len = 0;
989         int ret;
990
991         for_each_sg(sgl, sg, sg_len, i) {
992                 ret = d40_size_2_dmalen(sg_dma_len(sg),
993                                         data_width1, data_width2);
994                 if (ret < 0)
995                         return ret;
996                 len += ret;
997         }
998         return len;
999 }
1000
1001
1002 #ifdef CONFIG_PM
1003 static void dma40_backup(void __iomem *baseaddr, u32 *backup,
1004                          u32 *regaddr, int num, bool save)
1005 {
1006         int i;
1007
1008         for (i = 0; i < num; i++) {
1009                 void __iomem *addr = baseaddr + regaddr[i];
1010
1011                 if (save)
1012                         backup[i] = readl_relaxed(addr);
1013                 else
1014                         writel_relaxed(backup[i], addr);
1015         }
1016 }
1017
1018 static void d40_save_restore_registers(struct d40_base *base, bool save)
1019 {
1020         int i;
1021
1022         /* Save/Restore channel specific registers */
1023         for (i = 0; i < base->num_phy_chans; i++) {
1024                 void __iomem *addr;
1025                 int idx;
1026
1027                 if (base->phy_res[i].reserved)
1028                         continue;
1029
1030                 addr = base->virtbase + D40_DREG_PCBASE + i * D40_DREG_PCDELTA;
1031                 idx = i * ARRAY_SIZE(d40_backup_regs_chan);
1032
1033                 dma40_backup(addr, &base->reg_val_backup_chan[idx],
1034                              d40_backup_regs_chan,
1035                              ARRAY_SIZE(d40_backup_regs_chan),
1036                              save);
1037         }
1038
1039         /* Save/Restore global registers */
1040         dma40_backup(base->virtbase, base->reg_val_backup,
1041                      d40_backup_regs, ARRAY_SIZE(d40_backup_regs),
1042                      save);
1043
1044         /* Save/Restore registers only existing on dma40 v3 and later */
1045         if (base->gen_dmac.backup)
1046                 dma40_backup(base->virtbase, base->reg_val_backup_v4,
1047                              base->gen_dmac.backup,
1048                         base->gen_dmac.backup_size,
1049                         save);
1050 }
1051 #else
1052 static void d40_save_restore_registers(struct d40_base *base, bool save)
1053 {
1054 }
1055 #endif
1056
1057 static int __d40_execute_command_phy(struct d40_chan *d40c,
1058                                      enum d40_command command)
1059 {
1060         u32 status;
1061         int i;
1062         void __iomem *active_reg;
1063         int ret = 0;
1064         unsigned long flags;
1065         u32 wmask;
1066
1067         if (command == D40_DMA_STOP) {
1068                 ret = __d40_execute_command_phy(d40c, D40_DMA_SUSPEND_REQ);
1069                 if (ret)
1070                         return ret;
1071         }
1072
1073         spin_lock_irqsave(&d40c->base->execmd_lock, flags);
1074
1075         if (d40c->phy_chan->num % 2 == 0)
1076                 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1077         else
1078                 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1079
1080         if (command == D40_DMA_SUSPEND_REQ) {
1081                 status = (readl(active_reg) &
1082                           D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1083                         D40_CHAN_POS(d40c->phy_chan->num);
1084
1085                 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1086                         goto done;
1087         }
1088
1089         wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
1090         writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
1091                active_reg);
1092
1093         if (command == D40_DMA_SUSPEND_REQ) {
1094
1095                 for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
1096                         status = (readl(active_reg) &
1097                                   D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1098                                 D40_CHAN_POS(d40c->phy_chan->num);
1099
1100                         cpu_relax();
1101                         /*
1102                          * Reduce the number of bus accesses while
1103                          * waiting for the DMA to suspend.
1104                          */
1105                         udelay(3);
1106
1107                         if (status == D40_DMA_STOP ||
1108                             status == D40_DMA_SUSPENDED)
1109                                 break;
1110                 }
1111
1112                 if (i == D40_SUSPEND_MAX_IT) {
1113                         chan_err(d40c,
1114                                 "unable to suspend the chl %d (log: %d) status %x\n",
1115                                 d40c->phy_chan->num, d40c->log_num,
1116                                 status);
1117                         dump_stack();
1118                         ret = -EBUSY;
1119                 }
1120
1121         }
1122 done:
1123         spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
1124         return ret;
1125 }
1126
1127 static void d40_term_all(struct d40_chan *d40c)
1128 {
1129         struct d40_desc *d40d;
1130         struct d40_desc *_d;
1131
1132         /* Release completed descriptors */
1133         while ((d40d = d40_first_done(d40c))) {
1134                 d40_desc_remove(d40d);
1135                 d40_desc_free(d40c, d40d);
1136         }
1137
1138         /* Release active descriptors */
1139         while ((d40d = d40_first_active_get(d40c))) {
1140                 d40_desc_remove(d40d);
1141                 d40_desc_free(d40c, d40d);
1142         }
1143
1144         /* Release queued descriptors waiting for transfer */
1145         while ((d40d = d40_first_queued(d40c))) {
1146                 d40_desc_remove(d40d);
1147                 d40_desc_free(d40c, d40d);
1148         }
1149
1150         /* Release pending descriptors */
1151         while ((d40d = d40_first_pending(d40c))) {
1152                 d40_desc_remove(d40d);
1153                 d40_desc_free(d40c, d40d);
1154         }
1155
1156         /* Release client owned descriptors */
1157         if (!list_empty(&d40c->client))
1158                 list_for_each_entry_safe(d40d, _d, &d40c->client, node) {
1159                         d40_desc_remove(d40d);
1160                         d40_desc_free(d40c, d40d);
1161                 }
1162
1163         /* Release descriptors in prepare queue */
1164         if (!list_empty(&d40c->prepare_queue))
1165                 list_for_each_entry_safe(d40d, _d,
1166                                          &d40c->prepare_queue, node) {
1167                         d40_desc_remove(d40d);
1168                         d40_desc_free(d40c, d40d);
1169                 }
1170
1171         d40c->pending_tx = 0;
1172 }
1173
1174 static void __d40_config_set_event(struct d40_chan *d40c,
1175                                    enum d40_events event_type, u32 event,
1176                                    int reg)
1177 {
1178         void __iomem *addr = chan_base(d40c) + reg;
1179         int tries;
1180         u32 status;
1181
1182         switch (event_type) {
1183
1184         case D40_DEACTIVATE_EVENTLINE:
1185
1186                 writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
1187                        | ~D40_EVENTLINE_MASK(event), addr);
1188                 break;
1189
1190         case D40_SUSPEND_REQ_EVENTLINE:
1191                 status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1192                           D40_EVENTLINE_POS(event);
1193
1194                 if (status == D40_DEACTIVATE_EVENTLINE ||
1195                     status == D40_SUSPEND_REQ_EVENTLINE)
1196                         break;
1197
1198                 writel((D40_SUSPEND_REQ_EVENTLINE << D40_EVENTLINE_POS(event))
1199                        | ~D40_EVENTLINE_MASK(event), addr);
1200
1201                 for (tries = 0 ; tries < D40_SUSPEND_MAX_IT; tries++) {
1202
1203                         status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1204                                   D40_EVENTLINE_POS(event);
1205
1206                         cpu_relax();
1207                         /*
1208                          * Reduce the number of bus accesses while
1209                          * waiting for the DMA to suspend.
1210                          */
1211                         udelay(3);
1212
1213                         if (status == D40_DEACTIVATE_EVENTLINE)
1214                                 break;
1215                 }
1216
1217                 if (tries == D40_SUSPEND_MAX_IT) {
1218                         chan_err(d40c,
1219                                 "unable to stop the event_line chl %d (log: %d)"
1220                                 "status %x\n", d40c->phy_chan->num,
1221                                  d40c->log_num, status);
1222                 }
1223                 break;
1224
1225         case D40_ACTIVATE_EVENTLINE:
1226         /*
1227          * The hardware sometimes doesn't register the enable when src and dst
1228          * event lines are active on the same logical channel.  Retry to ensure
1229          * it does.  Usually only one retry is sufficient.
1230          */
1231                 tries = 100;
1232                 while (--tries) {
1233                         writel((D40_ACTIVATE_EVENTLINE <<
1234                                 D40_EVENTLINE_POS(event)) |
1235                                 ~D40_EVENTLINE_MASK(event), addr);
1236
1237                         if (readl(addr) & D40_EVENTLINE_MASK(event))
1238                                 break;
1239                 }
1240
1241                 if (tries != 99)
1242                         dev_dbg(chan2dev(d40c),
1243                                 "[%s] workaround enable S%cLNK (%d tries)\n",
1244                                 __func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D',
1245                                 100 - tries);
1246
1247                 WARN_ON(!tries);
1248                 break;
1249
1250         case D40_ROUND_EVENTLINE:
1251                 BUG();
1252                 break;
1253
1254         }
1255 }
1256
1257 static void d40_config_set_event(struct d40_chan *d40c,
1258                                  enum d40_events event_type)
1259 {
1260         /* Enable event line connected to device (or memcpy) */
1261         if ((d40c->dma_cfg.dir ==  STEDMA40_PERIPH_TO_MEM) ||
1262             (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH)) {
1263                 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
1264
1265                 __d40_config_set_event(d40c, event_type, event,
1266                                        D40_CHAN_REG_SSLNK);
1267         }
1268
1269         if (d40c->dma_cfg.dir !=  STEDMA40_PERIPH_TO_MEM) {
1270                 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
1271
1272                 __d40_config_set_event(d40c, event_type, event,
1273                                        D40_CHAN_REG_SDLNK);
1274         }
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 * (1 << 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 d40_chan *d40c)
1437 {
1438         int res = 0;
1439         unsigned long flags;
1440
1441         if (!d40c->busy)
1442                 return 0;
1443
1444         pm_runtime_get_sync(d40c->base->dev);
1445         spin_lock_irqsave(&d40c->lock, flags);
1446
1447         res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1448
1449         pm_runtime_mark_last_busy(d40c->base->dev);
1450         pm_runtime_put_autosuspend(d40c->base->dev);
1451         spin_unlock_irqrestore(&d40c->lock, flags);
1452         return res;
1453 }
1454
1455 static int d40_resume(struct d40_chan *d40c)
1456 {
1457         int res = 0;
1458         unsigned long flags;
1459
1460         if (!d40c->busy)
1461                 return 0;
1462
1463         spin_lock_irqsave(&d40c->lock, flags);
1464         pm_runtime_get_sync(d40c->base->dev);
1465
1466         /* If bytes left to transfer or linked tx resume job */
1467         if (d40_residue(d40c) || d40_tx_is_linked(d40c))
1468                 res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1469
1470         pm_runtime_mark_last_busy(d40c->base->dev);
1471         pm_runtime_put_autosuspend(d40c->base->dev);
1472         spin_unlock_irqrestore(&d40c->lock, flags);
1473         return res;
1474 }
1475
1476 static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
1477 {
1478         struct d40_chan *d40c = container_of(tx->chan,
1479                                              struct d40_chan,
1480                                              chan);
1481         struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
1482         unsigned long flags;
1483         dma_cookie_t cookie;
1484
1485         spin_lock_irqsave(&d40c->lock, flags);
1486         cookie = dma_cookie_assign(tx);
1487         d40_desc_queue(d40c, d40d);
1488         spin_unlock_irqrestore(&d40c->lock, flags);
1489
1490         return cookie;
1491 }
1492
1493 static int d40_start(struct d40_chan *d40c)
1494 {
1495         return d40_channel_execute_command(d40c, D40_DMA_RUN);
1496 }
1497
1498 static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
1499 {
1500         struct d40_desc *d40d;
1501         int err;
1502
1503         /* Start queued jobs, if any */
1504         d40d = d40_first_queued(d40c);
1505
1506         if (d40d != NULL) {
1507                 if (!d40c->busy) {
1508                         d40c->busy = true;
1509                         pm_runtime_get_sync(d40c->base->dev);
1510                 }
1511
1512                 /* Remove from queue */
1513                 d40_desc_remove(d40d);
1514
1515                 /* Add to active queue */
1516                 d40_desc_submit(d40c, d40d);
1517
1518                 /* Initiate DMA job */
1519                 d40_desc_load(d40c, d40d);
1520
1521                 /* Start dma job */
1522                 err = d40_start(d40c);
1523
1524                 if (err)
1525                         return NULL;
1526         }
1527
1528         return d40d;
1529 }
1530
1531 /* called from interrupt context */
1532 static void dma_tc_handle(struct d40_chan *d40c)
1533 {
1534         struct d40_desc *d40d;
1535
1536         /* Get first active entry from list */
1537         d40d = d40_first_active_get(d40c);
1538
1539         if (d40d == NULL)
1540                 return;
1541
1542         if (d40d->cyclic) {
1543                 /*
1544                  * If this was a paritially loaded list, we need to reloaded
1545                  * it, and only when the list is completed.  We need to check
1546                  * for done because the interrupt will hit for every link, and
1547                  * not just the last one.
