Merge remote-tracking branch 'lsk/v3.10/topic/gator' into linux-linaro-lsk
[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         bool callback_active;
1591         dma_async_tx_callback callback;
1592         void *callback_param;
1593
1594         spin_lock_irqsave(&d40c->lock, flags);
1595
1596         /* Get first entry from the done list */
1597         d40d = d40_first_done(d40c);
1598         if (d40d == NULL) {
1599                 /* Check if we have reached here for cyclic job */
1600                 d40d = d40_first_active_get(d40c);
1601                 if (d40d == NULL || !d40d->cyclic)
1602                         goto err;
1603         }
1604
1605         if (!d40d->cyclic)
1606                 dma_cookie_complete(&d40d->txd);
1607
1608         /*
1609          * If terminating a channel pending_tx is set to zero.
1610          * This prevents any finished active jobs to return to the client.
1611          */
1612         if (d40c->pending_tx == 0) {
1613                 spin_unlock_irqrestore(&d40c->lock, flags);
1614                 return;
1615         }
1616
1617         /* Callback to client */
1618         callback_active = !!(d40d->txd.flags & DMA_PREP_INTERRUPT);
1619         callback = d40d->txd.callback;
1620         callback_param = d40d->txd.callback_param;
1621
1622         if (!d40d->cyclic) {
1623                 if (async_tx_test_ack(&d40d->txd)) {
1624                         d40_desc_remove(d40d);
1625                         d40_desc_free(d40c, d40d);
1626                 } else if (!d40d->is_in_client_list) {
1627                         d40_desc_remove(d40d);
1628                         d40_lcla_free_all(d40c, d40d);
1629                         list_add_tail(&d40d->node, &d40c->client);
1630                         d40d->is_in_client_list = true;
1631                 }
1632         }
1633
1634         d40c->pending_tx--;
1635
1636         if (d40c->pending_tx)
1637                 tasklet_schedule(&d40c->tasklet);
1638
1639         spin_unlock_irqrestore(&d40c->lock, flags);
1640
1641         if (callback_active && callback)
1642                 callback(callback_param);
1643
1644         return;
1645
1646 err:
1647         /* Rescue manouver if receiving double interrupts */
1648         if (d40c->pending_tx > 0)
1649                 d40c->pending_tx--;
1650         spin_unlock_irqrestore(&d40c->lock, flags);
1651 }
1652
1653 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1654 {
1655         int i;
1656         u32 idx;
1657         u32 row;
1658         long chan = -1;
1659         struct d40_chan *d40c;
1660         unsigned long flags;
1661         struct d40_base *base = data;
1662         u32 regs[base->gen_dmac.il_size];
1663         struct d40_interrupt_lookup *il = base->gen_dmac.il;
1664         u32 il_size = base->gen_dmac.il_size;
1665
1666         spin_lock_irqsave(&base->interrupt_lock, flags);
1667
1668         /* Read interrupt status of both logical and physical channels */
1669         for (i = 0; i < il_size; i++)
1670                 regs[i] = readl(base->virtbase + il[i].src);
1671
1672         for (;;) {
1673
1674                 chan = find_next_bit((unsigned long *)regs,
1675                                      BITS_PER_LONG * il_size, chan + 1);
1676
1677                 /* No more set bits found? */
1678                 if (chan == BITS_PER_LONG * il_size)
1679                         break;
1680
1681                 row = chan / BITS_PER_LONG;
1682                 idx = chan & (BITS_PER_LONG - 1);
1683
1684                 if (il[row].offset == D40_PHY_CHAN)
1685                         d40c = base->lookup_phy_chans[idx];
1686                 else
1687                         d40c = base->lookup_log_chans[il[row].offset + idx];
1688
1689                 if (!d40c) {
1690                         /*
1691                          * No error because this can happen if something else
1692                          * in the system is using the channel.
1693                          */
1694                         continue;
1695                 }
1696
1697                 /* ACK interrupt */
1698                 writel(1 << idx, base->virtbase + il[row].clr);
1699
1700                 spin_lock(&d40c->lock);
1701
1702                 if (!il[row].is_error)
1703                         dma_tc_handle(d40c);
1704                 else
1705                         d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1706                                 chan, il[row].offset, idx);
1707
1708                 spin_unlock(&d40c->lock);
1709         }
1710
1711         spin_unlock_irqrestore(&base->interrupt_lock, flags);
1712
1713         return IRQ_HANDLED;
1714 }
1715
1716 static int d40_validate_conf(struct d40_chan *d40c,
1717                              struct stedma40_chan_cfg *conf)
1718 {
1719         int res = 0;
1720         u32 dst_event_group = D40_TYPE_TO_GROUP(conf->dst_dev_type);
1721         u32 src_event_group = D40_TYPE_TO_GROUP(conf->src_dev_type);
1722         bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1723
1724         if (!conf->dir) {
1725                 chan_err(d40c, "Invalid direction.\n");
1726                 res = -EINVAL;
1727         }
1728
1729         if (conf->dst_dev_type != STEDMA40_DEV_DST_MEMORY &&
1730             d40c->base->plat_data->dev_tx[conf->dst_dev_type] == 0 &&
1731             d40c->runtime_addr == 0) {
1732
1733                 chan_err(d40c, "Invalid TX channel address (%d)\n",
1734                          conf->dst_dev_type);
1735                 res = -EINVAL;
1736         }
1737
1738         if (conf->src_dev_type != STEDMA40_DEV_SRC_MEMORY &&
1739             d40c->base->plat_data->dev_rx[conf->src_dev_type] == 0 &&
1740             d40c->runtime_addr == 0) {
1741                 chan_err(d40c, "Invalid RX channel address (%d)\n",
1742                         conf->src_dev_type);
1743                 res = -EINVAL;
1744         }
1745
1746         if (conf->dir == STEDMA40_MEM_TO_PERIPH &&
1747             dst_event_group == STEDMA40_DEV_DST_MEMORY) {
1748                 chan_err(d40c, "Invalid dst\n");
1749                 res = -EINVAL;
1750         }
1751
1752         if (conf->dir == STEDMA40_PERIPH_TO_MEM &&
1753             src_event_group == STEDMA40_DEV_SRC_MEMORY) {
1754                 chan_err(d40c, "Invalid src\n");
1755                 res = -EINVAL;
1756         }
1757
1758         if (src_event_group == STEDMA40_DEV_SRC_MEMORY &&
1759             dst_event_group == STEDMA40_DEV_DST_MEMORY && is_log) {
1760                 chan_err(d40c, "No event line\n");
1761                 res = -EINVAL;
1762         }
1763
1764         if (conf->dir == STEDMA40_PERIPH_TO_PERIPH &&
1765             (src_event_group != dst_event_group)) {
1766                 chan_err(d40c, "Invalid event group\n");
1767                 res = -EINVAL;
1768         }
1769
1770         if (conf->dir == STEDMA40_PERIPH_TO_PERIPH) {
1771                 /*
1772                  * DMAC HW supports it. Will be added to this driver,
1773                  * in case any dma client requires it.
