Merge branch 'for-3.9' of git://linux-nfs.org/~bfields/linux
[platform/adaptation/renesas_rcar/renesas_kernel.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                 pm_runtime_mark_last_busy(d40c->base->dev);
1572                 pm_runtime_put_autosuspend(d40c->base->dev);
1573
1574                 d40_desc_remove(d40d);
1575                 d40_desc_done(d40c, d40d);
1576         }
1577
1578         d40c->pending_tx++;
1579         tasklet_schedule(&d40c->tasklet);
1580
1581 }
1582
1583 static void dma_tasklet(unsigned long data)
1584 {
1585         struct d40_chan *d40c = (struct d40_chan *) data;
1586         struct d40_desc *d40d;
1587         unsigned long flags;
1588         dma_async_tx_callback callback;
1589         void *callback_param;
1590
1591         spin_lock_irqsave(&d40c->lock, flags);
1592
1593         /* Get first entry from the done list */
1594         d40d = d40_first_done(d40c);
1595         if (d40d == NULL) {
1596                 /* Check if we have reached here for cyclic job */
1597                 d40d = d40_first_active_get(d40c);
1598                 if (d40d == NULL || !d40d->cyclic)
1599                         goto err;
1600         }
1601
1602         if (!d40d->cyclic)
1603                 dma_cookie_complete(&d40d->txd);
1604
1605         /*
1606          * If terminating a channel pending_tx is set to zero.
1607          * This prevents any finished active jobs to return to the client.
1608          */
1609         if (d40c->pending_tx == 0) {
1610                 spin_unlock_irqrestore(&d40c->lock, flags);
1611                 return;
1612         }
1613
1614         /* Callback to client */
1615         callback = d40d->txd.callback;
1616         callback_param = d40d->txd.callback_param;
1617
1618         if (!d40d->cyclic) {
1619                 if (async_tx_test_ack(&d40d->txd)) {
1620                         d40_desc_remove(d40d);
1621                         d40_desc_free(d40c, d40d);
1622                 } else if (!d40d->is_in_client_list) {
1623                         d40_desc_remove(d40d);
1624                         d40_lcla_free_all(d40c, d40d);
1625                         list_add_tail(&d40d->node, &d40c->client);
1626                         d40d->is_in_client_list = true;
1627                 }
1628         }
1629
1630         d40c->pending_tx--;
1631
1632         if (d40c->pending_tx)
1633                 tasklet_schedule(&d40c->tasklet);
1634
1635         spin_unlock_irqrestore(&d40c->lock, flags);
1636
1637         if (callback && (d40d->txd.flags & DMA_PREP_INTERRUPT))
1638                 callback(callback_param);
1639
1640         return;
1641
1642 err:
1643         /* Rescue manouver if receiving double interrupts */
1644         if (d40c->pending_tx > 0)
1645                 d40c->pending_tx--;
1646         spin_unlock_irqrestore(&d40c->lock, flags);
1647 }
1648
1649 static irqreturn_t d40_handle_interrupt(int irq, void *data)
1650 {
1651         int i;
1652         u32 idx;
1653         u32 row;
1654         long chan = -1;
1655         struct d40_chan *d40c;
1656         unsigned long flags;
1657         struct d40_base *base = data;
1658         u32 regs[base->gen_dmac.il_size];
1659         struct d40_interrupt_lookup *il = base->gen_dmac.il;
1660         u32 il_size = base->gen_dmac.il_size;
1661
1662         spin_lock_irqsave(&base->interrupt_lock, flags);
1663
1664         /* Read interrupt status of both logical and physical channels */
1665         for (i = 0; i < il_size; i++)
1666                 regs[i] = readl(base->virtbase + il[i].src);
1667
1668         for (;;) {
1669
1670                 chan = find_next_bit((unsigned long *)regs,
1671                                      BITS_PER_LONG * il_size, chan + 1);
1672
1673                 /* No more set bits found? */
1674                 if (chan == BITS_PER_LONG * il_size)
1675                         break;
1676
1677                 row = chan / BITS_PER_LONG;
1678                 idx = chan & (BITS_PER_LONG - 1);
1679
1680                 if (il[row].offset == D40_PHY_CHAN)
1681                         d40c = base->lookup_phy_chans[idx];
1682                 else
1683                         d40c = base->lookup_log_chans[il[row].offset + idx];
1684
1685                 if (!d40c) {
1686                         /*
1687                          * No error because this can happen if something else
1688                          * in the system is using the channel.
1689                          */
1690                         continue;
1691                 }
1692
1693                 /* ACK interrupt */
1694                 writel(1 << idx, base->virtbase + il[row].clr);
1695
1696                 spin_lock(&d40c->lock);
1697
1698                 if (!il[row].is_error)
1699                         dma_tc_handle(d40c);
1700                 else
1701                         d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1702                                 chan, il[row].offset, idx);
1703
1704                 spin_unlock(&d40c->lock);
1705         }
1706
1707         spin_unlock_irqrestore(&base->interrupt_lock, flags);
1708
1709         return IRQ_HANDLED;
1710 }
1711
1712 static int d40_validate_conf(struct d40_chan *d40c,
1713                              struct stedma40_chan_cfg *conf)
1714 {
1715         int res = 0;
1716         u32 dst_event_group = D40_TYPE_TO_GROUP(conf->dst_dev_type);
1717         u32 src_event_group = D40_TYPE_TO_GROUP(conf->src_dev_type);
1718         bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1719
1720         if (!conf->dir) {
1721                 chan_err(d40c, "Invalid direction.\n");
1722                 res = -EINVAL;
1723         }
1724
1725         if (conf->dst_dev_type != STEDMA40_DEV_DST_MEMORY &&
1726             d40c->base->plat_data->dev_tx[conf->dst_dev_type] == 0 &&
1727             d40c->runtime_addr == 0) {
1728
1729                 chan_err(d40c, "Invalid TX channel address (%d)\n",
1730                          conf->dst_dev_type);
1731                 res = -EINVAL;
1732         }
1733
1734         if (conf->src_dev_type != STEDMA40_DEV_SRC_MEMORY &&
1735             d40c->base->plat_data->dev_rx[conf->src_dev_type] == 0 &&
1736             d40c->runtime_addr == 0) {
1737                 chan_err(d40c, "Invalid RX channel address (%d)\n",
1738                         conf->src_dev_type);
1739                 res = -EINVAL;
1740         }
1741
1742         if (conf->dir == STEDMA40_MEM_TO_PERIPH &&
1743             dst_event_group == STEDMA40_DEV_DST_MEMORY) {
1744                 chan_err(d40c, "Invalid dst\n");
1745                 res = -EINVAL;
1746         }
1747
1748         if (conf->dir == STEDMA40_PERIPH_TO_MEM &&
1749             src_event_group == STEDMA40_DEV_SRC_MEMORY) {
1750                 chan_err(d40c, "Invalid src\n");
1751                 res = -EINVAL;
1752         }
1753
1754         if (src_event_group == STEDMA40_DEV_SRC_MEMORY &&
1755             dst_event_group == STEDMA40_DEV_DST_MEMORY && is_log) {
1756                 chan_err(d40c, "No event line\n");
1757                 res = -EINVAL;
1758         }
1759
1760         if (conf->dir == STEDMA40_PERIPH_TO_PERIPH &&
1761             (src_event_group != dst_event_group)) {
1762                 chan_err(d40c, "Invalid event group\n");
1763                 res = -EINVAL;
1764         }
1765
1766         if (conf->dir == STEDMA40_PERIPH_TO_PERIPH) {
1767                 /*
1768                  * DMAC HW supports it. Will be added to this driver,
1769                  * in case any dma client requires it.