1548                  */
1549                 if (d40d->lli_current < d40d->lli_len
1550                     && !d40_tx_is_linked(d40c)
1551                     && !d40_residue(d40c)) {
1552                         d40_lcla_free_all(d40c, d40d);
1553                         d40_desc_load(d40c, d40d);
1554                         (void) d40_start(d40c);
1555
1556                         if (d40d->lli_current == d40d->lli_len)
1557                                 d40d->lli_current = 0;
1558                 }
1559         } else {
1560                 d40_lcla_free_all(d40c, d40d);
1561
1562                 if (d40d->lli_current < d40d->lli_len) {
1563                         d40_desc_load(d40c, d40d);
1564                         /* Start dma job */
1565                         (void) d40_start(d40c);
1566                         return;
1567                 }
1568
1569                 if (d40_queue_start(d40c) == NULL) {
1570                         d40c->busy = false;
1571
1572                         pm_runtime_mark_last_busy(d40c->base->dev);
1573                         pm_runtime_put_autosuspend(d40c->base->dev);
1574                 }
1575
1576                 d40_desc_remove(d40d);
1577                 d40_desc_done(d40c, d40d);
1578         }
1579
1580         d40c->pending_tx++;
1581         tasklet_schedule(&d40c->tasklet);
1582
1583 }
1584
1585 static void dma_tasklet(unsigned long data)
1586 {
1587         struct d40_chan *d40c = (struct d40_chan *) data;
1588         struct d40_desc *d40d;
1589         unsigned long flags;
1590         dma_async_tx_callback callback;
1591         void *callback_param;
1592
1593         spin_lock_irqsave(&d40c->lock, flags);
1594
1595         /* Get first entry from the done list */
1596         d40d = d40_first_done(d40c);
1597         if (d40d == NULL) {
1598                 /* Check if we have reached here for cyclic job */
1599                 d40d = d40_first_active_get(d40c);
1600                 if (d40d == NULL || !d40d->cyclic)
1601                         goto err;
1602         }
1603
1604         if (!d40d->cyclic)
1605                 dma_cookie_complete(&d40d->txd);
1606
1607         /*
1608          * If terminating a channel pending_tx is set to zero.
1609          * This prevents any finished active jobs to return to the client.
1610          */
1611         if (d40c->pending_tx == 0) {
1612                 spin_unlock_irqrestore(&d40c->lock, flags);
1613                 return;
1614         }
1615
1616         /* Callback to client */
1617         callback = d40d->txd.callback;
1618         callback_param = d40d->txd.callback_param;
1619
1620         if (!d40d->cyclic) {
1621                 if (async_tx_test_ack(&d40d->txd)) {
1622                         d40_desc_remove(d40d);
1623                         d40_desc_free(d40c, d40d);
1624                 } else if (!d40d->is_in_client_list) {
1625                         d40_desc_remove(d40d);
1626                         d40_lcla_free_all(d40c, d40d);
1627                         list_add_tail(&d40d->node, &d40c->client);
1628                         d40d->is_in_client_list = true;
1629                 }
1630         }
1631
1632         d40c->pending_tx--;
1633
1634         if (d40c->pending_tx)
1635                 tasklet_schedule(&d40c->tasklet);
1636
1637         spin_unlock_irqrestore(&d40c->lock, flags);
1638
1639         if (callback && (d40d->txd.flags & DMA_PREP_INTERRUPT))
1640                 callback(callback_param);
1641
1642         return;
1643
1644 err:
1645         /* Rescue manouver if receiving double interrupts */
1646         if (d40c->pending_tx > 0)
1647                 d40c->pending_tx--;
1648         spin_unlock_irqrestore(&d40c->lock, flags);
1649 }
1650
1651 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1652 {
1653         int i;
1654         u32 idx;
1655         u32 row;
1656         long chan = -1;
1657         struct d40_chan *d40c;
1658         unsigned long flags;
1659         struct d40_base *base = data;
1660         u32 regs[base->gen_dmac.il_size];
1661         struct d40_interrupt_lookup *il = base->gen_dmac.il;
1662         u32 il_size = base->gen_dmac.il_size;
1663
1664         spin_lock_irqsave(&base->interrupt_lock, flags);
1665
1666         /* Read interrupt status of both logical and physical channels */
1667         for (i = 0; i < il_size; i++)
1668                 regs[i] = readl(base->virtbase + il[i].src);
1669
1670         for (;;) {
1671
1672                 chan = find_next_bit((unsigned long *)regs,
1673                                      BITS_PER_LONG * il_size, chan + 1);
1674
1675                 /* No more set bits found? */
1676                 if (chan == BITS_PER_LONG * il_size)
1677                         break;
1678
1679                 row = chan / BITS_PER_LONG;
1680                 idx = chan & (BITS_PER_LONG - 1);
1681
1682                 if (il[row].offset == D40_PHY_CHAN)
1683                         d40c = base->lookup_phy_chans[idx];
1684                 else
1685                         d40c = base->lookup_log_chans[il[row].offset + idx];
1686
1687                 if (!d40c) {
1688                         /*
1689                          * No error because this can happen if something else
1690                          * in the system is using the channel.
1691                          */
1692                         continue;
1693                 }
1694
1695                 /* ACK interrupt */
1696                 writel(1 << idx, base->virtbase + il[row].clr);
1697
1698                 spin_lock(&d40c->lock);
1699
1700                 if (!il[row].is_error)
1701                         dma_tc_handle(d40c);
1702                 else
1703                         d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1704                                 chan, il[row].offset, idx);
1705
1706                 spin_unlock(&d40c->lock);
1707         }
1708
1709         spin_unlock_irqrestore(&base->interrupt_lock, flags);
1710
1711         return IRQ_HANDLED;
1712 }
1713
1714 static int d40_validate_conf(struct d40_chan *d40c,
1715                              struct stedma40_chan_cfg *conf)
1716 {
1717         int res = 0;
1718         u32 dst_event_group = D40_TYPE_TO_GROUP(conf->dst_dev_type);
1719         u32 src_event_group = D40_TYPE_TO_GROUP(conf->src_dev_type);
1720         bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1721
1722         if (!conf->dir) {
1723                 chan_err(d40c, "Invalid direction.\n");
1724                 res = -EINVAL;
1725         }
1726
1727         if (conf->dst_dev_type != STEDMA40_DEV_DST_MEMORY &&
1728             d40c->base->plat_data->dev_tx[conf->dst_dev_type] == 0 &&
1729             d40c->runtime_addr == 0) {
1730
1731                 chan_err(d40c, "Invalid TX channel address (%d)\n",
1732                          conf->dst_dev_type);
1733                 res = -EINVAL;
1734         }
1735
1736         if (conf->src_dev_type != STEDMA40_DEV_SRC_MEMORY &&
1737             d40c->base->plat_data->dev_rx[conf->src_dev_type] == 0 &&
1738             d40c->runtime_addr == 0) {
1739                 chan_err(d40c, "Invalid RX channel address (%d)\n",
1740                         conf->src_dev_type);
1741                 res = -EINVAL;
1742         }
1743
1744         if (conf->dir == STEDMA40_MEM_TO_PERIPH &&
1745             dst_event_group == STEDMA40_DEV_DST_MEMORY) {
1746                 chan_err(d40c, "Invalid dst\n");
1747                 res = -EINVAL;
1748         }
1749
1750         if (conf->dir == STEDMA40_PERIPH_TO_MEM &&
1751             src_event_group == STEDMA40_DEV_SRC_MEMORY) {
1752                 chan_err(d40c, "Invalid src\n");
1753                 res = -EINVAL;
1754         }
1755
1756         if (src_event_group == STEDMA40_DEV_SRC_MEMORY &&
1757             dst_event_group == STEDMA40_DEV_DST_MEMORY && is_log) {
1758                 chan_err(d40c, "No event line\n");
1759                 res = -EINVAL;
1760         }
1761
1762         if (conf->dir == STEDMA40_PERIPH_TO_PERIPH &&
1763             (src_event_group != dst_event_group)) {
1764                 chan_err(d40c, "Invalid event group\n");
1765                 res = -EINVAL;
1766         }
1767
1768         if (conf->dir == STEDMA40_PERIPH_TO_PERIPH) {
1769                 /*
1770                  * DMAC HW supports it. Will be added to this driver,
1771                  * in case any dma client requires it.