1774                  */
1775                 chan_err(d40c, "periph to periph not supported\n");
1776                 res = -EINVAL;
1777         }
1778
1779         if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1780             (1 << conf->src_info.data_width) !=
1781             d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1782             (1 << conf->dst_info.data_width)) {
1783                 /*
1784                  * The DMAC hardware only supports
1785                  * src (burst x width) == dst (burst x width)
1786                  */
1787
1788                 chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1789                 res = -EINVAL;
1790         }
1791
1792         return res;
1793 }
1794
1795 static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1796                                bool is_src, int log_event_line, bool is_log,
1797                                bool *first_user)
1798 {
1799         unsigned long flags;
1800         spin_lock_irqsave(&phy->lock, flags);
1801
1802         *first_user = ((phy->allocated_src | phy->allocated_dst)
1803                         == D40_ALLOC_FREE);
1804
1805         if (!is_log) {
1806                 /* Physical interrupts are masked per physical full channel */
1807                 if (phy->allocated_src == D40_ALLOC_FREE &&
1808                     phy->allocated_dst == D40_ALLOC_FREE) {
1809                         phy->allocated_dst = D40_ALLOC_PHY;
1810                         phy->allocated_src = D40_ALLOC_PHY;
1811                         goto found;
1812                 } else
1813                         goto not_found;
1814         }
1815
1816         /* Logical channel */
1817         if (is_src) {
1818                 if (phy->allocated_src == D40_ALLOC_PHY)
1819                         goto not_found;
1820
1821                 if (phy->allocated_src == D40_ALLOC_FREE)
1822                         phy->allocated_src = D40_ALLOC_LOG_FREE;
1823
1824                 if (!(phy->allocated_src & (1 << log_event_line))) {
1825                         phy->allocated_src |= 1 << log_event_line;
1826                         goto found;
1827                 } else
1828                         goto not_found;
1829         } else {
1830                 if (phy->allocated_dst == D40_ALLOC_PHY)
1831                         goto not_found;
1832
1833                 if (phy->allocated_dst == D40_ALLOC_FREE)
1834                         phy->allocated_dst = D40_ALLOC_LOG_FREE;
1835
1836                 if (!(phy->allocated_dst & (1 << log_event_line))) {
1837                         phy->allocated_dst |= 1 << log_event_line;
1838                         goto found;
1839                 } else
1840                         goto not_found;
1841         }
1842
1843 not_found:
1844         spin_unlock_irqrestore(&phy->lock, flags);
1845         return false;
1846 found:
1847         spin_unlock_irqrestore(&phy->lock, flags);
1848         return true;
1849 }
1850
1851 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1852                                int log_event_line)
1853 {
1854         unsigned long flags;
1855         bool is_free = false;
1856
1857         spin_lock_irqsave(&phy->lock, flags);
1858         if (!log_event_line) {
1859                 phy->allocated_dst = D40_ALLOC_FREE;
1860                 phy->allocated_src = D40_ALLOC_FREE;
1861                 is_free = true;
1862                 goto out;
1863         }
1864
1865         /* Logical channel */
1866         if (is_src) {
1867                 phy->allocated_src &= ~(1 << log_event_line);
1868                 if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1869                         phy->allocated_src = D40_ALLOC_FREE;
1870         } else {
1871                 phy->allocated_dst &= ~(1 << log_event_line);
1872                 if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1873                         phy->allocated_dst = D40_ALLOC_FREE;
1874         }
1875
1876         is_free = ((phy->allocated_src | phy->allocated_dst) ==
1877                    D40_ALLOC_FREE);
1878
1879 out:
1880         spin_unlock_irqrestore(&phy->lock, flags);
1881
1882         return is_free;
1883 }
1884
1885 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1886 {
1887         int dev_type;
1888         int event_group;
1889         int event_line;
1890         struct d40_phy_res *phys;
1891         int i;
1892         int j;
1893         int log_num;
1894         int num_phy_chans;
1895         bool is_src;
1896         bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1897
1898         phys = d40c->base->phy_res;
1899         num_phy_chans = d40c->base->num_phy_chans;
1900
1901         if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1902                 dev_type = d40c->dma_cfg.src_dev_type;
1903                 log_num = 2 * dev_type;
1904                 is_src = true;
1905         } else if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1906                    d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1907                 /* dst event lines are used for logical memcpy */
1908                 dev_type = d40c->dma_cfg.dst_dev_type;
1909                 log_num = 2 * dev_type + 1;
1910                 is_src = false;
1911         } else
1912                 return -EINVAL;
1913
1914         event_group = D40_TYPE_TO_GROUP(dev_type);
1915         event_line = D40_TYPE_TO_EVENT(dev_type);
1916
1917         if (!is_log) {
1918                 if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1919                         /* Find physical half channel */
1920                         if (d40c->dma_cfg.use_fixed_channel) {
1921                                 i = d40c->dma_cfg.phy_channel;
1922                                 if (d40_alloc_mask_set(&phys[i], is_src,
1923                                                        0, is_log,
1924                                                        first_phy_user))
1925                                         goto found_phy;
1926                         } else {
1927                                 for (i = 0; i < num_phy_chans; i++) {
1928                                         if (d40_alloc_mask_set(&phys[i], is_src,
1929                                                        0, is_log,
1930                                                        first_phy_user))
1931                                                 goto found_phy;
1932                                 }
1933                         }
1934                 } else
1935                         for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1936                                 int phy_num = j  + event_group * 2;
1937                                 for (i = phy_num; i < phy_num + 2; i++) {
1938                                         if (d40_alloc_mask_set(&phys[i],
1939                                                                is_src,
1940                                                                0,
1941                                                                is_log,
1942                                                                first_phy_user))
1943                                                 goto found_phy;
1944                                 }
1945                         }
1946                 return -EINVAL;
1947 found_phy:
1948                 d40c->phy_chan = &phys[i];
1949                 d40c->log_num = D40_PHY_CHAN;
1950                 goto out;
1951         }
1952         if (dev_type == -1)
1953                 return -EINVAL;
1954
1955         /* Find logical channel */
1956         for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1957                 int phy_num = j + event_group * 2;
1958
1959                 if (d40c->dma_cfg.use_fixed_channel) {
1960                         i = d40c->dma_cfg.phy_channel;
1961
1962                         if ((i != phy_num) && (i != phy_num + 1)) {
1963                                 dev_err(chan2dev(d40c),
1964                                         "invalid fixed phy channel %d\n", i);
1965                                 return -EINVAL;
1966                         }
1967
1968                         if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1969                                                is_log, first_phy_user))
1970                                 goto found_log;
1971
1972                         dev_err(chan2dev(d40c),
1973                                 "could not allocate fixed phy channel %d\n", i);
1974                         return -EINVAL;
1975                 }
1976
1977                 /*
1978                  * Spread logical channels across all available physical rather
1979                  * than pack every logical channel at the first available phy
1980                  * channels.
1981                  */
1982                 if (is_src) {
1983                         for (i = phy_num; i < phy_num + 2; i++) {
1984                                 if (d40_alloc_mask_set(&phys[i], is_src,
1985                                                        event_line, is_log,
1986                                                        first_phy_user))
1987                                         goto found_log;
1988                         }
1989                 } else {
1990                         for (i = phy_num + 1; i >= phy_num; i--) {
1991                                 if (d40_alloc_mask_set(&phys[i], is_src,
1992                                                        event_line, is_log,
1993                                                        first_phy_user))
1994                                         goto found_log;
1995                         }
1996                 }
1997         }
1998         return -EINVAL;
1999
2000 found_log:
2001         d40c->phy_chan = &phys[i];
2002         d40c->log_num = log_num;
2003 out:
2004
2005         if (is_log)
2006                 d40c->base->lookup_log_chans[d40c->log_num] = d40c;
2007         else
2008                 d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
2009
2010         return 0;
2011
2012 }
2013
2014 static int d40_config_memcpy(struct d40_chan *d40c)
2015 {
2016         dma_cap_mask_t cap = d40c->chan.device->cap_mask;
2017
2018         if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
2019                 d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_log;
2020                 d40c->dma_cfg.src_dev_type = STEDMA40_DEV_SRC_MEMORY;
2021                 d40c->dma_cfg.dst_dev_type = d40c->base->plat_data->
2022                         memcpy[d40c->chan.chan_id];
2023
2024         } else if (dma_has_cap(DMA_MEMCPY, cap) &&
2025                    dma_has_cap(DMA_SLAVE, cap)) {
2026                 d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_phy;
2027         } else {
2028                 chan_err(d40c, "No memcpy\n");
2029                 return -EINVAL;
2030         }
2031
2032         return 0;
2033 }
2034
2035 static int d40_free_dma(struct d40_chan *d40c)
2036 {
2037
2038         int res = 0;
2039         u32 event;
2040         struct d40_phy_res *phy = d40c->phy_chan;
2041         bool is_src;
2042
2043         /* Terminate all queued and active transfers */
2044         d40_term_all(d40c);
2045
2046         if (phy == NULL) {
2047                 chan_err(d40c, "phy == null\n");
2048                 return -EINVAL;
2049         }
2050
2051         if (phy->allocated_src == D40_ALLOC_FREE &&
2052             phy->allocated_dst == D40_ALLOC_FREE) {
2053                 chan_err(d40c, "channel already free\n");
2054                 return -EINVAL;
2055         }
2056
2057         if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
2058             d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
2059                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
2060                 is_src = false;
2061         } else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
2062                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