1770                  */
1771                 chan_err(d40c, "periph to periph not supported\n");
1772                 res = -EINVAL;
1773         }
1774
1775         if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1776             (1 << conf->src_info.data_width) !=
1777             d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1778             (1 << conf->dst_info.data_width)) {
1779                 /*
1780                  * The DMAC hardware only supports
1781                  * src (burst x width) == dst (burst x width)
1782                  */
1783
1784                 chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1785                 res = -EINVAL;
1786         }
1787
1788         return res;
1789 }
1790
1791 static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1792                                bool is_src, int log_event_line, bool is_log,
1793                                bool *first_user)
1794 {
1795         unsigned long flags;
1796         spin_lock_irqsave(&phy->lock, flags);
1797
1798         *first_user = ((phy->allocated_src | phy->allocated_dst)
1799                         == D40_ALLOC_FREE);
1800
1801         if (!is_log) {
1802                 /* Physical interrupts are masked per physical full channel */
1803                 if (phy->allocated_src == D40_ALLOC_FREE &&
1804                     phy->allocated_dst == D40_ALLOC_FREE) {
1805                         phy->allocated_dst = D40_ALLOC_PHY;
1806                         phy->allocated_src = D40_ALLOC_PHY;
1807                         goto found;
1808                 } else
1809                         goto not_found;
1810         }
1811
1812         /* Logical channel */
1813         if (is_src) {
1814                 if (phy->allocated_src == D40_ALLOC_PHY)
1815                         goto not_found;
1816
1817                 if (phy->allocated_src == D40_ALLOC_FREE)
1818                         phy->allocated_src = D40_ALLOC_LOG_FREE;
1819
1820                 if (!(phy->allocated_src & (1 << log_event_line))) {
1821                         phy->allocated_src |= 1 << log_event_line;
1822                         goto found;
1823                 } else
1824                         goto not_found;
1825         } else {
1826                 if (phy->allocated_dst == D40_ALLOC_PHY)
1827                         goto not_found;
1828
1829                 if (phy->allocated_dst == D40_ALLOC_FREE)
1830                         phy->allocated_dst = D40_ALLOC_LOG_FREE;
1831
1832                 if (!(phy->allocated_dst & (1 << log_event_line))) {
1833                         phy->allocated_dst |= 1 << log_event_line;
1834                         goto found;
1835                 } else
1836                         goto not_found;
1837         }
1838
1839 not_found:
1840         spin_unlock_irqrestore(&phy->lock, flags);
1841         return false;
1842 found:
1843         spin_unlock_irqrestore(&phy->lock, flags);
1844         return true;
1845 }
1846
1847 static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1848                                int log_event_line)
1849 {
1850         unsigned long flags;
1851         bool is_free = false;
1852
1853         spin_lock_irqsave(&phy->lock, flags);
1854         if (!log_event_line) {
1855                 phy->allocated_dst = D40_ALLOC_FREE;
1856                 phy->allocated_src = D40_ALLOC_FREE;
1857                 is_free = true;
1858                 goto out;
1859         }
1860
1861         /* Logical channel */
1862         if (is_src) {
1863                 phy->allocated_src &= ~(1 << log_event_line);
1864                 if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1865                         phy->allocated_src = D40_ALLOC_FREE;
1866         } else {
1867                 phy->allocated_dst &= ~(1 << log_event_line);
1868                 if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1869                         phy->allocated_dst = D40_ALLOC_FREE;
1870         }
1871
1872         is_free = ((phy->allocated_src | phy->allocated_dst) ==
1873                    D40_ALLOC_FREE);
1874
1875 out:
1876         spin_unlock_irqrestore(&phy->lock, flags);
1877
1878         return is_free;
1879 }
1880
1881 static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1882 {
1883         int dev_type;
1884         int event_group;
1885         int event_line;
1886         struct d40_phy_res *phys;
1887         int i;
1888         int j;
1889         int log_num;
1890         int num_phy_chans;
1891         bool is_src;
1892         bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1893
1894         phys = d40c->base->phy_res;
1895         num_phy_chans = d40c->base->num_phy_chans;
1896
1897         if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1898                 dev_type = d40c->dma_cfg.src_dev_type;
1899                 log_num = 2 * dev_type;
1900                 is_src = true;
1901         } else if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1902                    d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1903                 /* dst event lines are used for logical memcpy */
1904                 dev_type = d40c->dma_cfg.dst_dev_type;
1905                 log_num = 2 * dev_type + 1;
1906                 is_src = false;
1907         } else
1908                 return -EINVAL;
1909
1910         event_group = D40_TYPE_TO_GROUP(dev_type);
1911         event_line = D40_TYPE_TO_EVENT(dev_type);
1912
1913         if (!is_log) {
1914                 if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1915                         /* Find physical half channel */
1916                         if (d40c->dma_cfg.use_fixed_channel) {
1917                                 i = d40c->dma_cfg.phy_channel;
1918                                 if (d40_alloc_mask_set(&phys[i], is_src,
1919                                                        0, is_log,
1920                                                        first_phy_user))
1921                                         goto found_phy;
1922                         } else {
1923                                 for (i = 0; i < num_phy_chans; i++) {
1924                                         if (d40_alloc_mask_set(&phys[i], is_src,
1925                                                        0, is_log,
1926                                                        first_phy_user))
1927                                                 goto found_phy;
1928                                 }
1929                         }
1930                 } else
1931                         for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1932                                 int phy_num = j  + event_group * 2;
1933                                 for (i = phy_num; i < phy_num + 2; i++) {
1934                                         if (d40_alloc_mask_set(&phys[i],
1935                                                                is_src,
1936                                                                0,
1937                                                                is_log,
1938                                                                first_phy_user))
1939                                                 goto found_phy;
1940                                 }
1941                         }
1942                 return -EINVAL;
1943 found_phy:
1944                 d40c->phy_chan = &phys[i];
1945                 d40c->log_num = D40_PHY_CHAN;
1946                 goto out;
1947         }
1948         if (dev_type == -1)
1949                 return -EINVAL;
1950
1951         /* Find logical channel */
1952         for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1953                 int phy_num = j + event_group * 2;
1954
1955                 if (d40c->dma_cfg.use_fixed_channel) {
1956                         i = d40c->dma_cfg.phy_channel;
1957
1958                         if ((i != phy_num) && (i != phy_num + 1)) {
1959                                 dev_err(chan2dev(d40c),
1960                                         "invalid fixed phy channel %d\n", i);
1961                                 return -EINVAL;
1962                         }
1963
1964                         if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1965                                                is_log, first_phy_user))
1966                                 goto found_log;
1967
1968                         dev_err(chan2dev(d40c),
1969                                 "could not allocate fixed phy channel %d\n", i);
1970                         return -EINVAL;
1971                 }
1972
1973                 /*
1974                  * Spread logical channels across all available physical rather
1975                  * than pack every logical channel at the first available phy
1976                  * channels.