1772                  */
1773                 chan_err(d40c, "periph to periph not supported\n");
1774                 res = -EINVAL;
1775         }
1776
1777         if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1778             (1 << conf->src_info.data_width) !=
1779             d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1780             (1 << conf->dst_info.data_width)) {
1781                 /*
1782                  * The DMAC hardware only supports
1783                  * src (burst x width) == dst (burst x width)
1784                  */
1785
1786                 chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1787                 res = -EINVAL;
1788         }
1789
1790         return res;
1791 }
1792
1793 static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1794                                bool is_src, int log_event_line, bool is_log,
1795                                bool *first_user)
1796 {
1797         unsigned long flags;
1798         spin_lock_irqsave(&phy->lock, flags);
1799
1800         *first_user = ((phy->allocated_src | phy->allocated_dst)
1801                         == D40_ALLOC_FREE);
1802
1803         if (!is_log) {
1804                 /* Physical interrupts are masked per physical full channel */
1805                 if (phy->allocated_src == D40_ALLOC_FREE &&
1806                     phy->allocated_dst == D40_ALLOC_FREE) {
1807                         phy->allocated_dst = D40_ALLOC_PHY;
1808                         phy->allocated_src = D40_ALLOC_PHY;
1809                         goto found;
1810                 } else
1811                         goto not_found;
1812         }
1813
1814         /* Logical channel */
1815         if (is_src) {
1816                 if (phy->allocated_src == D40_ALLOC_PHY)
1817                         goto not_found;
1818
1819                 if (phy->allocated_src == D40_ALLOC_FREE)
1820                         phy->allocated_src = D40_ALLOC_LOG_FREE;
1821
1822                 if (!(phy->allocated_src & (1 << log_event_line))) {
1823                         phy->allocated_src |= 1 << log_event_line;
1824                         goto found;
1825                 } else
1826                         goto not_found;
1827         } else {
1828                 if (phy->allocated_dst == D40_ALLOC_PHY)
1829                         goto not_found;
1830
1831                 if (phy->allocated_dst == D40_ALLOC_FREE)
1832                         phy->allocated_dst = D40_ALLOC_LOG_FREE;
1833
1834                 if (!(phy->allocated_dst & (1 << log_event_line))) {
1835                         phy->allocated_dst |= 1 << log_event_line;
1836                         goto found;
1837                 } else
1838                         goto not_found;
1839         }
1840
1841 not_found:
1842         spin_unlock_irqrestore(&phy->lock, flags);
1843         return false;
1844 found:
1845         spin_unlock_irqrestore(&phy->lock, flags);
1846         return true;
1847 }
1848
1849 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1850                                int log_event_line)
1851 {
1852         unsigned long flags;
1853         bool is_free = false;
1854
1855         spin_lock_irqsave(&phy->lock, flags);
1856         if (!log_event_line) {
1857                 phy->allocated_dst = D40_ALLOC_FREE;
1858                 phy->allocated_src = D40_ALLOC_FREE;
1859                 is_free = true;
1860                 goto out;
1861         }
1862
1863         /* Logical channel */
1864         if (is_src) {
1865                 phy->allocated_src &= ~(1 << log_event_line);
1866                 if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1867                         phy->allocated_src = D40_ALLOC_FREE;
1868         } else {
1869                 phy->allocated_dst &= ~(1 << log_event_line);
1870                 if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1871                         phy->allocated_dst = D40_ALLOC_FREE;
1872         }
1873
1874         is_free = ((phy->allocated_src | phy->allocated_dst) ==
1875                    D40_ALLOC_FREE);
1876
1877 out:
1878         spin_unlock_irqrestore(&phy->lock, flags);
1879
1880         return is_free;
1881 }
1882
1883 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1884 {
1885         int dev_type;
1886         int event_group;
1887         int event_line;
1888         struct d40_phy_res *phys;
1889         int i;
1890         int j;
1891         int log_num;
1892         int num_phy_chans;
1893         bool is_src;
1894         bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1895
1896         phys = d40c->base->phy_res;
1897         num_phy_chans = d40c->base->num_phy_chans;
1898
1899         if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1900                 dev_type = d40c->dma_cfg.src_dev_type;
1901                 log_num = 2 * dev_type;
1902                 is_src = true;
1903         } else if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1904                    d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1905                 /* dst event lines are used for logical memcpy */
1906                 dev_type = d40c->dma_cfg.dst_dev_type;
1907                 log_num = 2 * dev_type + 1;
1908                 is_src = false;
1909         } else
1910                 return -EINVAL;
1911
1912         event_group = D40_TYPE_TO_GROUP(dev_type);
1913         event_line = D40_TYPE_TO_EVENT(dev_type);
1914
1915         if (!is_log) {
1916                 if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1917                         /* Find physical half channel */
1918                         if (d40c->dma_cfg.use_fixed_channel) {
1919                                 i = d40c->dma_cfg.phy_channel;
1920                                 if (d40_alloc_mask_set(&phys[i], is_src,
1921                                                        0, is_log,
1922                                                        first_phy_user))
1923                                         goto found_phy;
1924                         } else {
1925                                 for (i = 0; i < num_phy_chans; i++) {
1926                                         if (d40_alloc_mask_set(&phys[i], is_src,
1927                                                        0, is_log,
1928                                                        first_phy_user))
1929                                                 goto found_phy;
1930                                 }
1931                         }
1932                 } else
1933                         for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1934                                 int phy_num = j  + event_group * 2;
1935                                 for (i = phy_num; i < phy_num + 2; i++) {
1936                                         if (d40_alloc_mask_set(&phys[i],
1937                                                                is_src,
1938                                                                0,
1939                                                                is_log,
1940                                                                first_phy_user))
1941                                                 goto found_phy;
1942                                 }
1943                         }
1944                 return -EINVAL;
1945 found_phy:
1946                 d40c->phy_chan = &phys[i];
1947                 d40c->log_num = D40_PHY_CHAN;
1948                 goto out;
1949         }
1950         if (dev_type == -1)
1951                 return -EINVAL;
1952
1953         /* Find logical channel */
1954         for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1955                 int phy_num = j + event_group * 2;
1956
1957                 if (d40c->dma_cfg.use_fixed_channel) {
1958                         i = d40c->dma_cfg.phy_channel;
1959
1960                         if ((i != phy_num) && (i != phy_num + 1)) {
1961                                 dev_err(chan2dev(d40c),
1962                                         "invalid fixed phy channel %d\n", i);
1963                                 return -EINVAL;
1964                         }
1965
1966                         if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1967                                                is_log, first_phy_user))
1968                                 goto found_log;
1969
1970                         dev_err(chan2dev(d40c),
1971                                 "could not allocate fixed phy channel %d\n", i);
1972                         return -EINVAL;
1973                 }
1974
1975                 /*
1976                  * Spread logical channels across all available physical rather
1977                  * than pack every logical channel at the first available phy
1978                  * channels.
1979                  */
1980                 if (is_src) {
1981                         for (i = phy_num; i < phy_num + 2; i++) {
1982                                 if (d40_alloc_mask_set(&phys[i], is_src,
1983                                                        event_line, is_log,
1984                                                        first_phy_user))
1985                                         goto found_log;
1986                         }
1987                 } else {
1988                         for (i = phy_num + 1; i >= phy_num; i--) {
1989                                 if (d40_alloc_mask_set(&phys[i], is_src,
1990                                                        event_line, is_log,
1991                                                        first_phy_user))
1992                                         goto found_log;
1993                         }
1994                 }
1995         }
1996         return -EINVAL;
1997
1998 found_log:
1999         d40c->phy_chan = &phys[i];
2000         d40c->log_num = log_num;
2001 out:
2002
2003         if (is_log)
2004                 d40c->base->lookup_log_chans[d40c->log_num] = d40c;
2005         else
2006                 d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
2007
2008         return 0;
2009
2010 }
2011
2012 static int d40_config_memcpy(struct d40_chan *d40c)
2013 {
2014         dma_cap_mask_t cap = d40c->chan.device->cap_mask;
2015
2016         if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
2017                 d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_log;
2018                 d40c->dma_cfg.src_dev_type = STEDMA40_DEV_SRC_MEMORY;
2019                 d40c->dma_cfg.dst_dev_type = d40c->base->plat_data->
2020                         memcpy[d40c->chan.chan_id];
2021
2022         } else if (dma_has_cap(DMA_MEMCPY, cap) &&
2023                    dma_has_cap(DMA_SLAVE, cap)) {
2024                 d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_phy;
2025         } else {
2026                 chan_err(d40c, "No memcpy\n");
2027                 return -EINVAL;
2028         }
2029
2030         return 0;
2031 }
2032
2033 static int d40_free_dma(struct d40_chan *d40c)
2034 {
2035
2036         int res = 0;
2037         u32 event;
2038         struct d40_phy_res *phy = d40c->phy_chan;
2039         bool is_src;
2040
2041         /* Terminate all queued and active transfers */
2042         d40_term_all(d40c);
2043
2044         if (phy == NULL) {
2045                 chan_err(d40c, "phy == null\n");
2046                 return -EINVAL;
2047         }
2048
2049         if (phy->allocated_src == D40_ALLOC_FREE &&
2050             phy->allocated_dst == D40_ALLOC_FREE) {
2051                 chan_err(d40c, "channel already free\n");
2052                 return -EINVAL;
2053         }
2054
2055         if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
2056             d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
2057                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
2058                 is_src = false;
2059         } else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
2060                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
2061                 is_src = true;
2062         } else {
2063                 chan_err(d40c, "Unknown direction\n");
2064                 return -EINVAL;
2065         }
2066
2067         pm_runtime_get_sync(d40c->base->dev);
2068         res = d40_channel_execute_command(d40c, D40_DMA_STOP);
2069         if (res) {
2070                 chan_err(d40c, "stop failed\n");
2071                 goto out;
2072         }
2073
2074         d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
2075
2076         if (chan_is_logical(d40c))
2077                 d40c->base->lookup_log_chans[d40c->log_num] = NULL;
2078         else
2079                 d40c->base->lookup_phy_chans[phy->num] = NULL;
2080
2081         if (d40c->busy) {
2082                 pm_runtime_mark_last_busy(d40c->base->dev);
2083                 pm_runtime_put_autosuspend(d40c->base->dev);
2084         }
2085
2086         d40c->busy = false;
2087         d40c->phy_chan = NULL;
2088         d40c->configured = false;
2089 out:
2090
2091         pm_runtime_mark_last_busy(d40c->base->dev);
2092         pm_runtime_put_autosuspend(d40c->base->dev);
2093         return res;
2094 }
2095
2096 static bool d40_is_paused(struct d40_chan *d40c)
2097 {
2098         void __iomem *chanbase = chan_base(d40c);
2099         bool is_paused = false;
2100         unsigned long flags;
2101         void __iomem *active_reg;
2102         u32 status;
2103         u32 event;
2104
2105         spin_lock_irqsave(&d40c->lock, flags);
2106
2107         if (chan_is_physical(d40c)) {
2108                 if (d40c->phy_chan->num % 2 == 0)
2109                         active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
2110                 else
2111                         active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