2063                 is_src = true;
2064         } else {
2065                 chan_err(d40c, "Unknown direction\n");
2066                 return -EINVAL;
2067         }
2068
2069         pm_runtime_get_sync(d40c->base->dev);
2070         res = d40_channel_execute_command(d40c, D40_DMA_STOP);
2071         if (res) {
2072                 chan_err(d40c, "stop failed\n");
2073                 goto out;
2074         }
2075
2076         d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
2077
2078         if (chan_is_logical(d40c))
2079                 d40c->base->lookup_log_chans[d40c->log_num] = NULL;
2080         else
2081                 d40c->base->lookup_phy_chans[phy->num] = NULL;
2082
2083         if (d40c->busy) {
2084                 pm_runtime_mark_last_busy(d40c->base->dev);
2085                 pm_runtime_put_autosuspend(d40c->base->dev);
2086         }
2087
2088         d40c->busy = false;
2089         d40c->phy_chan = NULL;
2090         d40c->configured = false;
2091 out:
2092
2093         pm_runtime_mark_last_busy(d40c->base->dev);
2094         pm_runtime_put_autosuspend(d40c->base->dev);
2095         return res;
2096 }
2097
2098 static bool d40_is_paused(struct d40_chan *d40c)
2099 {
2100         void __iomem *chanbase = chan_base(d40c);
2101         bool is_paused = false;
2102         unsigned long flags;
2103         void __iomem *active_reg;
2104         u32 status;
2105         u32 event;
2106
2107         spin_lock_irqsave(&d40c->lock, flags);
2108
2109         if (chan_is_physical(d40c)) {
2110                 if (d40c->phy_chan->num % 2 == 0)
2111                         active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
2112                 else
2113                         active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
2114
2115                 status = (readl(active_reg) &
2116                           D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
2117                         D40_CHAN_POS(d40c->phy_chan->num);
2118                 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
2119                         is_paused = true;
2120
2121                 goto _exit;
2122         }
2123
2124         if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
2125             d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
2126                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
2127                 status = readl(chanbase + D40_CHAN_REG_SDLNK);
2128         } else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
2129                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
2130                 status = readl(chanbase + D40_CHAN_REG_SSLNK);
2131         } else {
2132                 chan_err(d40c, "Unknown direction\n");
2133                 goto _exit;
2134         }
2135
2136         status = (status & D40_EVENTLINE_MASK(event)) >>
2137                 D40_EVENTLINE_POS(event);
2138
2139         if (status != D40_DMA_RUN)
2140                 is_paused = true;
2141 _exit:
2142         spin_unlock_irqrestore(&d40c->lock, flags);
2143         return is_paused;
2144
2145 }
2146
2147 static u32 stedma40_residue(struct dma_chan *chan)
2148 {
2149         struct d40_chan *d40c =
2150                 container_of(chan, struct d40_chan, chan);
2151         u32 bytes_left;
2152         unsigned long flags;
2153
2154         spin_lock_irqsave(&d40c->lock, flags);
2155         bytes_left = d40_residue(d40c);
2156         spin_unlock_irqrestore(&d40c->lock, flags);
2157
2158         return bytes_left;
2159 }
2160
2161 static int
2162 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
2163                 struct scatterlist *sg_src, struct scatterlist *sg_dst,
2164                 unsigned int sg_len, dma_addr_t src_dev_addr,
2165                 dma_addr_t dst_dev_addr)
2166 {
2167         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2168         struct stedma40_half_channel_info *src_info = &cfg->src_info;
2169         struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2170         int ret;
2171
2172         ret = d40_log_sg_to_lli(sg_src, sg_len,
2173                                 src_dev_addr,
2174                                 desc->lli_log.src,
2175                                 chan->log_def.lcsp1,
2176                                 src_info->data_width,
2177                                 dst_info->data_width);
2178
2179         ret = d40_log_sg_to_lli(sg_dst, sg_len,
2180                                 dst_dev_addr,
2181                                 desc->lli_log.dst,
2182                                 chan->log_def.lcsp3,
2183                                 dst_info->data_width,
2184                                 src_info->data_width);
2185
2186         return ret < 0 ? ret : 0;
2187 }
2188
2189 static int
2190 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2191                 struct scatterlist *sg_src, struct scatterlist *sg_dst,
2192                 unsigned int sg_len, dma_addr_t src_dev_addr,
2193                 dma_addr_t dst_dev_addr)
2194 {
2195         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2196         struct stedma40_half_channel_info *src_info = &cfg->src_info;
2197         struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2198         unsigned long flags = 0;
2199         int ret;
2200
2201         if (desc->cyclic)
2202                 flags |= LLI_CYCLIC | LLI_TERM_INT;
2203
2204         ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2205                                 desc->lli_phy.src,
2206                                 virt_to_phys(desc->lli_phy.src),
2207                                 chan->src_def_cfg,
2208                                 src_info, dst_info, flags);
2209
2210         ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2211                                 desc->lli_phy.dst,
2212                                 virt_to_phys(desc->lli_phy.dst),
2213                                 chan->dst_def_cfg,
2214                                 dst_info, src_info, flags);
2215
2216         dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2217                                    desc->lli_pool.size, DMA_TO_DEVICE);
2218
2219         return ret < 0 ? ret : 0;
2220 }
2221
2222 static struct d40_desc *
2223 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2224               unsigned int sg_len, unsigned long dma_flags)
2225 {
2226         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2227         struct d40_desc *desc;
2228         int ret;
2229
2230         desc = d40_desc_get(chan);
2231         if (!desc)
2232                 return NULL;
2233
2234         desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2235                                         cfg->dst_info.data_width);
2236         if (desc->lli_len < 0) {
2237                 chan_err(chan, "Unaligned size\n");
2238                 goto err;
2239         }
2240
2241         ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2242         if (ret < 0) {
2243                 chan_err(chan, "Could not allocate lli\n");
2244                 goto err;
2245         }
2246
2247         desc->lli_current = 0;
2248         desc->txd.flags = dma_flags;
2249         desc->txd.tx_submit = d40_tx_submit;
2250
2251         dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2252
2253         return desc;
2254
2255 err:
2256         d40_desc_free(chan, desc);
2257         return NULL;
2258 }
2259
2260 static dma_addr_t
2261 d40_get_dev_addr(struct d40_chan *chan, enum dma_transfer_direction direction)
2262 {
2263         struct stedma40_platform_data *plat = chan->base->plat_data;
2264         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2265         dma_addr_t addr = 0;
2266
2267         if (chan->runtime_addr)
2268                 return chan->runtime_addr;
2269
2270         if (direction == DMA_DEV_TO_MEM)
2271                 addr = plat->dev_rx[cfg->src_dev_type];
2272         else if (direction == DMA_MEM_TO_DEV)
2273                 addr = plat->dev_tx[cfg->dst_dev_type];
2274
2275         return addr;
2276 }
2277
2278 static struct dma_async_tx_descriptor *
2279 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2280             struct scatterlist *sg_dst, unsigned int sg_len,
2281             enum dma_transfer_direction direction, unsigned long dma_flags)
2282 {
2283         struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2284         dma_addr_t src_dev_addr = 0;
2285         dma_addr_t dst_dev_addr = 0;
2286         struct d40_desc *desc;
2287         unsigned long flags;
2288         int ret;
2289
2290         if (!chan->phy_chan) {
2291                 chan_err(chan, "Cannot prepare unallocated channel\n");
2292                 return NULL;
2293         }
2294
2295         spin_lock_irqsave(&chan->lock, flags);
2296
2297         desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2298         if (desc == NULL)
2299                 goto err;
2300
2301         if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2302                 desc->cyclic = true;
2303
2304         if (direction != DMA_TRANS_NONE) {
2305                 dma_addr_t dev_addr = d40_get_dev_addr(chan, direction);
2306
2307                 if (direction == DMA_DEV_TO_MEM)
2308                         src_dev_addr = dev_addr;
2309                 else if (direction == DMA_MEM_TO_DEV)
2310                         dst_dev_addr = dev_addr;
2311         }
2312
2313         if (chan_is_logical(chan))
2314                 ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2315                                       sg_len, src_dev_addr, dst_dev_addr);
2316         else
2317                 ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2318                                       sg_len, src_dev_addr, dst_dev_addr);
2319
2320         if (ret) {
2321                 chan_err(chan, "Failed to prepare %s sg job: %d\n",
2322                          chan_is_logical(chan) ? "log" : "phy", ret);
2323                 goto err;
2324         }
2325
2326         /*
2327          * add descriptor to the prepare queue in order to be able
2328          * to free them later in terminate_all
2329          */
2330         list_add_tail(&desc->node, &chan->prepare_queue);
2331
2332         spin_unlock_irqrestore(&chan->lock, flags);
2333
2334         return &desc->txd;
2335
2336 err:
2337         if (desc)
2338                 d40_desc_free(chan, desc);
2339         spin_unlock_irqrestore(&chan->lock, flags);
2340         return NULL;
2341 }
2342
2343 bool stedma40_filter(struct dma_chan *chan, void *data)
2344 {
2345         struct stedma40_chan_cfg *info = data;
2346         struct d40_chan *d40c =
2347                 container_of(chan, struct d40_chan, chan);
2348         int err;
2349
2350         if (data) {
2351                 err = d40_validate_conf(d40c, info);
2352                 if (!err)
2353                         d40c->dma_cfg = *info;
2354         } else
2355                 err = d40_config_memcpy(d40c);
2356
2357         if (!err)
2358                 d40c->configured = true;
2359
2360         return err == 0;
2361 }
2362 EXPORT_SYMBOL(stedma40_filter);
2363
2364 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2365 {
2366         bool realtime = d40c->dma_cfg.realtime;
2367         bool highprio = d40c->dma_cfg.high_priority;
2368         u32 rtreg;
2369         u32 event = D40_TYPE_TO_EVENT(dev_type);
2370         u32 group = D40_TYPE_TO_GROUP(dev_type);
2371         u32 bit = 1 << event;
2372         u32 prioreg;
2373         struct d40_gen_dmac *dmac = &d40c->base->gen_dmac;
2374
2375         rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear;
2376         /*
2377          * Due to a hardware bug, in some cases a logical channel triggered by
2378          * a high priority destination event line can generate extra packet
2379          * transactions.
2380          *
2381          * The workaround is to not set the high priority level for the
2382          * destination event lines that trigger logical channels.