1977                  */
1978                 if (is_src) {
1979                         for (i = phy_num; i < phy_num + 2; i++) {
1980                                 if (d40_alloc_mask_set(&phys[i], is_src,
1981                                                        event_line, is_log,
1982                                                        first_phy_user))
1983                                         goto found_log;
1984                         }
1985                 } else {
1986                         for (i = phy_num + 1; i >= phy_num; i--) {
1987                                 if (d40_alloc_mask_set(&phys[i], is_src,
1988                                                        event_line, is_log,
1989                                                        first_phy_user))
1990                                         goto found_log;
1991                         }
1992                 }
1993         }
1994         return -EINVAL;
1995
1996 found_log:
1997         d40c->phy_chan = &phys[i];
1998         d40c->log_num = log_num;
1999 out:
2000
2001         if (is_log)
2002                 d40c->base->lookup_log_chans[d40c->log_num] = d40c;
2003         else
2004                 d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
2005
2006         return 0;
2007
2008 }
2009
2010 static int d40_config_memcpy(struct d40_chan *d40c)
2011 {
2012         dma_cap_mask_t cap = d40c->chan.device->cap_mask;
2013
2014         if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
2015                 d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_log;
2016                 d40c->dma_cfg.src_dev_type = STEDMA40_DEV_SRC_MEMORY;
2017                 d40c->dma_cfg.dst_dev_type = d40c->base->plat_data->
2018                         memcpy[d40c->chan.chan_id];
2019
2020         } else if (dma_has_cap(DMA_MEMCPY, cap) &&
2021                    dma_has_cap(DMA_SLAVE, cap)) {
2022                 d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_phy;
2023         } else {
2024                 chan_err(d40c, "No memcpy\n");
2025                 return -EINVAL;
2026         }
2027
2028         return 0;
2029 }
2030
2031 static int d40_free_dma(struct d40_chan *d40c)
2032 {
2033
2034         int res = 0;
2035         u32 event;
2036         struct d40_phy_res *phy = d40c->phy_chan;
2037         bool is_src;
2038
2039         /* Terminate all queued and active transfers */
2040         d40_term_all(d40c);
2041
2042         if (phy == NULL) {
2043                 chan_err(d40c, "phy == null\n");
2044                 return -EINVAL;
2045         }
2046
2047         if (phy->allocated_src == D40_ALLOC_FREE &&
2048             phy->allocated_dst == D40_ALLOC_FREE) {
2049                 chan_err(d40c, "channel already free\n");
2050                 return -EINVAL;
2051         }
2052
2053         if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
2054             d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
2055                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
2056                 is_src = false;
2057         } else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
2058                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
2059                 is_src = true;
2060         } else {
2061                 chan_err(d40c, "Unknown direction\n");
2062                 return -EINVAL;
2063         }
2064
2065         pm_runtime_get_sync(d40c->base->dev);
2066         res = d40_channel_execute_command(d40c, D40_DMA_STOP);
2067         if (res) {
2068                 chan_err(d40c, "stop failed\n");
2069                 goto out;
2070         }
2071
2072         d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
2073
2074         if (chan_is_logical(d40c))
2075                 d40c->base->lookup_log_chans[d40c->log_num] = NULL;
2076         else
2077                 d40c->base->lookup_phy_chans[phy->num] = NULL;
2078
2079         if (d40c->busy) {
2080                 pm_runtime_mark_last_busy(d40c->base->dev);
2081                 pm_runtime_put_autosuspend(d40c->base->dev);
2082         }
2083
2084         d40c->busy = false;
2085         d40c->phy_chan = NULL;
2086         d40c->configured = false;
2087 out:
2088
2089         pm_runtime_mark_last_busy(d40c->base->dev);
2090         pm_runtime_put_autosuspend(d40c->base->dev);
2091         return res;
2092 }
2093
2094 static bool d40_is_paused(struct d40_chan *d40c)
2095 {
2096         void __iomem *chanbase = chan_base(d40c);
2097         bool is_paused = false;
2098         unsigned long flags;
2099         void __iomem *active_reg;
2100         u32 status;
2101         u32 event;
2102
2103         spin_lock_irqsave(&d40c->lock, flags);
2104
2105         if (chan_is_physical(d40c)) {
2106                 if (d40c->phy_chan->num % 2 == 0)
2107                         active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
2108                 else
2109                         active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
2110
2111                 status = (readl(active_reg) &
2112                           D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
2113                         D40_CHAN_POS(d40c->phy_chan->num);
2114                 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
2115                         is_paused = true;
2116
2117                 goto _exit;
2118         }
2119
2120         if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
2121             d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
2122                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
2123                 status = readl(chanbase + D40_CHAN_REG_SDLNK);
2124         } else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
2125                 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
2126                 status = readl(chanbase + D40_CHAN_REG_SSLNK);
2127         } else {
2128                 chan_err(d40c, "Unknown direction\n");
2129                 goto _exit;
2130         }
2131
2132         status = (status & D40_EVENTLINE_MASK(event)) >>
2133                 D40_EVENTLINE_POS(event);
2134
2135         if (status != D40_DMA_RUN)
2136                 is_paused = true;
2137 _exit:
2138         spin_unlock_irqrestore(&d40c->lock, flags);
2139         return is_paused;
2140
2141 }
2142
2143 static u32 stedma40_residue(struct dma_chan *chan)
2144 {
2145         struct d40_chan *d40c =
2146                 container_of(chan, struct d40_chan, chan);
2147         u32 bytes_left;
2148         unsigned long flags;
2149
2150         spin_lock_irqsave(&d40c->lock, flags);
2151         bytes_left = d40_residue(d40c);
2152         spin_unlock_irqrestore(&d40c->lock, flags);
2153
2154         return bytes_left;
2155 }
2156
2157 static int
2158 d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
2159                 struct scatterlist *sg_src, struct scatterlist *sg_dst,
2160                 unsigned int sg_len, dma_addr_t src_dev_addr,
2161                 dma_addr_t dst_dev_addr)
2162 {
2163         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2164         struct stedma40_half_channel_info *src_info = &cfg->src_info;
2165         struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2166         int ret;
2167
2168         ret = d40_log_sg_to_lli(sg_src, sg_len,
2169                                 src_dev_addr,
2170                                 desc->lli_log.src,
2171                                 chan->log_def.lcsp1,
2172                                 src_info->data_width,
2173                                 dst_info->data_width);
2174
2175         ret = d40_log_sg_to_lli(sg_dst, sg_len,
2176                                 dst_dev_addr,
2177                                 desc->lli_log.dst,
2178                                 chan->log_def.lcsp3,
2179                                 dst_info->data_width,
2180                                 src_info->data_width);
2181
2182         return ret < 0 ? ret : 0;
2183 }
2184
2185 static int
2186 d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2187                 struct scatterlist *sg_src, struct scatterlist *sg_dst,
2188                 unsigned int sg_len, dma_addr_t src_dev_addr,
2189                 dma_addr_t dst_dev_addr)
2190 {
2191         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2192         struct stedma40_half_channel_info *src_info = &cfg->src_info;
2193         struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2194         unsigned long flags = 0;
2195         int ret;
2196
2197         if (desc->cyclic)
2198                 flags |= LLI_CYCLIC | LLI_TERM_INT;
2199
2200         ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2201                                 desc->lli_phy.src,
2202                                 virt_to_phys(desc->lli_phy.src),
2203                                 chan->src_def_cfg,
2204                                 src_info, dst_info, flags);
2205
2206         ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2207                                 desc->lli_phy.dst,
2208                                 virt_to_phys(desc->lli_phy.dst),
2209                                 chan->dst_def_cfg,
2210                                 dst_info, src_info, flags);
2211
2212         dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2213                                    desc->lli_pool.size, DMA_TO_DEVICE);
2214