2112
2113                 status = (readl(active_reg) &
2114                           D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
2115                         D40_CHAN_POS(d40c->phy_chan->num);
2116                 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
2117                         is_paused = true;
2118
2119                 goto _exit;
2120         }
2121
2122         if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
2123             d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
2124                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
2125                 status = readl(chanbase + D40_CHAN_REG_SDLNK);
2126         } else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
2127                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
2128                 status = readl(chanbase + D40_CHAN_REG_SSLNK);
2129         } else {
2130                 chan_err(d40c, "Unknown direction\n");
2131                 goto _exit;
2132         }
2133
2134         status = (status & D40_EVENTLINE_MASK(event)) >>
2135                 D40_EVENTLINE_POS(event);
2136
2137         if (status != D40_DMA_RUN)
2138                 is_paused = true;
2139 _exit:
2140         spin_unlock_irqrestore(&d40c->lock, flags);
2141         return is_paused;
2142
2143 }
2144
2145 static u32 stedma40_residue(struct dma_chan *chan)
2146 {
2147         struct d40_chan *d40c =
2148                 container_of(chan, struct d40_chan, chan);
2149         u32 bytes_left;
2150         unsigned long flags;
2151
2152         spin_lock_irqsave(&d40c->lock, flags);
2153         bytes_left = d40_residue(d40c);
2154         spin_unlock_irqrestore(&d40c->lock, flags);
2155
2156         return bytes_left;
2157 }
2158
2159 static int
2160 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
2161                 struct scatterlist *sg_src, struct scatterlist *sg_dst,
2162                 unsigned int sg_len, dma_addr_t src_dev_addr,
2163                 dma_addr_t dst_dev_addr)
2164 {
2165         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2166         struct stedma40_half_channel_info *src_info = &cfg->src_info;
2167         struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2168         int ret;
2169
2170         ret = d40_log_sg_to_lli(sg_src, sg_len,
2171                                 src_dev_addr,
2172                                 desc->lli_log.src,
2173                                 chan->log_def.lcsp1,
2174                                 src_info->data_width,
2175                                 dst_info->data_width);
2176
2177         ret = d40_log_sg_to_lli(sg_dst, sg_len,
2178                                 dst_dev_addr,
2179                                 desc->lli_log.dst,
2180                                 chan->log_def.lcsp3,
2181                                 dst_info->data_width,
2182                                 src_info->data_width);
2183
2184         return ret < 0 ? ret : 0;
2185 }
2186
2187 static int
2188 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2189                 struct scatterlist *sg_src, struct scatterlist *sg_dst,
2190                 unsigned int sg_len, dma_addr_t src_dev_addr,
2191                 dma_addr_t dst_dev_addr)
2192 {
2193         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2194         struct stedma40_half_channel_info *src_info = &cfg->src_info;
2195         struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2196         unsigned long flags = 0;
2197         int ret;
2198
2199         if (desc->cyclic)
2200                 flags |= LLI_CYCLIC | LLI_TERM_INT;
2201
2202         ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2203                                 desc->lli_phy.src,
2204                                 virt_to_phys(desc->lli_phy.src),
2205                                 chan->src_def_cfg,
2206                                 src_info, dst_info, flags);
2207
2208         ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2209                                 desc->lli_phy.dst,
2210                                 virt_to_phys(desc->lli_phy.dst),
2211                                 chan->dst_def_cfg,
2212                                 dst_info, src_info, flags);
2213
2214         dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2215                                    desc->lli_pool.size, DMA_TO_DEVICE);
2216
2217         return ret < 0 ? ret : 0;
2218 }
2219
2220 static struct d40_desc *
2221 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2222               unsigned int sg_len, unsigned long dma_flags)
2223 {
2224         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2225         struct d40_desc *desc;
2226         int ret;
2227
2228         desc = d40_desc_get(chan);
2229         if (!desc)
2230                 return NULL;
2231
2232         desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2233                                         cfg->dst_info.data_width);
2234         if (desc->lli_len < 0) {
2235                 chan_err(chan, "Unaligned size\n");
2236                 goto err;
2237         }
2238
2239         ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2240         if (ret < 0) {
2241                 chan_err(chan, "Could not allocate lli\n");
2242                 goto err;
2243         }
2244
2245         desc->lli_current = 0;
2246         desc->txd.flags = dma_flags;
2247         desc->txd.tx_submit = d40_tx_submit;
2248
2249         dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2250
2251         return desc;
2252
2253 err:
2254         d40_desc_free(chan, desc);
2255         return NULL;
2256 }
2257
2258 static dma_addr_t
2259 d40_get_dev_addr(struct d40_chan *chan, enum dma_transfer_direction direction)
2260 {
2261         struct stedma40_platform_data *plat = chan->base->plat_data;
2262         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2263         dma_addr_t addr = 0;
2264
2265         if (chan->runtime_addr)
2266                 return chan->runtime_addr;
2267
2268         if (direction == DMA_DEV_TO_MEM)
2269                 addr = plat->dev_rx[cfg->src_dev_type];
2270         else if (direction == DMA_MEM_TO_DEV)
2271                 addr = plat->dev_tx[cfg->dst_dev_type];
2272
2273         return addr;
2274 }
2275
2276 static struct dma_async_tx_descriptor *
2277 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2278             struct scatterlist *sg_dst, unsigned int sg_len,
2279             enum dma_transfer_direction direction, unsigned long dma_flags)
2280 {
2281         struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2282         dma_addr_t src_dev_addr = 0;
2283         dma_addr_t dst_dev_addr = 0;
2284         struct d40_desc *desc;
2285         unsigned long flags;
2286         int ret;
2287
2288         if (!chan->phy_chan) {
2289                 chan_err(chan, "Cannot prepare unallocated channel\n");
2290                 return NULL;
2291         }
2292
2293         spin_lock_irqsave(&chan->lock, flags);
2294
2295         desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2296         if (desc == NULL)
2297                 goto err;
2298
2299         if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2300                 desc->cyclic = true;
2301
2302         if (direction != DMA_TRANS_NONE) {
2303                 dma_addr_t dev_addr = d40_get_dev_addr(chan, direction);
2304
2305                 if (direction == DMA_DEV_TO_MEM)
2306                         src_dev_addr = dev_addr;
2307                 else if (direction == DMA_MEM_TO_DEV)
2308                         dst_dev_addr = dev_addr;
2309         }
2310
2311         if (chan_is_logical(chan))
2312                 ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2313                                       sg_len, src_dev_addr, dst_dev_addr);
2314         else
2315                 ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2316                                       sg_len, src_dev_addr, dst_dev_addr);
2317
2318         if (ret) {
2319                 chan_err(chan, "Failed to prepare %s sg job: %d\n",
2320                          chan_is_logical(chan) ? "log" : "phy", ret);
2321                 goto err;
2322         }
2323
2324         /*
2325          * add descriptor to the prepare queue in order to be able
2326          * to free them later in terminate_all
2327          */
2328         list_add_tail(&desc->node, &chan->prepare_queue);
2329
2330         spin_unlock_irqrestore(&chan->lock, flags);
2331
2332         return &desc->txd;
2333
2334 err:
2335         if (desc)
2336                 d40_desc_free(chan, desc);
2337         spin_unlock_irqrestore(&chan->lock, flags);
2338         return NULL;
2339 }
2340
2341 bool stedma40_filter(struct dma_chan *chan, void *data)
2342 {
2343         struct stedma40_chan_cfg *info = data;
2344         struct d40_chan *d40c =
2345                 container_of(chan, struct d40_chan, chan);
2346         int err;
2347
2348         if (data) {
2349                 err = d40_validate_conf(d40c, info);
2350                 if (!err)
2351                         d40c->dma_cfg = *info;
2352         } else
2353                 err = d40_config_memcpy(d40c);
2354
2355         if (!err)
2356                 d40c->configured = true;
2357
2358         return err == 0;
2359 }
2360 EXPORT_SYMBOL(stedma40_filter);
2361
2362 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2363 {
2364         bool realtime = d40c->dma_cfg.realtime;
2365         bool highprio = d40c->dma_cfg.high_priority;
2366         u32 rtreg;
2367         u32 event = D40_TYPE_TO_EVENT(dev_type);
2368         u32 group = D40_TYPE_TO_GROUP(dev_type);
2369         u32 bit = 1 << event;
2370         u32 prioreg;
2371         struct d40_gen_dmac *dmac = &d40c->base->gen_dmac;
2372
2373         rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear;
2374         /*
2375          * Due to a hardware bug, in some cases a logical channel triggered by
2376          * a high priority destination event line can generate extra packet
2377          * transactions.
2378          *
2379          * The workaround is to not set the high priority level for the
2380          * destination event lines that trigger logical channels.
2381          */
2382         if (!src && chan_is_logical(d40c))
2383                 highprio = false;
2384
2385         prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear;
2386
2387         /* Destination event lines are stored in the upper halfword */
2388         if (!src)
2389                 bit <<= 16;
2390
2391         writel(bit, d40c->base->virtbase + prioreg + group * 4);
2392         writel(bit, d40c->base->virtbase + rtreg + group * 4);
2393 }
2394
2395 static void d40_set_prio_realtime(struct d40_chan *d40c)
2396 {
2397         if (d40c->base->rev < 3)
2398                 return;
2399
2400         if ((d40c->dma_cfg.dir ==  STEDMA40_PERIPH_TO_MEM) ||
2401             (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH))
2402                 __d40_set_prio_rt(d40c, d40c->dma_cfg.src_dev_type, true);
2403
2404         if ((d40c->dma_cfg.dir ==  STEDMA40_MEM_TO_PERIPH) ||
2405             (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH))
2406                 __d40_set_prio_rt(d40c, d40c->dma_cfg.dst_dev_type, false);
2407 }
2408
2409 /* DMA ENGINE functions */
2410 static int d40_alloc_chan_resources(struct dma_chan *chan)
2411 {
2412         int err;
2413         unsigned long flags;
2414         struct d40_chan *d40c =
2415                 container_of(chan, struct d40_chan, chan);
2416         bool is_free_phy;
2417         spin_lock_irqsave(&d40c->lock, flags);
2418
2419         dma_cookie_init(chan);
2420
2421         /* If no dma configuration is set use default configuration (memcpy) */
2422         if (!d40c->configured) {
2423                 err = d40_config_memcpy(d40c);
2424                 if (err) {
2425                         chan_err(d40c, "Failed to configure memcpy channel\n");
2426                         goto fail;
2427                 }
2428         }
2429
2430         err = d40_allocate_channel(d40c, &is_free_phy);
2431         if (err) {
2432                 chan_err(d40c, "Failed to allocate channel\n");
2433                 d40c->configured = false;
2434                 goto fail;
2435         }
2436
2437         pm_runtime_get_sync(d40c->base->dev);
2438         /* Fill in basic CFG register values */
2439         d40_phy_cfg(&d40c->dma_cfg, &d40c->src_def_cfg,
2440                     &d40c->dst_def_cfg, chan_is_logical(d40c));
2441
2442         d40_set_prio_realtime(d40c);
2443
2444         if (chan_is_logical(d40c)) {
2445                 d40_log_cfg(&d40c->dma_cfg,
2446                             &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2447
2448                 if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM)
2449                         d40c->lcpa = d40c->base->lcpa_base +
2450                                 d40c->dma_cfg.src_dev_type * D40_LCPA_CHAN_SIZE;
2451                 else
2452                         d40c->lcpa = d40c->base->lcpa_base +
2453                                 d40c->dma_cfg.dst_dev_type *
2454                                 D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2455         }
2456
2457         dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2458                  chan_is_logical(d40c) ? "logical" : "physical",
2459                  d40c->phy_chan->num,
2460                  d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2461
2462
2463         /*
2464          * Only write channel configuration to the DMA if the physical
2465          * resource is free. In case of multiple logical channels
2466          * on the same physical resource, only the first write is necessary.