2383          */
2384         if (!src && chan_is_logical(d40c))
2385                 highprio = false;
2386
2387         prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear;
2388
2389         /* Destination event lines are stored in the upper halfword */
2390         if (!src)
2391                 bit <<= 16;
2392
2393         writel(bit, d40c->base->virtbase + prioreg + group * 4);
2394         writel(bit, d40c->base->virtbase + rtreg + group * 4);
2395 }
2396
2397 static void d40_set_prio_realtime(struct d40_chan *d40c)
2398 {
2399         if (d40c->base->rev < 3)
2400                 return;
2401
2402         if ((d40c->dma_cfg.dir ==  STEDMA40_PERIPH_TO_MEM) ||
2403             (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH))
2404                 __d40_set_prio_rt(d40c, d40c->dma_cfg.src_dev_type, true);
2405
2406         if ((d40c->dma_cfg.dir ==  STEDMA40_MEM_TO_PERIPH) ||
2407             (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH))
2408                 __d40_set_prio_rt(d40c, d40c->dma_cfg.dst_dev_type, false);
2409 }
2410
2411 /* DMA ENGINE functions */
2412 static int d40_alloc_chan_resources(struct dma_chan *chan)
2413 {
2414         int err;
2415         unsigned long flags;
2416         struct d40_chan *d40c =
2417                 container_of(chan, struct d40_chan, chan);
2418         bool is_free_phy;
2419         spin_lock_irqsave(&d40c->lock, flags);
2420
2421         dma_cookie_init(chan);
2422
2423         /* If no dma configuration is set use default configuration (memcpy) */
2424         if (!d40c->configured) {
2425                 err = d40_config_memcpy(d40c);
2426                 if (err) {
2427                         chan_err(d40c, "Failed to configure memcpy channel\n");
2428                         goto fail;
2429                 }
2430         }
2431
2432         err = d40_allocate_channel(d40c, &is_free_phy);
2433         if (err) {
2434                 chan_err(d40c, "Failed to allocate channel\n");
2435                 d40c->configured = false;
2436                 goto fail;
2437         }
2438
2439         pm_runtime_get_sync(d40c->base->dev);
2440         /* Fill in basic CFG register values */
2441         d40_phy_cfg(&d40c->dma_cfg, &d40c->src_def_cfg,
2442                     &d40c->dst_def_cfg, chan_is_logical(d40c));
2443
2444         d40_set_prio_realtime(d40c);
2445
2446         if (chan_is_logical(d40c)) {
2447                 d40_log_cfg(&d40c->dma_cfg,
2448                             &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2449
2450                 if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM)
2451                         d40c->lcpa = d40c->base->lcpa_base +
2452                                 d40c->dma_cfg.src_dev_type * D40_LCPA_CHAN_SIZE;
2453                 else
2454                         d40c->lcpa = d40c->base->lcpa_base +
2455                                 d40c->dma_cfg.dst_dev_type *
2456                                 D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2457         }
2458
2459         dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2460                  chan_is_logical(d40c) ? "logical" : "physical",
2461                  d40c->phy_chan->num,
2462                  d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2463
2464
2465         /*
2466          * Only write channel configuration to the DMA if the physical
2467          * resource is free. In case of multiple logical channels
2468          * on the same physical resource, only the first write is necessary.
2469          */
2470         if (is_free_phy)
2471                 d40_config_write(d40c);
2472 fail:
2473         pm_runtime_mark_last_busy(d40c->base->dev);
2474         pm_runtime_put_autosuspend(d40c->base->dev);
2475         spin_unlock_irqrestore(&d40c->lock, flags);
2476         return err;
2477 }
2478
2479 static void d40_free_chan_resources(struct dma_chan *chan)
2480 {
2481         struct d40_chan *d40c =
2482                 container_of(chan, struct d40_chan, chan);
2483         int err;
2484         unsigned long flags;
2485
2486         if (d40c->phy_chan == NULL) {
2487                 chan_err(d40c, "Cannot free unallocated channel\n");
2488                 return;
2489         }
2490
2491         spin_lock_irqsave(&d40c->lock, flags);
2492
2493         err = d40_free_dma(d40c);
2494
2495         if (err)
2496                 chan_err(d40c, "Failed to free channel\n");
2497         spin_unlock_irqrestore(&d40c->lock, flags);
2498 }
2499
2500 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2501                                                        dma_addr_t dst,
2502                                                        dma_addr_t src,
2503                                                        size_t size,
2504                                                        unsigned long dma_flags)
2505 {
2506         struct scatterlist dst_sg;
2507         struct scatterlist src_sg;
2508
2509         sg_init_table(&dst_sg, 1);
2510         sg_init_table(&src_sg, 1);
2511
2512         sg_dma_address(&dst_sg) = dst;
2513         sg_dma_address(&src_sg) = src;
2514
2515         sg_dma_len(&dst_sg) = size;
2516         sg_dma_len(&src_sg) = size;
2517
2518         return d40_prep_sg(chan, &src_sg, &dst_sg, 1, DMA_NONE, dma_flags);
2519 }
2520
2521 static struct dma_async_tx_descriptor *
2522 d40_prep_memcpy_sg(struct dma_chan *chan,
2523                    struct scatterlist *dst_sg, unsigned int dst_nents,
2524                    struct scatterlist *src_sg, unsigned int src_nents,
2525                    unsigned long dma_flags)
2526 {
2527         if (dst_nents != src_nents)
2528                 return NULL;
2529
2530         return d40_prep_sg(chan, src_sg, dst_sg, src_nents, DMA_NONE, dma_flags);
2531 }
2532
2533 static struct dma_async_tx_descriptor *
2534 d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2535                   unsigned int sg_len, enum dma_transfer_direction direction,
2536                   unsigned long dma_flags, void *context)
2537 {
2538         if (!is_slave_direction(direction))
2539                 return NULL;
2540
2541         return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2542 }
2543
2544 static struct dma_async_tx_descriptor *
2545 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2546                      size_t buf_len, size_t period_len,
2547                      enum dma_transfer_direction direction, unsigned long flags,
2548                      void *context)
2549 {
2550         unsigned int periods = buf_len / period_len;
2551         struct dma_async_tx_descriptor *txd;
2552         struct scatterlist *sg;
2553         int i;
2554
2555         sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2556         for (i = 0; i < periods; i++) {
2557                 sg_dma_address(&sg[i]) = dma_addr;
2558                 sg_dma_len(&sg[i]) = period_len;
2559                 dma_addr += period_len;
2560         }
2561
2562         sg[periods].offset = 0;
2563         sg_dma_len(&sg[periods]) = 0;
2564         sg[periods].page_link =
2565                 ((unsigned long)sg | 0x01) & ~0x02;
2566
2567         txd = d40_prep_sg(chan, sg, sg, periods, direction,
2568                           DMA_PREP_INTERRUPT);
2569
2570         kfree(sg);
2571
2572         return txd;
2573 }
2574
2575 static enum dma_status d40_tx_status(struct dma_chan *chan,
2576                                      dma_cookie_t cookie,
2577                                      struct dma_tx_state *txstate)
2578 {
2579         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2580         enum dma_status ret;
2581
2582         if (d40c->phy_chan == NULL) {
2583                 chan_err(d40c, "Cannot read status of unallocated channel\n");
2584                 return -EINVAL;
2585         }
2586
2587         ret = dma_cookie_status(chan, cookie, txstate);
2588         if (ret != DMA_SUCCESS)
2589                 dma_set_residue(txstate, stedma40_residue(chan));
2590
2591         if (d40_is_paused(d40c))
2592                 ret = DMA_PAUSED;
2593
2594         return ret;
2595 }
2596
2597 static void d40_issue_pending(struct dma_chan *chan)
2598 {
2599         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2600         unsigned long flags;
2601
2602         if (d40c->phy_chan == NULL) {
2603                 chan_err(d40c, "Channel is not allocated!\n");
2604                 return;
2605         }
2606
2607         spin_lock_irqsave(&d40c->lock, flags);
2608
2609         list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2610
2611         /* Busy means that queued jobs are already being processed */
2612         if (!d40c->busy)
2613                 (void) d40_queue_start(d40c);
2614
2615         spin_unlock_irqrestore(&d40c->lock, flags);
2616 }
2617
2618 static void d40_terminate_all(struct dma_chan *chan)
2619 {
2620         unsigned long flags;
2621         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2622         int ret;
2623
2624         spin_lock_irqsave(&d40c->lock, flags);
2625
2626         pm_runtime_get_sync(d40c->base->dev);
2627         ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2628         if (ret)
2629                 chan_err(d40c, "Failed to stop channel\n");
2630
2631         d40_term_all(d40c);
2632         pm_runtime_mark_last_busy(d40c->base->dev);
2633         pm_runtime_put_autosuspend(d40c->base->dev);
2634         if (d40c->busy) {
2635                 pm_runtime_mark_last_busy(d40c->base->dev);
2636                 pm_runtime_put_autosuspend(d40c->base->dev);
2637         }
2638         d40c->busy = false;
2639
2640         spin_unlock_irqrestore(&d40c->lock, flags);
2641 }
2642
2643 static int