2215         return ret < 0 ? ret : 0;
2216 }
2217
2218 static struct d40_desc *
2219 d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2220               unsigned int sg_len, unsigned long dma_flags)
2221 {
2222         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2223         struct d40_desc *desc;
2224         int ret;
2225
2226         desc = d40_desc_get(chan);
2227         if (!desc)
2228                 return NULL;
2229
2230         desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2231                                         cfg->dst_info.data_width);
2232         if (desc->lli_len < 0) {
2233                 chan_err(chan, "Unaligned size\n");
2234                 goto err;
2235         }
2236
2237         ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2238         if (ret < 0) {
2239                 chan_err(chan, "Could not allocate lli\n");
2240                 goto err;
2241         }
2242
2243         desc->lli_current = 0;
2244         desc->txd.flags = dma_flags;
2245         desc->txd.tx_submit = d40_tx_submit;
2246
2247         dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2248
2249         return desc;
2250
2251 err:
2252         d40_desc_free(chan, desc);
2253         return NULL;
2254 }
2255
2256 static dma_addr_t
2257 d40_get_dev_addr(struct d40_chan *chan, enum dma_transfer_direction direction)
2258 {
2259         struct stedma40_platform_data *plat = chan->base->plat_data;
2260         struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2261         dma_addr_t addr = 0;
2262
2263         if (chan->runtime_addr)
2264                 return chan->runtime_addr;
2265
2266         if (direction == DMA_DEV_TO_MEM)
2267                 addr = plat->dev_rx[cfg->src_dev_type];
2268         else if (direction == DMA_MEM_TO_DEV)
2269                 addr = plat->dev_tx[cfg->dst_dev_type];
2270
2271         return addr;
2272 }
2273
2274 static struct dma_async_tx_descriptor *
2275 d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2276             struct scatterlist *sg_dst, unsigned int sg_len,
2277             enum dma_transfer_direction direction, unsigned long dma_flags)
2278 {
2279         struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2280         dma_addr_t src_dev_addr = 0;
2281         dma_addr_t dst_dev_addr = 0;
2282         struct d40_desc *desc;
2283         unsigned long flags;
2284         int ret;
2285
2286         if (!chan->phy_chan) {
2287                 chan_err(chan, "Cannot prepare unallocated channel\n");
2288                 return NULL;
2289         }
2290
2291         spin_lock_irqsave(&chan->lock, flags);
2292
2293         desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2294         if (desc == NULL)
2295                 goto err;
2296
2297         if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2298                 desc->cyclic = true;
2299
2300         if (direction != DMA_TRANS_NONE) {
2301                 dma_addr_t dev_addr = d40_get_dev_addr(chan, direction);
2302
2303                 if (direction == DMA_DEV_TO_MEM)
2304                         src_dev_addr = dev_addr;
2305                 else if (direction == DMA_MEM_TO_DEV)
2306                         dst_dev_addr = dev_addr;
2307         }
2308
2309         if (chan_is_logical(chan))
2310                 ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2311                                       sg_len, src_dev_addr, dst_dev_addr);
2312         else
2313                 ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2314                                       sg_len, src_dev_addr, dst_dev_addr);
2315
2316         if (ret) {
2317                 chan_err(chan, "Failed to prepare %s sg job: %d\n",
2318                          chan_is_logical(chan) ? "log" : "phy", ret);
2319                 goto err;
2320         }
2321
2322         /*
2323          * add descriptor to the prepare queue in order to be able
2324          * to free them later in terminate_all
2325          */
2326         list_add_tail(&desc->node, &chan->prepare_queue);
2327
2328         spin_unlock_irqrestore(&chan->lock, flags);
2329
2330         return &desc->txd;
2331
2332 err:
2333         if (desc)
2334                 d40_desc_free(chan, desc);
2335         spin_unlock_irqrestore(&chan->lock, flags);
2336         return NULL;
2337 }
2338
2339 bool stedma40_filter(struct dma_chan *chan, void *data)
2340 {
2341         struct stedma40_chan_cfg *info = data;
2342         struct d40_chan *d40c =
2343                 container_of(chan, struct d40_chan, chan);
2344         int err;
2345
2346         if (data) {
2347                 err = d40_validate_conf(d40c, info);
2348                 if (!err)
2349                         d40c->dma_cfg = *info;
2350         } else
2351                 err = d40_config_memcpy(d40c);
2352
2353         if (!err)
2354                 d40c->configured = true;
2355
2356         return err == 0;
2357 }
2358 EXPORT_SYMBOL(stedma40_filter);
2359
2360 static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2361 {
2362         bool realtime = d40c->dma_cfg.realtime;
2363         bool highprio = d40c->dma_cfg.high_priority;
2364         u32 rtreg;
2365         u32 event = D40_TYPE_TO_EVENT(dev_type);
2366         u32 group = D40_TYPE_TO_GROUP(dev_type);
2367         u32 bit = 1 << event;
2368         u32 prioreg;
2369         struct d40_gen_dmac *dmac = &d40c->base->gen_dmac;
2370
2371         rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear;
2372         /*
2373          * Due to a hardware bug, in some cases a logical channel triggered by
2374          * a high priority destination event line can generate extra packet
2375          * transactions.
2376          *
2377          * The workaround is to not set the high priority level for the
2378          * destination event lines that trigger logical channels.
2379          */
2380         if (!src && chan_is_logical(d40c))
2381                 highprio = false;
2382
2383         prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear;
2384
2385         /* Destination event lines are stored in the upper halfword */
2386         if (!src)
2387                 bit <<= 16;
2388
2389         writel(bit, d40c->base->virtbase + prioreg + group * 4);
2390         writel(bit, d40c->base->virtbase + rtreg + group * 4);
2391 }
2392
2393 static void d40_set_prio_realtime(struct d40_chan *d40c)
2394 {
2395         if (d40c->base->rev < 3)
2396                 return;
2397
2398         if ((d40c->dma_cfg.dir ==  STEDMA40_PERIPH_TO_MEM) ||
2399             (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH))
2400                 __d40_set_prio_rt(d40c, d40c->dma_cfg.src_dev_type, true);
2401
2402         if ((d40c->dma_cfg.dir ==  STEDMA40_MEM_TO_PERIPH) ||
2403             (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH))
2404                 __d40_set_prio_rt(d40c, d40c->dma_cfg.dst_dev_type, false);
2405 }
2406
2407 /* DMA ENGINE functions */
2408 static int d40_alloc_chan_resources(struct dma_chan *chan)
2409 {
2410         int err;
2411         unsigned long flags;
2412         struct d40_chan *d40c =
2413                 container_of(chan, struct d40_chan, chan);
2414         bool is_free_phy;
2415         spin_lock_irqsave(&d40c->lock, flags);
2416
2417         dma_cookie_init(chan);
2418
2419         /* If no dma configuration is set use default configuration (memcpy) */
2420         if (!d40c->configured) {
2421                 err = d40_config_memcpy(d40c);
2422                 if (err) {
2423                         chan_err(d40c, "Failed to configure memcpy channel\n");
2424                         goto fail;
2425                 }
2426         }
2427
2428         err = d40_allocate_channel(d40c, &is_free_phy);
2429         if (err) {
2430                 chan_err(d40c, "Failed to allocate channel\n");
2431                 d40c->configured = false;
2432                 goto fail;
2433         }
2434
2435         pm_runtime_get_sync(d40c->base->dev);
2436         /* Fill in basic CFG register values */
2437         d40_phy_cfg(&d40c->dma_cfg, &d40c->src_def_cfg,
2438                     &d40c->dst_def_cfg, chan_is_logical(d40c));
2439
2440         d40_set_prio_realtime(d40c);
2441
2442         if (chan_is_logical(d40c)) {
2443                 d40_log_cfg(&d40c->dma_cfg,
2444                             &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2445
2446                 if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM)
2447                         d40c->lcpa = d40c->base->lcpa_base +
2448                                 d40c->dma_cfg.src_dev_type * D40_LCPA_CHAN_SIZE;
2449                 else
2450                         d40c->lcpa = d40c->base->lcpa_base +
2451                                 d40c->dma_cfg.dst_dev_type *
2452                                 D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2453         }
2454
2455         dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2456                  chan_is_logical(d40c) ? "logical" : "physical",
2457                  d40c->phy_chan->num,
2458                  d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2459
2460
2461         /*
2462          * Only write channel configuration to the DMA if the physical
2463          * resource is free. In case of multiple logical channels
2464          * on the same physical resource, only the first write is necessary.