2467          */
2468         if (is_free_phy)
2469                 d40_config_write(d40c);
2470 fail:
2471         pm_runtime_mark_last_busy(d40c->base->dev);
2472         pm_runtime_put_autosuspend(d40c->base->dev);
2473         spin_unlock_irqrestore(&d40c->lock, flags);
2474         return err;
2475 }
2476
2477 static void d40_free_chan_resources(struct dma_chan *chan)
2478 {
2479         struct d40_chan *d40c =
2480                 container_of(chan, struct d40_chan, chan);
2481         int err;
2482         unsigned long flags;
2483
2484         if (d40c->phy_chan == NULL) {
2485                 chan_err(d40c, "Cannot free unallocated channel\n");
2486                 return;
2487         }
2488
2489         spin_lock_irqsave(&d40c->lock, flags);
2490
2491         err = d40_free_dma(d40c);
2492
2493         if (err)
2494                 chan_err(d40c, "Failed to free channel\n");
2495         spin_unlock_irqrestore(&d40c->lock, flags);
2496 }
2497
2498 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2499                                                        dma_addr_t dst,
2500                                                        dma_addr_t src,
2501                                                        size_t size,
2502                                                        unsigned long dma_flags)
2503 {
2504         struct scatterlist dst_sg;
2505         struct scatterlist src_sg;
2506
2507         sg_init_table(&dst_sg, 1);
2508         sg_init_table(&src_sg, 1);
2509
2510         sg_dma_address(&dst_sg) = dst;
2511         sg_dma_address(&src_sg) = src;
2512
2513         sg_dma_len(&dst_sg) = size;
2514         sg_dma_len(&src_sg) = size;
2515
2516         return d40_prep_sg(chan, &src_sg, &dst_sg, 1, DMA_NONE, dma_flags);
2517 }
2518
2519 static struct dma_async_tx_descriptor *
2520 d40_prep_memcpy_sg(struct dma_chan *chan,
2521                    struct scatterlist *dst_sg, unsigned int dst_nents,
2522                    struct scatterlist *src_sg, unsigned int src_nents,
2523                    unsigned long dma_flags)
2524 {
2525         if (dst_nents != src_nents)
2526                 return NULL;
2527
2528         return d40_prep_sg(chan, src_sg, dst_sg, src_nents, DMA_NONE, dma_flags);
2529 }
2530
2531 static struct dma_async_tx_descriptor *
2532 d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2533                   unsigned int sg_len, enum dma_transfer_direction direction,
2534                   unsigned long dma_flags, void *context)
2535 {
2536         if (!is_slave_direction(direction))
2537                 return NULL;
2538
2539         return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2540 }
2541
2542 static struct dma_async_tx_descriptor *
2543 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2544                      size_t buf_len, size_t period_len,
2545                      enum dma_transfer_direction direction, unsigned long flags,
2546                      void *context)
2547 {
2548         unsigned int periods = buf_len / period_len;
2549         struct dma_async_tx_descriptor *txd;
2550         struct scatterlist *sg;
2551         int i;
2552
2553         sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2554         for (i = 0; i < periods; i++) {
2555                 sg_dma_address(&sg[i]) = dma_addr;
2556                 sg_dma_len(&sg[i]) = period_len;
2557                 dma_addr += period_len;
2558         }
2559
2560         sg[periods].offset = 0;
2561         sg_dma_len(&sg[periods]) = 0;
2562         sg[periods].page_link =
2563                 ((unsigned long)sg | 0x01) & ~0x02;
2564
2565         txd = d40_prep_sg(chan, sg, sg, periods, direction,
2566                           DMA_PREP_INTERRUPT);
2567
2568         kfree(sg);
2569
2570         return txd;
2571 }
2572
2573 static enum dma_status d40_tx_status(struct dma_chan *chan,
2574                                      dma_cookie_t cookie,
2575                                      struct dma_tx_state *txstate)
2576 {
2577         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2578         enum dma_status ret;
2579
2580         if (d40c->phy_chan == NULL) {
2581                 chan_err(d40c, "Cannot read status of unallocated channel\n");
2582                 return -EINVAL;
2583         }
2584
2585         ret = dma_cookie_status(chan, cookie, txstate);
2586         if (ret != DMA_SUCCESS)
2587                 dma_set_residue(txstate, stedma40_residue(chan));
2588
2589         if (d40_is_paused(d40c))
2590                 ret = DMA_PAUSED;
2591
2592         return ret;
2593 }
2594
2595 static void d40_issue_pending(struct dma_chan *chan)
2596 {
2597         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2598         unsigned long flags;
2599
2600         if (d40c->phy_chan == NULL) {
2601                 chan_err(d40c, "Channel is not allocated!\n");
2602                 return;
2603         }
2604
2605         spin_lock_irqsave(&d40c->lock, flags);
2606
2607         list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2608
2609         /* Busy means that queued jobs are already being processed */
2610         if (!d40c->busy)
2611                 (void) d40_queue_start(d40c);
2612
2613         spin_unlock_irqrestore(&d40c->lock, flags);
2614 }
2615
2616 static void d40_terminate_all(struct dma_chan *chan)
2617 {
2618         unsigned long flags;
2619         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2620         int ret;
2621
2622         spin_lock_irqsave(&d40c->lock, flags);
2623
2624         pm_runtime_get_sync(d40c->base->dev);
2625         ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2626         if (ret)
2627                 chan_err(d40c, "Failed to stop channel\n");
2628
2629         d40_term_all(d40c);
2630         pm_runtime_mark_last_busy(d40c->base->dev);
2631         pm_runtime_put_autosuspend(d40c->base->dev);
2632         if (d40c->busy) {
2633                 pm_runtime_mark_last_busy(d40c->base->dev);
2634                 pm_runtime_put_autosuspend(d40c->base->dev);
2635         }
2636         d40c->busy = false;
2637
2638         spin_unlock_irqrestore(&d40c->lock, flags);
2639 }
2640
2641 static int
2642 dma40_config_to_halfchannel(struct d40_chan *d40c,
2643                             struct stedma40_half_channel_info *info,
2644                             enum dma_slave_buswidth width,
2645                             u32 maxburst)
2646 {
2647         enum stedma40_periph_data_width addr_width;
2648         int psize;
2649
2650         switch (width) {
2651         case DMA_SLAVE_BUSWIDTH_1_BYTE:
2652                 addr_width = STEDMA40_BYTE_WIDTH;
2653                 break;
2654         case DMA_SLAVE_BUSWIDTH_2_BYTES:
2655                 addr_width = STEDMA40_HALFWORD_WIDTH;
2656                 break;
2657         case DMA_SLAVE_BUSWIDTH_4_BYTES:
2658                 addr_width = STEDMA40_WORD_WIDTH;
2659                 break;
2660         case DMA_SLAVE_BUSWIDTH_8_BYTES:
2661                 addr_width = STEDMA40_DOUBLEWORD_WIDTH;
2662                 break;
2663         default:
2664                 dev_err(d40c->base->dev,
2665                         "illegal peripheral address width "
2666                         "requested (%d)\n",
2667                         width);
2668                 return -EINVAL;
2669         }
2670
2671         if (chan_is_logical(d40c)) {
2672                 if (maxburst >= 16)
2673                         psize = STEDMA40_PSIZE_LOG_16;
2674                 else if (maxburst >= 8)
2675                         psize = STEDMA40_PSIZE_LOG_8;
2676                 else if (maxburst >= 4)
2677                         psize = STEDMA40_PSIZE_LOG_4;
2678                 else
2679                         psize = STEDMA40_PSIZE_LOG_1;
2680         } else {
2681                 if (maxburst >= 16)
2682                         psize = STEDMA40_PSIZE_PHY_16;
2683                 else if (maxburst >= 8)
2684                         psize = STEDMA40_PSIZE_PHY_8;
2685                 else if (maxburst >= 4)
2686                         psize = STEDMA40_PSIZE_PHY_4;
2687                 else
2688                         psize = STEDMA40_PSIZE_PHY_1;
2689         }
2690
2691         info->data_width = addr_width;
2692         info->psize = psize;
2693         info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2694
2695         return 0;
2696 }
2697
2698 /* Runtime reconfiguration extension */
2699 static int d40_set_runtime_config(struct dma_chan *chan,
2700                                   struct dma_slave_config *config)
2701 {
2702         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2703         struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2704         enum dma_slave_buswidth src_addr_width, dst_addr_width;
2705         dma_addr_t config_addr;
2706         u32 src_maxburst, dst_maxburst;
2707         int ret;
2708
2709         src_addr_width = config->src_addr_width;
2710         src_maxburst = config->src_maxburst;
2711         dst_addr_width = config->dst_addr_width;
2712         dst_maxburst = config->dst_maxburst;
2713
2714         if (config->direction == DMA_DEV_TO_MEM) {
2715                 dma_addr_t dev_addr_rx =
2716                         d40c->base->plat_data->dev_rx[cfg->src_dev_type];
2717
2718                 config_addr = config->src_addr;
2719                 if (dev_addr_rx)
2720                         dev_dbg(d40c->base->dev,
2721                                 "channel has a pre-wired RX address %08x "
2722                                 "overriding with %08x\n",
2723                                 dev_addr_rx, config_addr);
2724                 if (cfg->dir != STEDMA40_PERIPH_TO_MEM)
2725                         dev_dbg(d40c->base->dev,
2726                                 "channel was not configured for peripheral "
2727                                 "to memory transfer (%d) overriding\n",
2728                                 cfg->dir);
2729                 cfg->dir = STEDMA40_PERIPH_TO_MEM;
2730
2731                 /* Configure the memory side */
2732                 if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2733                         dst_addr_width = src_addr_width;
2734                 if (dst_maxburst == 0)
2735                         dst_maxburst = src_maxburst;
2736
2737         } else if (config->direction == DMA_MEM_TO_DEV) {
2738                 dma_addr_t dev_addr_tx =
2739                         d40c->base->plat_data->dev_tx[cfg->dst_dev_type];
2740
2741                 config_addr = config->dst_addr;
2742                 if (dev_addr_tx)
2743                         dev_dbg(d40c->base->dev,
2744                                 "channel has a pre-wired TX address %08x "
2745                                 "overriding with %08x\n",
2746                                 dev_addr_tx, config_addr);
2747                 if (cfg->dir != STEDMA40_MEM_TO_PERIPH)
2748                         dev_dbg(d40c->base->dev,
2749                                 "channel was not configured for memory "
2750                                 "to peripheral transfer (%d) overriding\n",
2751                                 cfg->dir);
2752                 cfg->dir = STEDMA40_MEM_TO_PERIPH;
2753
2754                 /* Configure the memory side */
2755                 if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2756                         src_addr_width = dst_addr_width;
2757                 if (src_maxburst == 0)
2758                         src_maxburst = dst_maxburst;
2759         } else {
2760                 dev_err(d40c->base->dev,
2761                         "unrecognized channel direction %d\n",
2762                         config->direction);
2763                 return -EINVAL;
2764         }
2765
2766         if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2767                 dev_err(d40c->base->dev,
2768                         "src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2769                         src_maxburst,
2770                         src_addr_width,
2771                         dst_maxburst,
2772                         dst_addr_width);
2773                 return -EINVAL;
2774         }
2775
2776         if (src_maxburst > 16) {
2777                 src_maxburst = 16;
2778                 dst_maxburst = src_maxburst * src_addr_width / dst_addr_width;
2779         } else if (dst_maxburst > 16) {
2780                 dst_maxburst = 16;
2781                 src_maxburst = dst_maxburst * dst_addr_width / src_addr_width;
2782         }
2783
2784         ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2785                                           src_addr_width,
2786                                           src_maxburst);