2644 dma40_config_to_halfchannel(struct d40_chan *d40c,
2645                             struct stedma40_half_channel_info *info,
2646                             enum dma_slave_buswidth width,
2647                             u32 maxburst)
2648 {
2649         enum stedma40_periph_data_width addr_width;
2650         int psize;
2651
2652         switch (width) {
2653         case DMA_SLAVE_BUSWIDTH_1_BYTE:
2654                 addr_width = STEDMA40_BYTE_WIDTH;
2655                 break;
2656         case DMA_SLAVE_BUSWIDTH_2_BYTES:
2657                 addr_width = STEDMA40_HALFWORD_WIDTH;
2658                 break;
2659         case DMA_SLAVE_BUSWIDTH_4_BYTES:
2660                 addr_width = STEDMA40_WORD_WIDTH;
2661                 break;
2662         case DMA_SLAVE_BUSWIDTH_8_BYTES:
2663                 addr_width = STEDMA40_DOUBLEWORD_WIDTH;
2664                 break;
2665         default:
2666                 dev_err(d40c->base->dev,
2667                         "illegal peripheral address width "
2668                         "requested (%d)\n",
2669                         width);
2670                 return -EINVAL;
2671         }
2672
2673         if (chan_is_logical(d40c)) {
2674                 if (maxburst >= 16)
2675                         psize = STEDMA40_PSIZE_LOG_16;
2676                 else if (maxburst >= 8)
2677                         psize = STEDMA40_PSIZE_LOG_8;
2678                 else if (maxburst >= 4)
2679                         psize = STEDMA40_PSIZE_LOG_4;
2680                 else
2681                         psize = STEDMA40_PSIZE_LOG_1;
2682         } else {
2683                 if (maxburst >= 16)
2684                         psize = STEDMA40_PSIZE_PHY_16;
2685                 else if (maxburst >= 8)
2686                         psize = STEDMA40_PSIZE_PHY_8;
2687                 else if (maxburst >= 4)
2688                         psize = STEDMA40_PSIZE_PHY_4;
2689                 else
2690                         psize = STEDMA40_PSIZE_PHY_1;
2691         }
2692
2693         info->data_width = addr_width;
2694         info->psize = psize;
2695         info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2696
2697         return 0;
2698 }
2699
2700 /* Runtime reconfiguration extension */
2701 static int d40_set_runtime_config(struct dma_chan *chan,
2702                                   struct dma_slave_config *config)
2703 {
2704         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2705         struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2706         enum dma_slave_buswidth src_addr_width, dst_addr_width;
2707         dma_addr_t config_addr;
2708         u32 src_maxburst, dst_maxburst;
2709         int ret;
2710
2711         src_addr_width = config->src_addr_width;
2712         src_maxburst = config->src_maxburst;
2713         dst_addr_width = config->dst_addr_width;
2714         dst_maxburst = config->dst_maxburst;
2715
2716         if (config->direction == DMA_DEV_TO_MEM) {
2717                 dma_addr_t dev_addr_rx =
2718                         d40c->base->plat_data->dev_rx[cfg->src_dev_type];
2719
2720                 config_addr = config->src_addr;
2721                 if (dev_addr_rx)
2722                         dev_dbg(d40c->base->dev,
2723                                 "channel has a pre-wired RX address %08x "
2724                                 "overriding with %08x\n",
2725                                 dev_addr_rx, config_addr);
2726                 if (cfg->dir != STEDMA40_PERIPH_TO_MEM)
2727                         dev_dbg(d40c->base->dev,
2728                                 "channel was not configured for peripheral "
2729                                 "to memory transfer (%d) overriding\n",
2730                                 cfg->dir);
2731                 cfg->dir = STEDMA40_PERIPH_TO_MEM;
2732
2733                 /* Configure the memory side */
2734                 if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2735                         dst_addr_width = src_addr_width;
2736                 if (dst_maxburst == 0)
2737                         dst_maxburst = src_maxburst;
2738
2739         } else if (config->direction == DMA_MEM_TO_DEV) {
2740                 dma_addr_t dev_addr_tx =
2741                         d40c->base->plat_data->dev_tx[cfg->dst_dev_type];
2742
2743                 config_addr = config->dst_addr;
2744                 if (dev_addr_tx)
2745                         dev_dbg(d40c->base->dev,
2746                                 "channel has a pre-wired TX address %08x "
2747                                 "overriding with %08x\n",
2748                                 dev_addr_tx, config_addr);
2749                 if (cfg->dir != STEDMA40_MEM_TO_PERIPH)
2750                         dev_dbg(d40c->base->dev,
2751                                 "channel was not configured for memory "
2752                                 "to peripheral transfer (%d) overriding\n",
2753                                 cfg->dir);
2754                 cfg->dir = STEDMA40_MEM_TO_PERIPH;
2755
2756                 /* Configure the memory side */
2757                 if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2758                         src_addr_width = dst_addr_width;
2759                 if (src_maxburst == 0)
2760                         src_maxburst = dst_maxburst;
2761         } else {
2762                 dev_err(d40c->base->dev,
2763                         "unrecognized channel direction %d\n",
2764                         config->direction);
2765                 return -EINVAL;
2766         }
2767
2768         if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2769                 dev_err(d40c->base->dev,
2770                         "src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2771                         src_maxburst,
2772                         src_addr_width,
2773                         dst_maxburst,
2774                         dst_addr_width);
2775                 return -EINVAL;
2776         }
2777
2778         if (src_maxburst > 16) {
2779                 src_maxburst = 16;
2780                 dst_maxburst = src_maxburst * src_addr_width / dst_addr_width;
2781         } else if (dst_maxburst > 16) {
2782                 dst_maxburst = 16;
2783                 src_maxburst = dst_maxburst * dst_addr_width / src_addr_width;
2784         }
2785
2786         ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2787                                           src_addr_width,
2788                                           src_maxburst);
2789         if (ret)
2790                 return ret;
2791
2792         ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2793                                           dst_addr_width,
2794                                           dst_maxburst);
2795         if (ret)
2796                 return ret;
2797
2798         /* Fill in register values */
2799         if (chan_is_logical(d40c))
2800                 d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2801         else
2802                 d40_phy_cfg(cfg, &d40c->src_def_cfg,
2803                             &d40c->dst_def_cfg, false);
2804
2805         /* These settings will take precedence later */
2806         d40c->runtime_addr = config_addr;
2807         d40c->runtime_direction = config->direction;
2808         dev_dbg(d40c->base->dev,
2809                 "configured channel %s for %s, data width %d/%d, "
2810                 "maxburst %d/%d elements, LE, no flow control\n",
2811                 dma_chan_name(chan),
2812                 (config->direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2813                 src_addr_width, dst_addr_width,
2814                 src_maxburst, dst_maxburst);
2815
2816         return 0;
2817 }
2818
2819 static int d40_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
2820                        unsigned long arg)
2821 {
2822         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2823
2824         if (d40c->phy_chan == NULL) {
2825                 chan_err(d40c, "Channel is not allocated!\n");
2826                 return -EINVAL;
2827         }
2828
2829         switch (cmd) {
2830         case DMA_TERMINATE_ALL:
2831                 d40_terminate_all(chan);
2832                 return 0;
2833         case DMA_PAUSE:
2834                 return d40_pause(d40c);
2835         case DMA_RESUME:
2836                 return d40_resume(d40c);
2837         case DMA_SLAVE_CONFIG:
2838                 return d40_set_runtime_config(chan,
2839                         (struct dma_slave_config *) arg);
2840         default:
2841                 break;
2842         }
2843
2844         /* Other commands are unimplemented */
2845         return -ENXIO;
2846 }
2847
2848 /* Initialization functions */
2849
2850 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2851                                  struct d40_chan *chans, int offset,
2852                                  int num_chans)
2853 {
2854         int i = 0;
2855         struct d40_chan *d40c;
2856
2857         INIT_LIST_HEAD(&dma->channels);
2858
2859         for (i = offset; i < offset + num_chans; i++) {
2860                 d40c = &chans[i];
2861                 d40c->base = base;
2862                 d40c->chan.device = dma;
2863
2864                 spin_lock_init(&d40c->lock);
2865
2866                 d40c->log_num = D40_PHY_CHAN;
2867
2868                 INIT_LIST_HEAD(&d40c->done);
2869                 INIT_LIST_HEAD(&d40c->active);
2870                 INIT_LIST_HEAD(&d40c->queue);
2871                 INIT_LIST_HEAD(&d40c->pending_queue);
2872                 INIT_LIST_HEAD(&d40c->client);
2873                 INIT_LIST_HEAD(&d40c->prepare_queue);
2874