2465          */
2466         if (is_free_phy)
2467                 d40_config_write(d40c);
2468 fail:
2469         pm_runtime_mark_last_busy(d40c->base->dev);
2470         pm_runtime_put_autosuspend(d40c->base->dev);
2471         spin_unlock_irqrestore(&d40c->lock, flags);
2472         return err;
2473 }
2474
2475 static void d40_free_chan_resources(struct dma_chan *chan)
2476 {
2477         struct d40_chan *d40c =
2478                 container_of(chan, struct d40_chan, chan);
2479         int err;
2480         unsigned long flags;
2481
2482         if (d40c->phy_chan == NULL) {
2483                 chan_err(d40c, "Cannot free unallocated channel\n");
2484                 return;
2485         }
2486
2487         spin_lock_irqsave(&d40c->lock, flags);
2488
2489         err = d40_free_dma(d40c);
2490
2491         if (err)
2492                 chan_err(d40c, "Failed to free channel\n");
2493         spin_unlock_irqrestore(&d40c->lock, flags);
2494 }
2495
2496 static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2497                                                        dma_addr_t dst,
2498                                                        dma_addr_t src,
2499                                                        size_t size,
2500                                                        unsigned long dma_flags)
2501 {
2502         struct scatterlist dst_sg;
2503         struct scatterlist src_sg;
2504
2505         sg_init_table(&dst_sg, 1);
2506         sg_init_table(&src_sg, 1);
2507
2508         sg_dma_address(&dst_sg) = dst;
2509         sg_dma_address(&src_sg) = src;
2510
2511         sg_dma_len(&dst_sg) = size;
2512         sg_dma_len(&src_sg) = size;
2513
2514         return d40_prep_sg(chan, &src_sg, &dst_sg, 1, DMA_NONE, dma_flags);
2515 }
2516
2517 static struct dma_async_tx_descriptor *
2518 d40_prep_memcpy_sg(struct dma_chan *chan,
2519                    struct scatterlist *dst_sg, unsigned int dst_nents,
2520                    struct scatterlist *src_sg, unsigned int src_nents,
2521                    unsigned long dma_flags)
2522 {
2523         if (dst_nents != src_nents)
2524                 return NULL;
2525
2526         return d40_prep_sg(chan, src_sg, dst_sg, src_nents, DMA_NONE, dma_flags);
2527 }
2528
2529 static struct dma_async_tx_descriptor *
2530 d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2531                   unsigned int sg_len, enum dma_transfer_direction direction,
2532                   unsigned long dma_flags, void *context)
2533 {
2534         if (!is_slave_direction(direction))
2535                 return NULL;
2536
2537         return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2538 }
2539
2540 static struct dma_async_tx_descriptor *
2541 dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2542                      size_t buf_len, size_t period_len,
2543                      enum dma_transfer_direction direction, unsigned long flags,
2544                      void *context)
2545 {
2546         unsigned int periods = buf_len / period_len;
2547         struct dma_async_tx_descriptor *txd;
2548         struct scatterlist *sg;
2549         int i;
2550
2551         sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2552         for (i = 0; i < periods; i++) {
2553                 sg_dma_address(&sg[i]) = dma_addr;
2554                 sg_dma_len(&sg[i]) = period_len;
2555                 dma_addr += period_len;
2556         }
2557
2558         sg[periods].offset = 0;
2559         sg_dma_len(&sg[periods]) = 0;
2560         sg[periods].page_link =
2561                 ((unsigned long)sg | 0x01) & ~0x02;
2562
2563         txd = d40_prep_sg(chan, sg, sg, periods, direction,
2564                           DMA_PREP_INTERRUPT);
2565
2566         kfree(sg);
2567
2568         return txd;
2569 }
2570
2571 static enum dma_status d40_tx_status(struct dma_chan *chan,
2572                                      dma_cookie_t cookie,
2573                                      struct dma_tx_state *txstate)
2574 {
2575         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2576         enum dma_status ret;
2577
2578         if (d40c->phy_chan == NULL) {
2579                 chan_err(d40c, "Cannot read status of unallocated channel\n");
2580                 return -EINVAL;
2581         }
2582
2583         ret = dma_cookie_status(chan, cookie, txstate);
2584         if (ret != DMA_SUCCESS)
2585                 dma_set_residue(txstate, stedma40_residue(chan));
2586
2587         if (d40_is_paused(d40c))
2588                 ret = DMA_PAUSED;
2589
2590         return ret;
2591 }
2592
2593 static void d40_issue_pending(struct dma_chan *chan)
2594 {
2595         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2596         unsigned long flags;
2597
2598         if (d40c->phy_chan == NULL) {
2599                 chan_err(d40c, "Channel is not allocated!\n");
2600                 return;
2601         }
2602
2603         spin_lock_irqsave(&d40c->lock, flags);
2604
2605         list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2606
2607         /* Busy means that queued jobs are already being processed */
2608         if (!d40c->busy)
2609                 (void) d40_queue_start(d40c);
2610
2611         spin_unlock_irqrestore(&d40c->lock, flags);
2612 }
2613
2614 static void d40_terminate_all(struct dma_chan *chan)
2615 {
2616         unsigned long flags;
2617         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2618         int ret;
2619
2620         spin_lock_irqsave(&d40c->lock, flags);
2621
2622         pm_runtime_get_sync(d40c->base->dev);
2623         ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2624         if (ret)
2625                 chan_err(d40c, "Failed to stop channel\n");
2626
2627         d40_term_all(d40c);
2628         pm_runtime_mark_last_busy(d40c->base->dev);
2629         pm_runtime_put_autosuspend(d40c->base->dev);
2630         if (d40c->busy) {
2631                 pm_runtime_mark_last_busy(d40c->base->dev);
2632                 pm_runtime_put_autosuspend(d40c->base->dev);
2633         }
2634         d40c->busy = false;
2635
2636         spin_unlock_irqrestore(&d40c->lock, flags);
2637 }
2638
2639 static int
2640 dma40_config_to_halfchannel(struct d40_chan *d40c,
2641                             struct stedma40_half_channel_info *info,
2642                             enum dma_slave_buswidth width,
2643                             u32 maxburst)
2644 {
2645         enum stedma40_periph_data_width addr_width;
2646         int psize;
2647
2648         switch (width) {
2649         case DMA_SLAVE_BUSWIDTH_1_BYTE:
2650                 addr_width = STEDMA40_BYTE_WIDTH;
2651                 break;
2652         case DMA_SLAVE_BUSWIDTH_2_BYTES:
2653                 addr_width = STEDMA40_HALFWORD_WIDTH;
2654                 break;
2655         case DMA_SLAVE_BUSWIDTH_4_BYTES:
2656                 addr_width = STEDMA40_WORD_WIDTH;
2657                 break;
2658         case DMA_SLAVE_BUSWIDTH_8_BYTES:
2659                 addr_width = STEDMA40_DOUBLEWORD_WIDTH;
2660                 break;
2661         default:
2662                 dev_err(d40c->base->dev,
2663                         "illegal peripheral address width "
2664                         "requested (%d)\n",
2665                         width);
2666                 return -EINVAL;
2667         }
2668
2669         if (chan_is_logical(d40c)) {
2670                 if (maxburst >= 16)
2671                         psize = STEDMA40_PSIZE_LOG_16;
2672                 else if (maxburst >= 8)
2673                         psize = STEDMA40_PSIZE_LOG_8;
2674                 else if (maxburst >= 4)
2675                         psize = STEDMA40_PSIZE_LOG_4;
2676                 else
2677                         psize = STEDMA40_PSIZE_LOG_1;
2678         } else {
2679                 if (maxburst >= 16)
2680                         psize = STEDMA40_PSIZE_PHY_16;
2681                 else if (maxburst >= 8)
2682                         psize = STEDMA40_PSIZE_PHY_8;
2683                 else if (maxburst >= 4)
2684                         psize = STEDMA40_PSIZE_PHY_4;
2685                 else
2686                         psize = STEDMA40_PSIZE_PHY_1;
2687         }
2688
2689         info->data_width = addr_width;
2690         info->psize = psize;
2691         info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2692
2693         return 0;
2694 }
2695
2696 /* Runtime reconfiguration extension */
2697 static int d40_set_runtime_config(struct dma_chan *chan,
2698                                   struct dma_slave_config *config)