2787         if (ret)
2788                 return ret;
2789
2790         ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2791                                           dst_addr_width,
2792                                           dst_maxburst);
2793         if (ret)
2794                 return ret;
2795
2796         /* Fill in register values */
2797         if (chan_is_logical(d40c))
2798                 d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2799         else
2800                 d40_phy_cfg(cfg, &d40c->src_def_cfg,
2801                             &d40c->dst_def_cfg, false);
2802
2803         /* These settings will take precedence later */
2804         d40c->runtime_addr = config_addr;
2805         d40c->runtime_direction = config->direction;
2806         dev_dbg(d40c->base->dev,
2807                 "configured channel %s for %s, data width %d/%d, "
2808                 "maxburst %d/%d elements, LE, no flow control\n",
2809                 dma_chan_name(chan),
2810                 (config->direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2811                 src_addr_width, dst_addr_width,
2812                 src_maxburst, dst_maxburst);
2813
2814         return 0;
2815 }
2816
2817 static int d40_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
2818                        unsigned long arg)
2819 {
2820         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2821
2822         if (d40c->phy_chan == NULL) {
2823                 chan_err(d40c, "Channel is not allocated!\n");
2824                 return -EINVAL;
2825         }
2826
2827         switch (cmd) {
2828         case DMA_TERMINATE_ALL:
2829                 d40_terminate_all(chan);
2830                 return 0;
2831         case DMA_PAUSE:
2832                 return d40_pause(d40c);
2833         case DMA_RESUME:
2834                 return d40_resume(d40c);
2835         case DMA_SLAVE_CONFIG:
2836                 return d40_set_runtime_config(chan,
2837                         (struct dma_slave_config *) arg);
2838         default:
2839                 break;
2840         }
2841
2842         /* Other commands are unimplemented */
2843         return -ENXIO;
2844 }
2845
2846 /* Initialization functions */
2847
2848 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2849                                  struct d40_chan *chans, int offset,
2850                                  int num_chans)
2851 {
2852         int i = 0;
2853         struct d40_chan *d40c;
2854
2855         INIT_LIST_HEAD(&dma->channels);
2856
2857         for (i = offset; i < offset + num_chans; i++) {
2858                 d40c = &chans[i];
2859                 d40c->base = base;
2860                 d40c->chan.device = dma;
2861
2862                 spin_lock_init(&d40c->lock);
2863
2864                 d40c->log_num = D40_PHY_CHAN;
2865
2866                 INIT_LIST_HEAD(&d40c->done);
2867                 INIT_LIST_HEAD(&d40c->active);
2868                 INIT_LIST_HEAD(&d40c->queue);
2869                 INIT_LIST_HEAD(&d40c->pending_queue);
2870                 INIT_LIST_HEAD(&d40c->client);
2871                 INIT_LIST_HEAD(&d40c->prepare_queue);
2872
2873                 tasklet_init(&d40c->tasklet, dma_tasklet,
2874                              (unsigned long) d40c);
2875
2876                 list_add_tail(&d40c->chan.device_node,
2877                               &dma->channels);
2878         }
2879 }
2880
2881 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2882 {
2883         if (dma_has_cap(DMA_SLAVE, dev->cap_mask))
2884                 dev->device_prep_slave_sg = d40_prep_slave_sg;
2885
2886         if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2887                 dev->device_prep_dma_memcpy = d40_prep_memcpy;
2888
2889                 /*
2890                  * This controller can only access address at even
2891                  * 32bit boundaries, i.e. 2^2
2892                  */
2893                 dev->copy_align = 2;
2894         }
2895
2896         if (dma_has_cap(DMA_SG, dev->cap_mask))
2897                 dev->device_prep_dma_sg = d40_prep_memcpy_sg;
2898
2899         if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2900                 dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2901
2902         dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2903         dev->device_free_chan_resources = d40_free_chan_resources;
2904         dev->device_issue_pending = d40_issue_pending;
2905         dev->device_tx_status = d40_tx_status;
2906         dev->device_control = d40_control;
2907         dev->dev = base->dev;
2908 }
2909
2910 static int __init d40_dmaengine_init(struct d40_base *base,
2911                                      int num_reserved_chans)
2912 {
2913         int err ;
2914
2915         d40_chan_init(base, &base->dma_slave, base->log_chans,
2916                       0, base->num_log_chans);
2917
2918         dma_cap_zero(base->dma_slave.cap_mask);
2919         dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2920         dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2921
2922         d40_ops_init(base, &base->dma_slave);
2923
2924         err = dma_async_device_register(&base->dma_slave);
2925
2926         if (err) {
2927                 d40_err(base->dev, "Failed to register slave channels\n");
2928                 goto failure1;
2929         }
2930
2931         d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2932                       base->num_log_chans, base->plat_data->memcpy_len);
2933
2934         dma_cap_zero(base->dma_memcpy.cap_mask);
2935         dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2936         dma_cap_set(DMA_SG, base->dma_memcpy.cap_mask);
2937
2938         d40_ops_init(base, &base->dma_memcpy);
2939
2940         err = dma_async_device_register(&base->dma_memcpy);
2941
2942         if (err) {
2943                 d40_err(base->dev,
2944                         "Failed to regsiter memcpy only channels\n");
2945                 goto failure2;
2946         }
2947
2948         d40_chan_init(base, &base->dma_both, base->phy_chans,
2949                       0, num_reserved_chans);
2950
2951         dma_cap_zero(base->dma_both.cap_mask);
2952         dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2953         dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2954         dma_cap_set(DMA_SG, base->dma_both.cap_mask);
2955         dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2956
2957         d40_ops_init(base, &base->dma_both);
2958         err = dma_async_device_register(&base->dma_both);
2959
2960         if (err) {
2961                 d40_err(base->dev,
2962                         "Failed to register logical and physical capable channels\n");
2963                 goto failure3;
2964         }
2965         return 0;
2966 failure3:
2967         dma_async_device_unregister(&base->dma_memcpy);
2968 failure2:
2969         dma_async_device_unregister(&base->dma_slave);
2970 failure1:
2971         return err;
2972 }
2973
2974 /* Suspend resume functionality */
2975 #ifdef CONFIG_PM
2976 static int dma40_pm_suspend(struct device *dev)
2977 {
2978         struct platform_device *pdev = to_platform_device(dev);
2979         struct d40_base *base = platform_get_drvdata(pdev);
2980         int ret = 0;
2981
2982         if (base->lcpa_regulator)
2983                 ret = regulator_disable(base->lcpa_regulator);
2984         return ret;
2985 }
2986
2987 static int dma40_runtime_suspend(struct device *dev)
2988 {
2989         struct platform_device *pdev = to_platform_device(dev);
2990         struct d40_base *base = platform_get_drvdata(pdev);
2991
2992         d40_save_restore_registers(base, true);
2993
2994         /* Don't disable/enable clocks for v1 due to HW bugs */
2995         if (base->rev != 1)
2996                 writel_relaxed(base->gcc_pwr_off_mask,
2997                                base->virtbase + D40_DREG_GCC);
2998
2999         return 0;
3000 }
3001
3002 static int dma40_runtime_resume(struct device *dev)
3003 {
3004         struct platform_device *pdev = to_platform_device(dev);
3005         struct d40_base *base = platform_get_drvdata(pdev);
3006
3007         if (base->initialized)
3008                 d40_save_restore_registers(base, false);
3009
3010         writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
3011                        base->virtbase + D40_DREG_GCC);
3012         return 0;
3013 }
3014
3015 static int dma40_resume(struct device *dev)
3016 {
3017         struct platform_device *pdev = to_platform_device(dev);
3018         struct d40_base *base = platform_get_drvdata(pdev);
3019         int ret = 0;
3020
3021         if (base->lcpa_regulator)
3022                 ret = regulator_enable(base->lcpa_regulator);
3023
3024         return ret;
3025 }
3026
3027 static const struct dev_pm_ops dma40_pm_ops = {
3028         .suspend                = dma40_pm_suspend,
3029         .runtime_suspend        = dma40_runtime_suspend,
3030         .runtime_resume         = dma40_runtime_resume,
3031         .resume                 = dma40_resume,
3032 };
3033 #define DMA40_PM_OPS    (&dma40_pm_ops)
3034 #else
3035 #define DMA40_PM_OPS    NULL
3036 #endif
3037
3038 /* Initialization functions. */
3039
3040 static int __init d40_phy_res_init(struct d40_base *base)
3041 {
3042         int i;
3043         int num_phy_chans_avail = 0;
3044         u32 val[2];
3045         int odd_even_bit = -2;
3046         int gcc = D40_DREG_GCC_ENA;
3047
3048         val[0] = readl(base->virtbase + D40_DREG_PRSME);
3049         val[1] = readl(base->virtbase + D40_DREG_PRSMO);
3050
3051         for (i = 0; i < base->num_phy_chans; i++) {
3052                 base->phy_res[i].num = i;
3053                 odd_even_bit += 2 * ((i % 2) == 0);
3054                 if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
3055                         /* Mark security only channels as occupied */
3056                         base->phy_res[i].allocated_src = D40_ALLOC_PHY;
3057                         base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
3058                         base->phy_res[i].reserved = true;
3059                         gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3060                                                        D40_DREG_GCC_SRC);
3061                         gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3062                                                        D40_DREG_GCC_DST);
3063
3064
3065                 } else {
3066                         base->phy_res[i].allocated_src = D40_ALLOC_FREE;
3067                         base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
3068                         base->phy_res[i].reserved = false;
3069                         num_phy_chans_avail++;
3070                 }
3071                 spin_lock_init(&base->phy_res[i].lock);
3072         }
3073
3074         /* Mark disabled channels as occupied */
3075         for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
3076                 int chan = base->plat_data->disabled_channels[i];
3077
3078                 base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
3079                 base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
3080                 base->phy_res[chan].reserved = true;
3081                 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3082                                                D40_DREG_GCC_SRC);
3083                 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3084                                                D40_DREG_GCC_DST);
3085                 num_phy_chans_avail--;
3086         }
3087
3088         /* Mark soft_lli channels */
3089         for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) {
3090                 int chan = base->plat_data->soft_lli_chans[i];
3091
3092                 base->phy_res[chan].use_soft_lli = true;
3093         }
3094
3095         dev_info(base->dev, "%d of %d physical DMA channels available\n",
3096                  num_phy_chans_avail, base->num_phy_chans);
3097
3098         /* Verify settings extended vs standard */
3099         val[0] = readl(base->virtbase + D40_DREG_PRTYP);
3100
3101         for (i = 0; i < base->num_phy_chans; i++) {
3102
3103                 if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
3104                     (val[0] & 0x3) != 1)
3105                         dev_info(base->dev,
3106                                  "[%s] INFO: channel %d is misconfigured (%d)\n",
3107                                  __func__, i, val[0] & 0x3);
3108
3109                 val[0] = val[0] >> 2;
3110         }
3111
3112         /*
3113          * To keep things simple, Enable all clocks initially.