2875                 tasklet_init(&d40c->tasklet, dma_tasklet,
2876                              (unsigned long) d40c);
2877
2878                 list_add_tail(&d40c->chan.device_node,
2879                               &dma->channels);
2880         }
2881 }
2882
2883 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2884 {
2885         if (dma_has_cap(DMA_SLAVE, dev->cap_mask))
2886                 dev->device_prep_slave_sg = d40_prep_slave_sg;
2887
2888         if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2889                 dev->device_prep_dma_memcpy = d40_prep_memcpy;
2890
2891                 /*
2892                  * This controller can only access address at even
2893                  * 32bit boundaries, i.e. 2^2
2894                  */
2895                 dev->copy_align = 2;
2896         }
2897
2898         if (dma_has_cap(DMA_SG, dev->cap_mask))
2899                 dev->device_prep_dma_sg = d40_prep_memcpy_sg;
2900
2901         if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2902                 dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2903
2904         dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2905         dev->device_free_chan_resources = d40_free_chan_resources;
2906         dev->device_issue_pending = d40_issue_pending;
2907         dev->device_tx_status = d40_tx_status;
2908         dev->device_control = d40_control;
2909         dev->dev = base->dev;
2910 }
2911
2912 static int __init d40_dmaengine_init(struct d40_base *base,
2913                                      int num_reserved_chans)
2914 {
2915         int err ;
2916
2917         d40_chan_init(base, &base->dma_slave, base->log_chans,
2918                       0, base->num_log_chans);
2919
2920         dma_cap_zero(base->dma_slave.cap_mask);
2921         dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2922         dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2923
2924         d40_ops_init(base, &base->dma_slave);
2925
2926         err = dma_async_device_register(&base->dma_slave);
2927
2928         if (err) {
2929                 d40_err(base->dev, "Failed to register slave channels\n");
2930                 goto failure1;
2931         }
2932
2933         d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2934                       base->num_log_chans, base->plat_data->memcpy_len);
2935
2936         dma_cap_zero(base->dma_memcpy.cap_mask);
2937         dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2938         dma_cap_set(DMA_SG, base->dma_memcpy.cap_mask);
2939
2940         d40_ops_init(base, &base->dma_memcpy);
2941
2942         err = dma_async_device_register(&base->dma_memcpy);
2943
2944         if (err) {
2945                 d40_err(base->dev,
2946                         "Failed to regsiter memcpy only channels\n");
2947                 goto failure2;
2948         }
2949
2950         d40_chan_init(base, &base->dma_both, base->phy_chans,
2951                       0, num_reserved_chans);
2952
2953         dma_cap_zero(base->dma_both.cap_mask);
2954         dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2955         dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2956         dma_cap_set(DMA_SG, base->dma_both.cap_mask);
2957         dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2958
2959         d40_ops_init(base, &base->dma_both);
2960         err = dma_async_device_register(&base->dma_both);
2961
2962         if (err) {
2963                 d40_err(base->dev,
2964                         "Failed to register logical and physical capable channels\n");
2965                 goto failure3;
2966         }
2967         return 0;
2968 failure3:
2969         dma_async_device_unregister(&base->dma_memcpy);
2970 failure2:
2971         dma_async_device_unregister(&base->dma_slave);
2972 failure1:
2973         return err;
2974 }
2975
2976 /* Suspend resume functionality */
2977 #ifdef CONFIG_PM
2978 static int dma40_pm_suspend(struct device *dev)
2979 {
2980         struct platform_device *pdev = to_platform_device(dev);
2981         struct d40_base *base = platform_get_drvdata(pdev);
2982         int ret = 0;
2983
2984         if (base->lcpa_regulator)
2985                 ret = regulator_disable(base->lcpa_regulator);
2986         return ret;
2987 }
2988
2989 static int dma40_runtime_suspend(struct device *dev)
2990 {
2991         struct platform_device *pdev = to_platform_device(dev);
2992         struct d40_base *base = platform_get_drvdata(pdev);
2993
2994         d40_save_restore_registers(base, true);
2995
2996         /* Don't disable/enable clocks for v1 due to HW bugs */
2997         if (base->rev != 1)
2998                 writel_relaxed(base->gcc_pwr_off_mask,
2999                                base->virtbase + D40_DREG_GCC);
3000
3001         return 0;
3002 }
3003
3004 static int dma40_runtime_resume(struct device *dev)
3005 {
3006         struct platform_device *pdev = to_platform_device(dev);
3007         struct d40_base *base = platform_get_drvdata(pdev);
3008
3009         if (base->initialized)
3010                 d40_save_restore_registers(base, false);
3011
3012         writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
3013                        base->virtbase + D40_DREG_GCC);
3014         return 0;
3015 }
3016
3017 static int dma40_resume(struct device *dev)
3018 {
3019         struct platform_device *pdev = to_platform_device(dev);
3020         struct d40_base *base = platform_get_drvdata(pdev);
3021         int ret = 0;
3022
3023         if (base->lcpa_regulator)
3024                 ret = regulator_enable(base->lcpa_regulator);
3025
3026         return ret;
3027 }
3028
3029 static const struct dev_pm_ops dma40_pm_ops = {
3030         .suspend                = dma40_pm_suspend,
3031         .runtime_suspend        = dma40_runtime_suspend,
3032         .runtime_resume         = dma40_runtime_resume,
3033         .resume                 = dma40_resume,
3034 };
3035 #define DMA40_PM_OPS    (&dma40_pm_ops)
3036 #else
3037 #define DMA40_PM_OPS    NULL
3038 #endif
3039
3040 /* Initialization functions. */
3041
3042 static int __init d40_phy_res_init(struct d40_base *base)
3043 {
3044         int i;
3045         int num_phy_chans_avail = 0;
3046         u32 val[2];
3047         int odd_even_bit = -2;
3048         int gcc = D40_DREG_GCC_ENA;
3049
3050         val[0] = readl(base->virtbase + D40_DREG_PRSME);
3051         val[1] = readl(base->virtbase + D40_DREG_PRSMO);
3052
3053         for (i = 0; i < base->num_phy_chans; i++) {
3054                 base->phy_res[i].num = i;
3055                 odd_even_bit += 2 * ((i % 2) == 0);
3056                 if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
3057                         /* Mark security only channels as occupied */
3058                         base->phy_res[i].allocated_src = D40_ALLOC_PHY;
3059                         base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
3060                         base->phy_res[i].reserved = true;
3061                         gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3062                                                        D40_DREG_GCC_SRC);
3063                         gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3064                                                        D40_DREG_GCC_DST);
3065
3066
3067                 } else {
3068                         base->phy_res[i].allocated_src = D40_ALLOC_FREE;
3069                         base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
3070                         base->phy_res[i].reserved = false;
3071                         num_phy_chans_avail++;
3072                 }
3073                 spin_lock_init(&base->phy_res[i].lock);
3074         }
3075
3076         /* Mark disabled channels as occupied */
3077         for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
3078                 int chan = base->plat_data->disabled_channels[i];
3079
3080                 base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
3081                 base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
3082                 base->phy_res[chan].reserved = true;
3083                 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3084                                                D40_DREG_GCC_SRC);
3085                 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3086                                                D40_DREG_GCC_DST);
3087                 num_phy_chans_avail--;
3088         }
3089
3090         /* Mark soft_lli channels */
3091         for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) {
3092                 int chan = base->plat_data->soft_lli_chans[i];
3093
3094                 base->phy_res[chan].use_soft_lli = true;
3095         }
3096
3097         dev_info(base->dev, "%d of %d physical DMA channels available\n",
3098                  num_phy_chans_avail, base->num_phy_chans);
3099
3100         /* Verify settings extended vs standard */
3101         val[0] = readl(base->virtbase + D40_DREG_PRTYP);
3102
3103         for (i = 0; i < base->num_phy_chans; i++) {
3104
3105                 if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
3106                     (val[0] & 0x3) != 1)
3107                         dev_info(base->dev,
3108                                  "[%s] INFO: channel %d is misconfigured (%d)\n",
3109                                  __func__, i, val[0] & 0x3);
3110
3111                 val[0] = val[0] >> 2;
3112         }
3113
3114         /*
3115          * To keep things simple, Enable all clocks initially.
3116          * The clocks will get managed later post channel allocation.
3117          * The clocks for the event lines on which reserved channels exists
3118          * are not managed here.