2699 {
2700         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2701         struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2702         enum dma_slave_buswidth src_addr_width, dst_addr_width;
2703         dma_addr_t config_addr;
2704         u32 src_maxburst, dst_maxburst;
2705         int ret;
2706
2707         src_addr_width = config->src_addr_width;
2708         src_maxburst = config->src_maxburst;
2709         dst_addr_width = config->dst_addr_width;
2710         dst_maxburst = config->dst_maxburst;
2711
2712         if (config->direction == DMA_DEV_TO_MEM) {
2713                 dma_addr_t dev_addr_rx =
2714                         d40c->base->plat_data->dev_rx[cfg->src_dev_type];
2715
2716                 config_addr = config->src_addr;
2717                 if (dev_addr_rx)
2718                         dev_dbg(d40c->base->dev,
2719                                 "channel has a pre-wired RX address %08x "
2720                                 "overriding with %08x\n",
2721                                 dev_addr_rx, config_addr);
2722                 if (cfg->dir != STEDMA40_PERIPH_TO_MEM)
2723                         dev_dbg(d40c->base->dev,
2724                                 "channel was not configured for peripheral "
2725                                 "to memory transfer (%d) overriding\n",
2726                                 cfg->dir);
2727                 cfg->dir = STEDMA40_PERIPH_TO_MEM;
2728
2729                 /* Configure the memory side */
2730                 if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2731                         dst_addr_width = src_addr_width;
2732                 if (dst_maxburst == 0)
2733                         dst_maxburst = src_maxburst;
2734
2735         } else if (config->direction == DMA_MEM_TO_DEV) {
2736                 dma_addr_t dev_addr_tx =
2737                         d40c->base->plat_data->dev_tx[cfg->dst_dev_type];
2738
2739                 config_addr = config->dst_addr;
2740                 if (dev_addr_tx)
2741                         dev_dbg(d40c->base->dev,
2742                                 "channel has a pre-wired TX address %08x "
2743                                 "overriding with %08x\n",
2744                                 dev_addr_tx, config_addr);
2745                 if (cfg->dir != STEDMA40_MEM_TO_PERIPH)
2746                         dev_dbg(d40c->base->dev,
2747                                 "channel was not configured for memory "
2748                                 "to peripheral transfer (%d) overriding\n",
2749                                 cfg->dir);
2750                 cfg->dir = STEDMA40_MEM_TO_PERIPH;
2751
2752                 /* Configure the memory side */
2753                 if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2754                         src_addr_width = dst_addr_width;
2755                 if (src_maxburst == 0)
2756                         src_maxburst = dst_maxburst;
2757         } else {
2758                 dev_err(d40c->base->dev,
2759                         "unrecognized channel direction %d\n",
2760                         config->direction);
2761                 return -EINVAL;
2762         }
2763
2764         if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2765                 dev_err(d40c->base->dev,
2766                         "src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2767                         src_maxburst,
2768                         src_addr_width,
2769                         dst_maxburst,
2770                         dst_addr_width);
2771                 return -EINVAL;
2772         }
2773
2774         if (src_maxburst > 16) {
2775                 src_maxburst = 16;
2776                 dst_maxburst = src_maxburst * src_addr_width / dst_addr_width;
2777         } else if (dst_maxburst > 16) {
2778                 dst_maxburst = 16;
2779                 src_maxburst = dst_maxburst * dst_addr_width / src_addr_width;
2780         }
2781
2782         ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2783                                           src_addr_width,
2784                                           src_maxburst);
2785         if (ret)
2786                 return ret;
2787
2788         ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2789                                           dst_addr_width,
2790                                           dst_maxburst);
2791         if (ret)
2792                 return ret;
2793
2794         /* Fill in register values */
2795         if (chan_is_logical(d40c))
2796                 d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2797         else
2798                 d40_phy_cfg(cfg, &d40c->src_def_cfg,
2799                             &d40c->dst_def_cfg, false);
2800
2801         /* These settings will take precedence later */
2802         d40c->runtime_addr = config_addr;
2803         d40c->runtime_direction = config->direction;
2804         dev_dbg(d40c->base->dev,
2805                 "configured channel %s for %s, data width %d/%d, "
2806                 "maxburst %d/%d elements, LE, no flow control\n",
2807                 dma_chan_name(chan),
2808                 (config->direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2809                 src_addr_width, dst_addr_width,
2810                 src_maxburst, dst_maxburst);
2811
2812         return 0;
2813 }
2814
2815 static int d40_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
2816                        unsigned long arg)
2817 {
2818         struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2819
2820         if (d40c->phy_chan == NULL) {
2821                 chan_err(d40c, "Channel is not allocated!\n");
2822                 return -EINVAL;
2823         }
2824
2825         switch (cmd) {
2826         case DMA_TERMINATE_ALL:
2827                 d40_terminate_all(chan);
2828                 return 0;
2829         case DMA_PAUSE:
2830                 return d40_pause(d40c);
2831         case DMA_RESUME:
2832                 return d40_resume(d40c);
2833         case DMA_SLAVE_CONFIG:
2834                 return d40_set_runtime_config(chan,
2835                         (struct dma_slave_config *) arg);
2836         default:
2837                 break;
2838         }
2839
2840         /* Other commands are unimplemented */
2841         return -ENXIO;
2842 }
2843
2844 /* Initialization functions */
2845
2846 static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2847                                  struct d40_chan *chans, int offset,
2848                                  int num_chans)
2849 {
2850         int i = 0;
2851         struct d40_chan *d40c;
2852
2853         INIT_LIST_HEAD(&dma->channels);
2854
2855         for (i = offset; i < offset + num_chans; i++) {
2856                 d40c = &chans[i];
2857                 d40c->base = base;
2858                 d40c->chan.device = dma;
2859
2860                 spin_lock_init(&d40c->lock);
2861
2862                 d40c->log_num = D40_PHY_CHAN;
2863
2864                 INIT_LIST_HEAD(&d40c->done);
2865                 INIT_LIST_HEAD(&d40c->active);
2866                 INIT_LIST_HEAD(&d40c->queue);
2867                 INIT_LIST_HEAD(&d40c->pending_queue);
2868                 INIT_LIST_HEAD(&d40c->client);
2869                 INIT_LIST_HEAD(&d40c->prepare_queue);
2870
2871                 tasklet_init(&d40c->tasklet, dma_tasklet,
2872                              (unsigned long) d40c);
2873
2874                 list_add_tail(&d40c->chan.device_node,
2875                               &dma->channels);
2876         }
2877 }
2878
2879 static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2880 {
2881         if (dma_has_cap(DMA_SLAVE, dev->cap_mask))
2882                 dev->device_prep_slave_sg = d40_prep_slave_sg;
2883
2884         if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2885                 dev->device_prep_dma_memcpy = d40_prep_memcpy;
2886
2887                 /*
2888                  * This controller can only access address at even
2889                  * 32bit boundaries, i.e. 2^2
2890                  */
2891                 dev->copy_align = 2;
2892         }
2893
2894         if (dma_has_cap(DMA_SG, dev->cap_mask))
2895                 dev->device_prep_dma_sg = d40_prep_memcpy_sg;
2896
2897         if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2898                 dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2899
2900         dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2901         dev->device_free_chan_resources = d40_free_chan_resources;