3114          * The clocks will get managed later post channel allocation.
3115          * The clocks for the event lines on which reserved channels exists
3116          * are not managed here.
3117          */
3118         writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3119         base->gcc_pwr_off_mask = gcc;
3120
3121         return num_phy_chans_avail;
3122 }
3123
3124 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
3125 {
3126         struct stedma40_platform_data *plat_data;
3127         struct clk *clk = NULL;
3128         void __iomem *virtbase = NULL;
3129         struct resource *res = NULL;
3130         struct d40_base *base = NULL;
3131         int num_log_chans = 0;
3132         int num_phy_chans;
3133         int clk_ret = -EINVAL;
3134         int i;
3135         u32 pid;
3136         u32 cid;
3137         u8 rev;
3138
3139         clk = clk_get(&pdev->dev, NULL);
3140         if (IS_ERR(clk)) {
3141                 d40_err(&pdev->dev, "No matching clock found\n");
3142                 goto failure;
3143         }
3144
3145         clk_ret = clk_prepare_enable(clk);
3146         if (clk_ret) {
3147                 d40_err(&pdev->dev, "Failed to prepare/enable clock\n");
3148                 goto failure;
3149         }
3150
3151         /* Get IO for DMAC base address */
3152         res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
3153         if (!res)
3154                 goto failure;
3155
3156         if (request_mem_region(res->start, resource_size(res),
3157                                D40_NAME " I/O base") == NULL)
3158                 goto failure;
3159
3160         virtbase = ioremap(res->start, resource_size(res));
3161         if (!virtbase)
3162                 goto failure;
3163
3164         /* This is just a regular AMBA PrimeCell ID actually */
3165         for (pid = 0, i = 0; i < 4; i++)
3166                 pid |= (readl(virtbase + resource_size(res) - 0x20 + 4 * i)
3167                         & 255) << (i * 8);
3168         for (cid = 0, i = 0; i < 4; i++)
3169                 cid |= (readl(virtbase + resource_size(res) - 0x10 + 4 * i)
3170                         & 255) << (i * 8);
3171
3172         if (cid != AMBA_CID) {
3173                 d40_err(&pdev->dev, "Unknown hardware! No PrimeCell ID\n");
3174                 goto failure;
3175         }
3176         if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
3177                 d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
3178                         AMBA_MANF_BITS(pid),
3179                         AMBA_VENDOR_ST);
3180                 goto failure;
3181         }
3182         /*
3183          * HW revision:
3184          * DB8500ed has revision 0
3185          * ? has revision 1
3186          * DB8500v1 has revision 2
3187          * DB8500v2 has revision 3
3188          * AP9540v1 has revision 4
3189          * DB8540v1 has revision 4
3190          */
3191         rev = AMBA_REV_BITS(pid);
3192
3193         plat_data = pdev->dev.platform_data;
3194
3195         /* The number of physical channels on this HW */
3196         if (plat_data->num_of_phy_chans)
3197                 num_phy_chans = plat_data->num_of_phy_chans;
3198         else
3199                 num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
3200
3201         dev_info(&pdev->dev, "hardware revision: %d @ 0x%x with %d physical channels\n",
3202                  rev, res->start, num_phy_chans);
3203
3204         if (rev < 2) {
3205                 d40_err(&pdev->dev, "hardware revision: %d is not supported",
3206                         rev);
3207                 goto failure;
3208         }
3209
3210         /* Count the number of logical channels in use */
3211         for (i = 0; i < plat_data->dev_len; i++)
3212                 if (plat_data->dev_rx[i] != 0)
3213                         num_log_chans++;
3214
3215         for (i = 0; i < plat_data->dev_len; i++)
3216                 if (plat_data->dev_tx[i] != 0)
3217                         num_log_chans++;
3218
3219         base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
3220                        (num_phy_chans + num_log_chans + plat_data->memcpy_len) *
3221                        sizeof(struct d40_chan), GFP_KERNEL);
3222
3223         if (base == NULL) {
3224                 d40_err(&pdev->dev, "Out of memory\n");
3225                 goto failure;
3226         }
3227
3228         base->rev = rev;
3229         base->clk = clk;
3230         base->num_phy_chans = num_phy_chans;
3231         base->num_log_chans = num_log_chans;
3232         base->phy_start = res->start;
3233         base->phy_size = resource_size(res);
3234         base->virtbase = virtbase;
3235         base->plat_data = plat_data;
3236         base->dev = &pdev->dev;
3237         base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3238         base->log_chans = &base->phy_chans[num_phy_chans];
3239
3240         if (base->plat_data->num_of_phy_chans == 14) {
3241                 base->gen_dmac.backup = d40_backup_regs_v4b;
3242                 base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
3243                 base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
3244                 base->gen_dmac.interrupt_clear = D40_DREG_CPCICR;
3245                 base->gen_dmac.realtime_en = D40_DREG_CRSEG1;
3246                 base->gen_dmac.realtime_clear = D40_DREG_CRCEG1;
3247                 base->gen_dmac.high_prio_en = D40_DREG_CPSEG1;
3248                 base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1;
3249                 base->gen_dmac.il = il_v4b;
3250                 base->gen_dmac.il_size = ARRAY_SIZE(il_v4b);
3251                 base->gen_dmac.init_reg = dma_init_reg_v4b;
3252                 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
3253         } else {
3254                 if (base->rev >= 3) {
3255                         base->gen_dmac.backup = d40_backup_regs_v4a;
3256                         base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
3257                 }
3258                 base->gen_dmac.interrupt_en = D40_DREG_PCMIS;
3259                 base->gen_dmac.interrupt_clear = D40_DREG_PCICR;
3260                 base->gen_dmac.realtime_en = D40_DREG_RSEG1;
3261                 base->gen_dmac.realtime_clear = D40_DREG_RCEG1;
3262                 base->gen_dmac.high_prio_en = D40_DREG_PSEG1;
3263                 base->gen_dmac.high_prio_clear = D40_DREG_PCEG1;
3264                 base->gen_dmac.il = il_v4a;
3265                 base->gen_dmac.il_size = ARRAY_SIZE(il_v4a);
3266                 base->gen_dmac.init_reg = dma_init_reg_v4a;
3267                 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a);
3268         }
3269
3270         base->phy_res = kzalloc(num_phy_chans * sizeof(struct d40_phy_res),
3271                                 GFP_KERNEL);
3272         if (!base->phy_res)
3273                 goto failure;
3274
3275         base->lookup_phy_chans = kzalloc(num_phy_chans *
3276                                          sizeof(struct d40_chan *),
3277                                          GFP_KERNEL);
3278         if (!base->lookup_phy_chans)
3279                 goto failure;
3280
3281         if (num_log_chans + plat_data->memcpy_len) {
3282                 /*
3283                  * The max number of logical channels are event lines for all
3284                  * src devices and dst devices
3285                  */
3286                 base->lookup_log_chans = kzalloc(plat_data->dev_len * 2 *
3287                                                  sizeof(struct d40_chan *),
3288                                                  GFP_KERNEL);
3289                 if (!base->lookup_log_chans)
3290                         goto failure;
3291         }
3292
3293         base->reg_val_backup_chan = kmalloc(base->num_phy_chans *
3294                                             sizeof(d40_backup_regs_chan),
3295                                             GFP_KERNEL);
3296         if (!base->reg_val_backup_chan)
3297                 goto failure;
3298
3299         base->lcla_pool.alloc_map =
3300                 kzalloc(num_phy_chans * sizeof(struct d40_desc *)
3301                         * D40_LCLA_LINK_PER_EVENT_GRP, GFP_KERNEL);
3302         if (!base->lcla_pool.alloc_map)
3303                 goto failure;
3304
3305         base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3306                                             0, SLAB_HWCACHE_ALIGN,
3307                                             NULL);
3308         if (base->desc_slab == NULL)
3309                 goto failure;
3310
3311         return base;
3312
3313 failure:
3314         if (!clk_ret)
3315                 clk_disable_unprepare(clk);
3316         if (!IS_ERR(clk))
3317                 clk_put(clk);
3318         if (virtbase)
3319                 iounmap(virtbase);
3320         if (res)
3321                 release_mem_region(res->start,
3322                                    resource_size(res));
3323         if (virtbase)
3324                 iounmap(virtbase);
3325
3326         if (base) {
3327                 kfree(base->lcla_pool.alloc_map);
3328                 kfree(base->reg_val_backup_chan);
3329                 kfree(base->lookup_log_chans);
3330                 kfree(base->lookup_phy_chans);
3331                 kfree(base->phy_res);
3332                 kfree(base);
3333         }
3334
3335         return NULL;
3336 }
3337
3338 static void __init d40_hw_init(struct d40_base *base)
3339 {
3340
3341         int i;
3342         u32 prmseo[2] = {0, 0};
3343         u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3344         u32 pcmis = 0;
3345         u32 pcicr = 0;
3346         struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg;
3347         u32 reg_size = base->gen_dmac.init_reg_size;
3348
3349         for (i = 0; i < reg_size; i++)
3350                 writel(dma_init_reg[i].val,
3351                        base->virtbase + dma_init_reg[i].reg);
3352
3353         /* Configure all our dma channels to default settings */
3354         for (i = 0; i < base->num_phy_chans; i++) {
3355
3356                 activeo[i % 2] = activeo[i % 2] << 2;
3357
3358                 if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3359                     == D40_ALLOC_PHY) {
3360                         activeo[i % 2] |= 3;
3361                         continue;
3362                 }
3363
3364                 /* Enable interrupt # */
3365                 pcmis = (pcmis << 1) | 1;
3366
3367                 /* Clear interrupt # */
3368                 pcicr = (pcicr << 1) | 1;
3369
3370                 /* Set channel to physical mode */
3371                 prmseo[i % 2] = prmseo[i % 2] << 2;
3372                 prmseo[i % 2] |= 1;
3373
3374         }
3375
3376         writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3377         writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3378         writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3379         writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3380
3381         /* Write which interrupt to enable */
3382         writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en);
3383
3384         /* Write which interrupt to clear */
3385         writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear);
3386
3387         /* These are __initdata and cannot be accessed after init */
3388         base->gen_dmac.init_reg = NULL;
3389         base->gen_dmac.init_reg_size = 0;
3390 }
3391
3392 static int __init d40_lcla_allocate(struct d40_base *base)
3393 {
3394         struct d40_lcla_pool *pool = &base->lcla_pool;
3395         unsigned long *page_list;
3396         int i, j;
3397         int ret = 0;
3398
3399         /*
3400          * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3401          * To full fill this hardware requirement without wasting 256 kb
3402          * we allocate pages until we get an aligned one.