3119          */
3120         writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3121         base->gcc_pwr_off_mask = gcc;
3122
3123         return num_phy_chans_avail;
3124 }
3125
3126 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
3127 {
3128         struct stedma40_platform_data *plat_data;
3129         struct clk *clk = NULL;
3130         void __iomem *virtbase = NULL;
3131         struct resource *res = NULL;
3132         struct d40_base *base = NULL;
3133         int num_log_chans = 0;
3134         int num_phy_chans;
3135         int clk_ret = -EINVAL;
3136         int i;
3137         u32 pid;
3138         u32 cid;
3139         u8 rev;
3140
3141         clk = clk_get(&pdev->dev, NULL);
3142         if (IS_ERR(clk)) {
3143                 d40_err(&pdev->dev, "No matching clock found\n");
3144                 goto failure;
3145         }
3146
3147         clk_ret = clk_prepare_enable(clk);
3148         if (clk_ret) {
3149                 d40_err(&pdev->dev, "Failed to prepare/enable clock\n");
3150                 goto failure;
3151         }
3152
3153         /* Get IO for DMAC base address */
3154         res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
3155         if (!res)
3156                 goto failure;
3157
3158         if (request_mem_region(res->start, resource_size(res),
3159                                D40_NAME " I/O base") == NULL)
3160                 goto failure;
3161
3162         virtbase = ioremap(res->start, resource_size(res));
3163         if (!virtbase)
3164                 goto failure;
3165
3166         /* This is just a regular AMBA PrimeCell ID actually */
3167         for (pid = 0, i = 0; i < 4; i++)
3168                 pid |= (readl(virtbase + resource_size(res) - 0x20 + 4 * i)
3169                         & 255) << (i * 8);
3170         for (cid = 0, i = 0; i < 4; i++)
3171                 cid |= (readl(virtbase + resource_size(res) - 0x10 + 4 * i)
3172                         & 255) << (i * 8);
3173
3174         if (cid != AMBA_CID) {
3175                 d40_err(&pdev->dev, "Unknown hardware! No PrimeCell ID\n");
3176                 goto failure;
3177         }
3178         if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
3179                 d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
3180                         AMBA_MANF_BITS(pid),
3181                         AMBA_VENDOR_ST);
3182                 goto failure;
3183         }
3184         /*
3185          * HW revision:
3186          * DB8500ed has revision 0
3187          * ? has revision 1
3188          * DB8500v1 has revision 2
3189          * DB8500v2 has revision 3
3190          * AP9540v1 has revision 4
3191          * DB8540v1 has revision 4
3192          */
3193         rev = AMBA_REV_BITS(pid);
3194
3195         plat_data = pdev->dev.platform_data;
3196
3197         /* The number of physical channels on this HW */
3198         if (plat_data->num_of_phy_chans)
3199                 num_phy_chans = plat_data->num_of_phy_chans;
3200         else
3201                 num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
3202
3203         dev_info(&pdev->dev, "hardware revision: %d @ 0x%x with %d physical channels\n",
3204                  rev, res->start, num_phy_chans);
3205
3206         if (rev < 2) {
3207                 d40_err(&pdev->dev, "hardware revision: %d is not supported",
3208                         rev);
3209                 goto failure;
3210         }
3211
3212         /* Count the number of logical channels in use */
3213         for (i = 0; i < plat_data->dev_len; i++)
3214                 if (plat_data->dev_rx[i] != 0)
3215                         num_log_chans++;
3216
3217         for (i = 0; i < plat_data->dev_len; i++)
3218                 if (plat_data->dev_tx[i] != 0)
3219                         num_log_chans++;
3220
3221         base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
3222                        (num_phy_chans + num_log_chans + plat_data->memcpy_len) *
3223                        sizeof(struct d40_chan), GFP_KERNEL);
3224
3225         if (base == NULL) {
3226                 d40_err(&pdev->dev, "Out of memory\n");
3227                 goto failure;
3228         }
3229
3230         base->rev = rev;
3231         base->clk = clk;
3232         base->num_phy_chans = num_phy_chans;
3233         base->num_log_chans = num_log_chans;
3234         base->phy_start = res->start;
3235         base->phy_size = resource_size(res);
3236         base->virtbase = virtbase;
3237         base->plat_data = plat_data;
3238         base->dev = &pdev->dev;
3239         base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3240         base->log_chans = &base->phy_chans[num_phy_chans];
3241
3242         if (base->plat_data->num_of_phy_chans == 14) {
3243                 base->gen_dmac.backup = d40_backup_regs_v4b;
3244                 base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
3245                 base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
3246                 base->gen_dmac.interrupt_clear = D40_DREG_CPCICR;
3247                 base->gen_dmac.realtime_en = D40_DREG_CRSEG1;
3248                 base->gen_dmac.realtime_clear = D40_DREG_CRCEG1;
3249                 base->gen_dmac.high_prio_en = D40_DREG_CPSEG1;
3250                 base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1;
3251                 base->gen_dmac.il = il_v4b;
3252                 base->gen_dmac.il_size = ARRAY_SIZE(il_v4b);
3253                 base->gen_dmac.init_reg = dma_init_reg_v4b;
3254                 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
3255         } else {
3256                 if (base->rev >= 3) {
3257                         base->gen_dmac.backup = d40_backup_regs_v4a;
3258                         base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
3259                 }
3260                 base->gen_dmac.interrupt_en = D40_DREG_PCMIS;
3261                 base->gen_dmac.interrupt_clear = D40_DREG_PCICR;
3262                 base->gen_dmac.realtime_en = D40_DREG_RSEG1;
3263                 base->gen_dmac.realtime_clear = D40_DREG_RCEG1;
3264                 base->gen_dmac.high_prio_en = D40_DREG_PSEG1;
3265                 base->gen_dmac.high_prio_clear = D40_DREG_PCEG1;
3266                 base->gen_dmac.il = il_v4a;
3267                 base->gen_dmac.il_size = ARRAY_SIZE(il_v4a);
3268                 base->gen_dmac.init_reg = dma_init_reg_v4a;
3269                 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a);
3270         }
3271
3272         base->phy_res = kzalloc(num_phy_chans * sizeof(struct d40_phy_res),
3273                                 GFP_KERNEL);
3274         if (!base->phy_res)
3275                 goto failure;
3276
3277         base->lookup_phy_chans = kzalloc(num_phy_chans *
3278                                          sizeof(struct d40_chan *),
3279                                          GFP_KERNEL);
3280         if (!base->lookup_phy_chans)
3281                 goto failure;
3282
3283         if (num_log_chans + plat_data->memcpy_len) {
3284                 /*
3285                  * The max number of logical channels are event lines for all
3286                  * src devices and dst devices
3287                  */
3288                 base->lookup_log_chans = kzalloc(plat_data->dev_len * 2 *
3289                                                  sizeof(struct d40_chan *),
3290                                                  GFP_KERNEL);
3291                 if (!base->lookup_log_chans)
3292                         goto failure;
3293         }
3294
3295         base->reg_val_backup_chan = kmalloc(base->num_phy_chans *
3296                                             sizeof(d40_backup_regs_chan),
3297                                             GFP_KERNEL);
3298         if (!base->reg_val_backup_chan)
3299                 goto failure;
3300
3301         base->lcla_pool.alloc_map =
3302                 kzalloc(num_phy_chans * sizeof(struct d40_desc *)
3303                         * D40_LCLA_LINK_PER_EVENT_GRP, GFP_KERNEL);
3304         if (!base->lcla_pool.alloc_map)
3305                 goto failure;
3306
3307         base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3308                                             0, SLAB_HWCACHE_ALIGN,
3309                                             NULL);
3310         if (base->desc_slab == NULL)
3311                 goto failure;
3312
3313         return base;
3314
3315 failure:
3316         if (!clk_ret)
3317                 clk_disable_unprepare(clk);
3318         if (!IS_ERR(clk))
3319                 clk_put(clk);
3320         if (virtbase)
3321                 iounmap(virtbase);
3322         if (res)
3323                 release_mem_region(res->start,
3324                                    resource_size(res));
3325         if (virtbase)
3326                 iounmap(virtbase);
3327
3328         if (base) {
3329                 kfree(base->lcla_pool.alloc_map);
3330                 kfree(base->reg_val_backup_chan);
3331                 kfree(base->lookup_log_chans);
3332                 kfree(base->lookup_phy_chans);
3333                 kfree(base->phy_res);
3334                 kfree(base);
3335         }
3336
3337         return NULL;
3338 }
3339
3340 static void __init d40_hw_init(struct d40_base *base)
3341 {
3342
3343         int i;
3344         u32 prmseo[2] = {0, 0};
3345         u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3346         u32 pcmis = 0;
3347         u32 pcicr = 0;
3348         struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg;
3349         u32 reg_size = base->gen_dmac.init_reg_size;
3350
3351         for (i = 0; i < reg_size; i++)
3352                 writel(dma_init_reg[i].val,
3353                        base->virtbase + dma_init_reg[i].reg);
3354
3355         /* Configure all our dma channels to default settings */
3356         for (i = 0; i < base->num_phy_chans; i++) {
3357
3358                 activeo[i % 2] = activeo[i % 2] << 2;
3359
3360                 if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3361                     == D40_ALLOC_PHY) {
3362                         activeo[i % 2] |= 3;
3363                         continue;
3364                 }
3365
3366                 /* Enable interrupt # */
3367                 pcmis = (pcmis << 1) | 1;
3368
3369                 /* Clear interrupt # */
3370                 pcicr = (pcicr << 1) | 1;
3371
3372                 /* Set channel to physical mode */
3373                 prmseo[i % 2] = prmseo[i % 2] << 2;
3374                 prmseo[i % 2] |= 1;
3375
3376         }
3377
3378         writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3379         writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3380         writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3381         writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3382
3383         /* Write which interrupt to enable */
3384         writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en);
3385
3386         /* Write which interrupt to clear */
3387         writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear);
3388
3389         /* These are __initdata and cannot be accessed after init */
3390         base->gen_dmac.init_reg = NULL;
3391         base->gen_dmac.init_reg_size = 0;
3392 }
3393
3394 static int __init d40_lcla_allocate(struct d40_base *base)
3395 {
3396         struct d40_lcla_pool *pool = &base->lcla_pool;
3397         unsigned long *page_list;
3398         int i, j;
3399         int ret = 0;
3400
3401         /*
3402          * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3403          * To full fill this hardware requirement without wasting 256 kb
3404          * we allocate pages until we get an aligned one.