2902         dev->device_issue_pending = d40_issue_pending;
2903         dev->device_tx_status = d40_tx_status;
2904         dev->device_control = d40_control;
2905         dev->dev = base->dev;
2906 }
2907
2908 static int __init d40_dmaengine_init(struct d40_base *base,
2909                                      int num_reserved_chans)
2910 {
2911         int err ;
2912
2913         d40_chan_init(base, &base->dma_slave, base->log_chans,
2914                       0, base->num_log_chans);
2915
2916         dma_cap_zero(base->dma_slave.cap_mask);
2917         dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2918         dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2919
2920         d40_ops_init(base, &base->dma_slave);
2921
2922         err = dma_async_device_register(&base->dma_slave);
2923
2924         if (err) {
2925                 d40_err(base->dev, "Failed to register slave channels\n");
2926                 goto failure1;
2927         }
2928
2929         d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2930                       base->num_log_chans, base->plat_data->memcpy_len);
2931
2932         dma_cap_zero(base->dma_memcpy.cap_mask);
2933         dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2934         dma_cap_set(DMA_SG, base->dma_memcpy.cap_mask);
2935
2936         d40_ops_init(base, &base->dma_memcpy);
2937
2938         err = dma_async_device_register(&base->dma_memcpy);
2939
2940         if (err) {
2941                 d40_err(base->dev,
2942                         "Failed to regsiter memcpy only channels\n");
2943                 goto failure2;
2944         }
2945
2946         d40_chan_init(base, &base->dma_both, base->phy_chans,
2947                       0, num_reserved_chans);
2948
2949         dma_cap_zero(base->dma_both.cap_mask);
2950         dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2951         dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2952         dma_cap_set(DMA_SG, base->dma_both.cap_mask);
2953         dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2954
2955         d40_ops_init(base, &base->dma_both);
2956         err = dma_async_device_register(&base->dma_both);
2957
2958         if (err) {
2959                 d40_err(base->dev,
2960                         "Failed to register logical and physical capable channels\n");
2961                 goto failure3;
2962         }
2963         return 0;
2964 failure3:
2965         dma_async_device_unregister(&base->dma_memcpy);
2966 failure2:
2967         dma_async_device_unregister(&base->dma_slave);
2968 failure1:
2969         return err;
2970 }
2971
2972 /* Suspend resume functionality */
2973 #ifdef CONFIG_PM
2974 static int dma40_pm_suspend(struct device *dev)
2975 {
2976         struct platform_device *pdev = to_platform_device(dev);
2977         struct d40_base *base = platform_get_drvdata(pdev);
2978         int ret = 0;
2979
2980         if (base->lcpa_regulator)
2981                 ret = regulator_disable(base->lcpa_regulator);
2982         return ret;
2983 }
2984
2985 static int dma40_runtime_suspend(struct device *dev)
2986 {
2987         struct platform_device *pdev = to_platform_device(dev);
2988         struct d40_base *base = platform_get_drvdata(pdev);
2989
2990         d40_save_restore_registers(base, true);
2991
2992         /* Don't disable/enable clocks for v1 due to HW bugs */
2993         if (base->rev != 1)
2994                 writel_relaxed(base->gcc_pwr_off_mask,
2995                                base->virtbase + D40_DREG_GCC);
2996
2997         return 0;
2998 }
2999
3000 static int dma40_runtime_resume(struct device *dev)
3001 {
3002         struct platform_device *pdev = to_platform_device(dev);
3003         struct d40_base *base = platform_get_drvdata(pdev);
3004
3005         if (base->initialized)
3006                 d40_save_restore_registers(base, false);
3007
3008         writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
3009                        base->virtbase + D40_DREG_GCC);
3010         return 0;
3011 }
3012
3013 static int dma40_resume(struct device *dev)
3014 {
3015         struct platform_device *pdev = to_platform_device(dev);
3016         struct d40_base *base = platform_get_drvdata(pdev);
3017         int ret = 0;
3018
3019         if (base->lcpa_regulator)
3020                 ret = regulator_enable(base->lcpa_regulator);
3021
3022         return ret;
3023 }
3024
3025 static const struct dev_pm_ops dma40_pm_ops = {
3026         .suspend                = dma40_pm_suspend,
3027         .runtime_suspend        = dma40_runtime_suspend,
3028         .runtime_resume         = dma40_runtime_resume,
3029         .resume                 = dma40_resume,
3030 };
3031 #define DMA40_PM_OPS    (&dma40_pm_ops)
3032 #else
3033 #define DMA40_PM_OPS    NULL
3034 #endif
3035
3036 /* Initialization functions. */
3037
3038 static int __init d40_phy_res_init(struct d40_base *base)
3039 {
3040         int i;
3041         int num_phy_chans_avail = 0;
3042         u32 val[2];
3043         int odd_even_bit = -2;
3044         int gcc = D40_DREG_GCC_ENA;
3045
3046         val[0] = readl(base->virtbase + D40_DREG_PRSME);
3047         val[1] = readl(base->virtbase + D40_DREG_PRSMO);
3048
3049         for (i = 0; i < base->num_phy_chans; i++) {
3050                 base->phy_res[i].num = i;
3051                 odd_even_bit += 2 * ((i % 2) == 0);
3052                 if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
3053                         /* Mark security only channels as occupied */
3054                         base->phy_res[i].allocated_src = D40_ALLOC_PHY;
3055                         base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
3056                         base->phy_res[i].reserved = true;
3057                         gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3058                                                        D40_DREG_GCC_SRC);
3059                         gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3060                                                        D40_DREG_GCC_DST);
3061
3062
3063                 } else {
3064                         base->phy_res[i].allocated_src = D40_ALLOC_FREE;
3065                         base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
3066                         base->phy_res[i].reserved = false;
3067                         num_phy_chans_avail++;
3068                 }
3069                 spin_lock_init(&base->phy_res[i].lock);
3070         }
3071
3072         /* Mark disabled channels as occupied */
3073         for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
3074                 int chan = base->plat_data->disabled_channels[i];
3075
3076                 base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
3077                 base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
3078                 base->phy_res[chan].reserved = true;
3079                 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3080                                                D40_DREG_GCC_SRC);
3081                 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3082                                                D40_DREG_GCC_DST);
3083                 num_phy_chans_avail--;
3084         }
3085
3086         /* Mark soft_lli channels */
3087         for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) {
3088                 int chan = base->plat_data->soft_lli_chans[i];
3089
3090                 base->phy_res[chan].use_soft_lli = true;
3091         }
3092
3093         dev_info(base->dev, "%d of %d physical DMA channels available\n",
3094                  num_phy_chans_avail, base->num_phy_chans);
3095
3096         /* Verify settings extended vs standard */
3097         val[0] = readl(base->virtbase + D40_DREG_PRTYP);
3098
3099         for (i = 0; i < base->num_phy_chans; i++) {
3100
3101                 if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
3102                     (val[0] & 0x3) != 1)
3103                         dev_info(base->dev,
3104                                  "[%s] INFO: channel %d is misconfigured (%d)\n",
3105                                  __func__, i, val[0] & 0x3);
3106
3107                 val[0] = val[0] >> 2;
3108         }
3109
3110         /*
3111          * To keep things simple, Enable all clocks initially.
3112          * The clocks will get managed later post channel allocation.
3113          * The clocks for the event lines on which reserved channels exists
3114          * are not managed here.