3403          */
3404         page_list = kmalloc(sizeof(unsigned long) * MAX_LCLA_ALLOC_ATTEMPTS,
3405                             GFP_KERNEL);
3406
3407         if (!page_list) {
3408                 ret = -ENOMEM;
3409                 goto failure;
3410         }
3411
3412         /* Calculating how many pages that are required */
3413         base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3414
3415         for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3416                 page_list[i] = __get_free_pages(GFP_KERNEL,
3417                                                 base->lcla_pool.pages);
3418                 if (!page_list[i]) {
3419
3420                         d40_err(base->dev, "Failed to allocate %d pages.\n",
3421                                 base->lcla_pool.pages);
3422
3423                         for (j = 0; j < i; j++)
3424                                 free_pages(page_list[j], base->lcla_pool.pages);
3425                         goto failure;
3426                 }
3427
3428                 if ((virt_to_phys((void *)page_list[i]) &
3429                      (LCLA_ALIGNMENT - 1)) == 0)
3430                         break;
3431         }
3432
3433         for (j = 0; j < i; j++)
3434                 free_pages(page_list[j], base->lcla_pool.pages);
3435
3436         if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3437                 base->lcla_pool.base = (void *)page_list[i];
3438         } else {
3439                 /*
3440                  * After many attempts and no succees with finding the correct
3441                  * alignment, try with allocating a big buffer.
3442                  */
3443                 dev_warn(base->dev,
3444                          "[%s] Failed to get %d pages @ 18 bit align.\n",
3445                          __func__, base->lcla_pool.pages);
3446                 base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3447                                                          base->num_phy_chans +
3448                                                          LCLA_ALIGNMENT,
3449                                                          GFP_KERNEL);
3450                 if (!base->lcla_pool.base_unaligned) {
3451                         ret = -ENOMEM;
3452                         goto failure;
3453                 }
3454
3455                 base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3456                                                  LCLA_ALIGNMENT);
3457         }
3458
3459         pool->dma_addr = dma_map_single(base->dev, pool->base,
3460                                         SZ_1K * base->num_phy_chans,
3461                                         DMA_TO_DEVICE);
3462         if (dma_mapping_error(base->dev, pool->dma_addr)) {
3463                 pool->dma_addr = 0;
3464                 ret = -ENOMEM;
3465                 goto failure;
3466         }
3467
3468         writel(virt_to_phys(base->lcla_pool.base),
3469                base->virtbase + D40_DREG_LCLA);
3470 failure:
3471         kfree(page_list);
3472         return ret;
3473 }
3474
3475 static int __init d40_probe(struct platform_device *pdev)
3476 {
3477         int err;
3478         int ret = -ENOENT;
3479         struct d40_base *base;
3480         struct resource *res = NULL;
3481         int num_reserved_chans;
3482         u32 val;
3483
3484         base = d40_hw_detect_init(pdev);
3485
3486         if (!base)
3487                 goto failure;
3488
3489         num_reserved_chans = d40_phy_res_init(base);
3490
3491         platform_set_drvdata(pdev, base);
3492
3493         spin_lock_init(&base->interrupt_lock);
3494         spin_lock_init(&base->execmd_lock);
3495
3496         /* Get IO for logical channel parameter address */
3497         res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
3498         if (!res) {
3499                 ret = -ENOENT;
3500                 d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
3501                 goto failure;
3502         }
3503         base->lcpa_size = resource_size(res);
3504         base->phy_lcpa = res->start;
3505
3506         if (request_mem_region(res->start, resource_size(res),
3507                                D40_NAME " I/O lcpa") == NULL) {
3508                 ret = -EBUSY;
3509                 d40_err(&pdev->dev,
3510                         "Failed to request LCPA region 0x%x-0x%x\n",
3511                         res->start, res->end);
3512                 goto failure;
3513         }
3514
3515         /* We make use of ESRAM memory for this. */
3516         val = readl(base->virtbase + D40_DREG_LCPA);
3517         if (res->start != val && val != 0) {
3518                 dev_warn(&pdev->dev,
3519                          "[%s] Mismatch LCPA dma 0x%x, def 0x%x\n",
3520                          __func__, val, res->start);
3521         } else
3522                 writel(res->start, base->virtbase + D40_DREG_LCPA);
3523
3524         base->lcpa_base = ioremap(res->start, resource_size(res));
3525         if (!base->lcpa_base) {
3526                 ret = -ENOMEM;
3527                 d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
3528                 goto failure;
3529         }
3530         /* If lcla has to be located in ESRAM we don't need to allocate */
3531         if (base->plat_data->use_esram_lcla) {
3532                 res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3533                                                         "lcla_esram");
3534                 if (!res) {
3535                         ret = -ENOENT;
3536                         d40_err(&pdev->dev,
3537                                 "No \"lcla_esram\" memory resource\n");
3538                         goto failure;
3539                 }
3540                 base->lcla_pool.base = ioremap(res->start,
3541                                                 resource_size(res));
3542                 if (!base->lcla_pool.base) {
3543                         ret = -ENOMEM;
3544                         d40_err(&pdev->dev, "Failed to ioremap LCLA region\n");
3545                         goto failure;
3546                 }
3547                 writel(res->start, base->virtbase + D40_DREG_LCLA);
3548
3549         } else {
3550                 ret = d40_lcla_allocate(base);
3551                 if (ret) {
3552                         d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
3553                         goto failure;
3554                 }
3555         }
3556
3557         spin_lock_init(&base->lcla_pool.lock);
3558
3559         base->irq = platform_get_irq(pdev, 0);
3560
3561         ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3562         if (ret) {
3563                 d40_err(&pdev->dev, "No IRQ defined\n");
3564                 goto failure;
3565         }
3566
3567         pm_runtime_irq_safe(base->dev);
3568         pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3569         pm_runtime_use_autosuspend(base->dev);
3570         pm_runtime_enable(base->dev);
3571         pm_runtime_resume(base->dev);
3572
3573         if (base->plat_data->use_esram_lcla) {
3574
3575                 base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3576                 if (IS_ERR(base->lcpa_regulator)) {
3577                         d40_err(&pdev->dev, "Failed to get lcpa_regulator\n");
3578                         base->lcpa_regulator = NULL;
3579                         goto failure;
3580                 }
3581
3582                 ret = regulator_enable(base->lcpa_regulator);
3583                 if (ret) {
3584                         d40_err(&pdev->dev,
3585                                 "Failed to enable lcpa_regulator\n");
3586                         regulator_put(base->lcpa_regulator);
3587                         base->lcpa_regulator = NULL;
3588                         goto failure;
3589                 }
3590         }
3591
3592         base->initialized = true;
3593         err = d40_dmaengine_init(base, num_reserved_chans);
3594         if (err)
3595                 goto failure;
3596
3597         base->dev->dma_parms = &base->dma_parms;
3598         err = dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
3599         if (err) {
3600                 d40_err(&pdev->dev, "Failed to set dma max seg size\n");
3601                 goto failure;
3602         }
3603
3604         d40_hw_init(base);
3605
3606         dev_info(base->dev, "initialized\n");
3607         return 0;
3608
3609 failure:
3610         if (base) {
3611                 if (base->desc_slab)
3612                         kmem_cache_destroy(base->desc_slab);
3613                 if (base->virtbase)
3614                         iounmap(base->virtbase);
3615
3616                 if (base->lcla_pool.base && base->plat_data->use_esram_lcla) {
3617                         iounmap(base->lcla_pool.base);
3618                         base->lcla_pool.base = NULL;
3619                 }
3620
3621                 if (base->lcla_pool.dma_addr)
3622                         dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3623                                          SZ_1K * base->num_phy_chans,
3624                                          DMA_TO_DEVICE);
3625
3626                 if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3627                         free_pages((unsigned long)base->lcla_pool.base,
3628                                    base->lcla_pool.pages);
3629
3630                 kfree(base->lcla_pool.base_unaligned);
3631
3632                 if (base->phy_lcpa)
3633                         release_mem_region(base->phy_lcpa,
3634                                            base->lcpa_size);
3635                 if (base->phy_start)
3636                         release_mem_region(base->phy_start,
3637                                            base->phy_size);
3638                 if (base->clk) {
3639                         clk_disable_unprepare(base->clk);
3640                         clk_put(base->clk);
3641                 }
3642
3643                 if (base->lcpa_regulator) {
3644                         regulator_disable(base->lcpa_regulator);
3645                         regulator_put(base->lcpa_regulator);
3646                 }
3647
3648                 kfree(base->lcla_pool.alloc_map);
3649                 kfree(base->lookup_log_chans);
3650                 kfree(base->lookup_phy_chans);
3651                 kfree(base->phy_res);
3652                 kfree(base);
3653         }
3654
3655         d40_err(&pdev->dev, "probe failed\n");
3656         return ret;
3657 }
3658
3659 static struct platform_driver d40_driver = {
3660         .driver = {
3661                 .owner = THIS_MODULE,
3662                 .name  = D40_NAME,
3663                 .pm = DMA40_PM_OPS,
3664         },
3665 };
3666
3667 static int __init stedma40_init(void)
3668 {
3669         return platform_driver_probe(&d40_driver, d40_probe);
3670 }
3671 subsys_initcall(stedma40_init);