3405          */
3406         page_list = kmalloc(sizeof(unsigned long) * MAX_LCLA_ALLOC_ATTEMPTS,
3407                             GFP_KERNEL);
3408
3409         if (!page_list) {
3410                 ret = -ENOMEM;
3411                 goto failure;
3412         }
3413
3414         /* Calculating how many pages that are required */
3415         base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3416
3417         for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3418                 page_list[i] = __get_free_pages(GFP_KERNEL,
3419                                                 base->lcla_pool.pages);
3420                 if (!page_list[i]) {
3421
3422                         d40_err(base->dev, "Failed to allocate %d pages.\n",
3423                                 base->lcla_pool.pages);
3424
3425                         for (j = 0; j < i; j++)
3426                                 free_pages(page_list[j], base->lcla_pool.pages);
3427                         goto failure;
3428                 }
3429
3430                 if ((virt_to_phys((void *)page_list[i]) &
3431                      (LCLA_ALIGNMENT - 1)) == 0)
3432                         break;
3433         }
3434
3435         for (j = 0; j < i; j++)
3436                 free_pages(page_list[j], base->lcla_pool.pages);
3437
3438         if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3439                 base->lcla_pool.base = (void *)page_list[i];
3440         } else {
3441                 /*
3442                  * After many attempts and no succees with finding the correct
3443                  * alignment, try with allocating a big buffer.
3444                  */
3445                 dev_warn(base->dev,
3446                          "[%s] Failed to get %d pages @ 18 bit align.\n",
3447                          __func__, base->lcla_pool.pages);
3448                 base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3449                                                          base->num_phy_chans +
3450                                                          LCLA_ALIGNMENT,
3451                                                          GFP_KERNEL);
3452                 if (!base->lcla_pool.base_unaligned) {
3453                         ret = -ENOMEM;
3454                         goto failure;
3455                 }
3456
3457                 base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3458                                                  LCLA_ALIGNMENT);
3459         }
3460
3461         pool->dma_addr = dma_map_single(base->dev, pool->base,
3462                                         SZ_1K * base->num_phy_chans,
3463                                         DMA_TO_DEVICE);
3464         if (dma_mapping_error(base->dev, pool->dma_addr)) {
3465                 pool->dma_addr = 0;
3466                 ret = -ENOMEM;
3467                 goto failure;
3468         }
3469
3470         writel(virt_to_phys(base->lcla_pool.base),
3471                base->virtbase + D40_DREG_LCLA);
3472 failure:
3473         kfree(page_list);
3474         return ret;
3475 }
3476
3477 static int __init d40_probe(struct platform_device *pdev)
3478 {
3479         int err;
3480         int ret = -ENOENT;
3481         struct d40_base *base;
3482         struct resource *res = NULL;
3483         int num_reserved_chans;
3484         u32 val;
3485
3486         base = d40_hw_detect_init(pdev);
3487
3488         if (!base)
3489                 goto failure;
3490
3491         num_reserved_chans = d40_phy_res_init(base);
3492
3493         platform_set_drvdata(pdev, base);
3494
3495         spin_lock_init(&base->interrupt_lock);
3496         spin_lock_init(&base->execmd_lock);
3497
3498         /* Get IO for logical channel parameter address */
3499         res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
3500         if (!res) {
3501                 ret = -ENOENT;
3502                 d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
3503                 goto failure;
3504         }
3505         base->lcpa_size = resource_size(res);
3506         base->phy_lcpa = res->start;
3507
3508         if (request_mem_region(res->start, resource_size(res),
3509                                D40_NAME " I/O lcpa") == NULL) {
3510                 ret = -EBUSY;
3511                 d40_err(&pdev->dev,
3512                         "Failed to request LCPA region 0x%x-0x%x\n",
3513                         res->start, res->end);
3514                 goto failure;
3515         }
3516
3517         /* We make use of ESRAM memory for this. */
3518         val = readl(base->virtbase + D40_DREG_LCPA);
3519         if (res->start != val && val != 0) {
3520                 dev_warn(&pdev->dev,
3521                          "[%s] Mismatch LCPA dma 0x%x, def 0x%x\n",
3522                          __func__, val, res->start);
3523         } else
3524                 writel(res->start, base->virtbase + D40_DREG_LCPA);
3525
3526         base->lcpa_base = ioremap(res->start, resource_size(res));
3527         if (!base->lcpa_base) {
3528                 ret = -ENOMEM;
3529                 d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
3530                 goto failure;
3531         }
3532         /* If lcla has to be located in ESRAM we don't need to allocate */
3533         if (base->plat_data->use_esram_lcla) {
3534                 res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3535                                                         "lcla_esram");
3536                 if (!res) {
3537                         ret = -ENOENT;
3538                         d40_err(&pdev->dev,
3539                                 "No \"lcla_esram\" memory resource\n");
3540                         goto failure;
3541                 }
3542                 base->lcla_pool.base = ioremap(res->start,
3543                                                 resource_size(res));
3544                 if (!base->lcla_pool.base) {
3545                         ret = -ENOMEM;
3546                         d40_err(&pdev->dev, "Failed to ioremap LCLA region\n");
3547                         goto failure;
3548                 }
3549                 writel(res->start, base->virtbase + D40_DREG_LCLA);
3550
3551         } else {
3552                 ret = d40_lcla_allocate(base);
3553                 if (ret) {
3554                         d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
3555                         goto failure;
3556                 }
3557         }
3558
3559         spin_lock_init(&base->lcla_pool.lock);
3560
3561         base->irq = platform_get_irq(pdev, 0);
3562
3563         ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3564         if (ret) {
3565                 d40_err(&pdev->dev, "No IRQ defined\n");
3566                 goto failure;
3567         }
3568
3569         pm_runtime_irq_safe(base->dev);
3570         pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3571         pm_runtime_use_autosuspend(base->dev);
3572         pm_runtime_enable(base->dev);
3573         pm_runtime_resume(base->dev);
3574
3575         if (base->plat_data->use_esram_lcla) {
3576
3577                 base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3578                 if (IS_ERR(base->lcpa_regulator)) {
3579                         d40_err(&pdev->dev, "Failed to get lcpa_regulator\n");
3580                         base->lcpa_regulator = NULL;
3581                         goto failure;
3582                 }
3583
3584                 ret = regulator_enable(base->lcpa_regulator);
3585                 if (ret) {
3586                         d40_err(&pdev->dev,
3587                                 "Failed to enable lcpa_regulator\n");
3588                         regulator_put(base->lcpa_regulator);
3589                         base->lcpa_regulator = NULL;
3590                         goto failure;
3591                 }
3592         }
3593
3594         base->initialized = true;
3595         err = d40_dmaengine_init(base, num_reserved_chans);
3596         if (err)
3597                 goto failure;
3598
3599         base->dev->dma_parms = &base->dma_parms;
3600         err = dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
3601         if (err) {
3602                 d40_err(&pdev->dev, "Failed to set dma max seg size\n");
3603                 goto failure;
3604         }
3605
3606         d40_hw_init(base);
3607
3608         dev_info(base->dev, "initialized\n");
3609         return 0;
3610
3611 failure:
3612         if (base) {
3613                 if (base->desc_slab)
3614                         kmem_cache_destroy(base->desc_slab);
3615                 if (base->virtbase)
3616                         iounmap(base->virtbase);
3617
3618                 if (base->lcla_pool.base && base->plat_data->use_esram_lcla) {
3619                         iounmap(base->lcla_pool.base);
3620                         base->lcla_pool.base = NULL;
3621                 }
3622
3623                 if (base->lcla_pool.dma_addr)
3624                         dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3625                                          SZ_1K * base->num_phy_chans,
3626                                          DMA_TO_DEVICE);
3627
3628                 if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3629                         free_pages((unsigned long)base->lcla_pool.base,
3630                                    base->lcla_pool.pages);
3631
3632                 kfree(base->lcla_pool.base_unaligned);
3633
3634                 if (base->phy_lcpa)
3635                         release_mem_region(base->phy_lcpa,
3636                                            base->lcpa_size);
3637                 if (base->phy_start)
3638                         release_mem_region(base->phy_start,
3639                                            base->phy_size);
3640                 if (base->clk) {
3641                         clk_disable_unprepare(base->clk);
3642                         clk_put(base->clk);
3643                 }
3644
3645                 if (base->lcpa_regulator) {
3646                         regulator_disable(base->lcpa_regulator);
3647                         regulator_put(base->lcpa_regulator);
3648                 }
3649
3650                 kfree(base->lcla_pool.alloc_map);
3651                 kfree(base->lookup_log_chans);
3652                 kfree(base->lookup_phy_chans);
3653                 kfree(base->phy_res);
3654                 kfree(base);
3655         }
3656
3657         d40_err(&pdev->dev, "probe failed\n");
3658         return ret;
3659 }
3660
3661 static struct platform_driver d40_driver = {
3662         .driver = {
3663                 .owner = THIS_MODULE,
3664                 .name  = D40_NAME,
3665                 .pm = DMA40_PM_OPS,
3666         },
3667 };
3668
3669 static int __init stedma40_init(void)
3670 {
3671         return platform_driver_probe(&d40_driver, d40_probe);
3672 }
3673 subsys_initcall(stedma40_init);