3115          */
3116         writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3117         base->gcc_pwr_off_mask = gcc;
3118
3119         return num_phy_chans_avail;
3120 }
3121
3122 static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
3123 {
3124         struct stedma40_platform_data *plat_data;
3125         struct clk *clk = NULL;
3126         void __iomem *virtbase = NULL;
3127         struct resource *res = NULL;
3128         struct d40_base *base = NULL;
3129         int num_log_chans = 0;
3130         int num_phy_chans;
3131         int clk_ret = -EINVAL;
3132         int i;
3133         u32 pid;
3134         u32 cid;
3135         u8 rev;
3136
3137         clk = clk_get(&pdev->dev, NULL);
3138         if (IS_ERR(clk)) {
3139                 d40_err(&pdev->dev, "No matching clock found\n");
3140                 goto failure;
3141         }
3142
3143         clk_ret = clk_prepare_enable(clk);
3144         if (clk_ret) {
3145                 d40_err(&pdev->dev, "Failed to prepare/enable clock\n");
3146                 goto failure;
3147         }
3148
3149         /* Get IO for DMAC base address */
3150         res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
3151         if (!res)
3152                 goto failure;
3153
3154         if (request_mem_region(res->start, resource_size(res),
3155                                D40_NAME " I/O base") == NULL)
3156                 goto failure;
3157
3158         virtbase = ioremap(res->start, resource_size(res));
3159         if (!virtbase)
3160                 goto failure;
3161
3162         /* This is just a regular AMBA PrimeCell ID actually */
3163         for (pid = 0, i = 0; i < 4; i++)
3164                 pid |= (readl(virtbase + resource_size(res) - 0x20 + 4 * i)
3165                         & 255) << (i * 8);
3166         for (cid = 0, i = 0; i < 4; i++)
3167                 cid |= (readl(virtbase + resource_size(res) - 0x10 + 4 * i)
3168                         & 255) << (i * 8);
3169
3170         if (cid != AMBA_CID) {
3171                 d40_err(&pdev->dev, "Unknown hardware! No PrimeCell ID\n");
3172                 goto failure;
3173         }
3174         if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
3175                 d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
3176                         AMBA_MANF_BITS(pid),
3177                         AMBA_VENDOR_ST);
3178                 goto failure;
3179         }
3180         /*
3181          * HW revision:
3182          * DB8500ed has revision 0
3183          * ? has revision 1
3184          * DB8500v1 has revision 2
3185          * DB8500v2 has revision 3
3186          * AP9540v1 has revision 4
3187          * DB8540v1 has revision 4
3188          */
3189         rev = AMBA_REV_BITS(pid);
3190
3191         plat_data = pdev->dev.platform_data;
3192
3193         /* The number of physical channels on this HW */
3194         if (plat_data->num_of_phy_chans)
3195                 num_phy_chans = plat_data->num_of_phy_chans;
3196         else
3197                 num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
3198
3199         dev_info(&pdev->dev, "hardware revision: %d @ 0x%x with %d physical channels\n",
3200                  rev, res->start, num_phy_chans);
3201
3202         if (rev < 2) {
3203                 d40_err(&pdev->dev, "hardware revision: %d is not supported",
3204                         rev);
3205                 goto failure;
3206         }
3207
3208         /* Count the number of logical channels in use */
3209         for (i = 0; i < plat_data->dev_len; i++)
3210                 if (plat_data->dev_rx[i] != 0)
3211                         num_log_chans++;
3212
3213         for (i = 0; i < plat_data->dev_len; i++)
3214                 if (plat_data->dev_tx[i] != 0)
3215                         num_log_chans++;
3216
3217         base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
3218                        (num_phy_chans + num_log_chans + plat_data->memcpy_len) *
3219                        sizeof(struct d40_chan), GFP_KERNEL);
3220
3221         if (base == NULL) {
3222                 d40_err(&pdev->dev, "Out of memory\n");
3223                 goto failure;
3224         }
3225
3226         base->rev = rev;
3227         base->clk = clk;
3228         base->num_phy_chans = num_phy_chans;
3229         base->num_log_chans = num_log_chans;
3230         base->phy_start = res->start;
3231         base->phy_size = resource_size(res);
3232         base->virtbase = virtbase;
3233         base->plat_data = plat_data;
3234         base->dev = &pdev->dev;
3235         base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3236         base->log_chans = &base->phy_chans[num_phy_chans];
3237
3238         if (base->plat_data->num_of_phy_chans == 14) {
3239                 base->gen_dmac.backup = d40_backup_regs_v4b;
3240                 base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
3241                 base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
3242                 base->gen_dmac.interrupt_clear = D40_DREG_CPCICR;
3243                 base->gen_dmac.realtime_en = D40_DREG_CRSEG1;
3244                 base->gen_dmac.realtime_clear = D40_DREG_CRCEG1;
3245                 base->gen_dmac.high_prio_en = D40_DREG_CPSEG1;
3246                 base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1;
3247                 base->gen_dmac.il = il_v4b;
3248                 base->gen_dmac.il_size = ARRAY_SIZE(il_v4b);
3249                 base->gen_dmac.init_reg = dma_init_reg_v4b;
3250                 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
3251         } else {
3252                 if (base->rev >= 3) {
3253                         base->gen_dmac.backup = d40_backup_regs_v4a;
3254                         base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
3255                 }
3256                 base->gen_dmac.interrupt_en = D40_DREG_PCMIS;
3257                 base->gen_dmac.interrupt_clear = D40_DREG_PCICR;
3258                 base->gen_dmac.realtime_en = D40_DREG_RSEG1;
3259                 base->gen_dmac.realtime_clear = D40_DREG_RCEG1;
3260                 base->gen_dmac.high_prio_en = D40_DREG_PSEG1;
3261                 base->gen_dmac.high_prio_clear = D40_DREG_PCEG1;
3262                 base->gen_dmac.il = il_v4a;
3263                 base->gen_dmac.il_size = ARRAY_SIZE(il_v4a);
3264                 base->gen_dmac.init_reg = dma_init_reg_v4a;
3265                 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a);
3266         }
3267
3268         base->phy_res = kzalloc(num_phy_chans * sizeof(struct d40_phy_res),
3269                                 GFP_KERNEL);
3270         if (!base->phy_res)
3271                 goto failure;
3272
3273         base->lookup_phy_chans = kzalloc(num_phy_chans *
3274                                          sizeof(struct d40_chan *),
3275                                          GFP_KERNEL);
3276         if (!base->lookup_phy_chans)
3277                 goto failure;
3278
3279         if (num_log_chans + plat_data->memcpy_len) {
3280                 /*
3281                  * The max number of logical channels are event lines for all
3282                  * src devices and dst devices
3283                  */
3284                 base->lookup_log_chans = kzalloc(plat_data->dev_len * 2 *
3285                                                  sizeof(struct d40_chan *),
3286                                                  GFP_KERNEL);
3287                 if (!base->lookup_log_chans)
3288                         goto failure;
3289         }
3290
3291         base->reg_val_backup_chan = kmalloc(base->num_phy_chans *
3292                                             sizeof(d40_backup_regs_chan),
3293                                             GFP_KERNEL);
3294         if (!base->reg_val_backup_chan)
3295                 goto failure;
3296
3297         base->lcla_pool.alloc_map =
3298                 kzalloc(num_phy_chans * sizeof(struct d40_desc *)
3299                         * D40_LCLA_LINK_PER_EVENT_GRP, GFP_KERNEL);
3300         if (!base->lcla_pool.alloc_map)
3301                 goto failure;
3302
3303         base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3304                                             0, SLAB_HWCACHE_ALIGN,
3305                                             NULL);
3306         if (base->desc_slab == NULL)
3307                 goto failure;
3308
3309         return base;
3310
3311 failure:
3312         if (!clk_ret)
3313                 clk_disable_unprepare(clk);
3314         if (!IS_ERR(clk))
3315                 clk_put(clk);
3316         if (virtbase)
3317                 iounmap(virtbase);
3318         if (res)
3319                 release_mem_region(res->start,
3320                                    resource_size(res));
3321         if (virtbase)
3322                 iounmap(virtbase);
3323
3324         if (base) {
3325                 kfree(base->lcla_pool.alloc_map);
3326                 kfree(base->reg_val_backup_chan);
3327                 kfree(base->lookup_log_chans);
3328                 kfree(base->lookup_phy_chans);
3329                 kfree(base->phy_res);
3330                 kfree(base);
3331         }
3332
3333         return NULL;
3334 }
3335
3336 static void __init d40_hw_init(struct d40_base *base)
3337 {
3338
3339         int i;
3340         u32 prmseo[2] = {0, 0};
3341         u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3342         u32 pcmis = 0;
3343         u32 pcicr = 0;
3344         struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg;
3345         u32 reg_size = base->gen_dmac.init_reg_size;
3346
3347         for (i = 0; i < reg_size; i++)
3348                 writel(dma_init_reg[i].val,
3349                        base->virtbase + dma_init_reg[i].reg);
3350
3351         /* Configure all our dma channels to default settings */
3352         for (i = 0; i < base->num_phy_chans; i++) {
3353
3354                 activeo[i % 2] = activeo[i % 2] << 2;
3355
3356                 if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3357                     == D40_ALLOC_PHY) {
3358                         activeo[i % 2] |= 3;
3359                         continue;
3360                 }
3361
3362                 /* Enable interrupt # */
3363                 pcmis = (pcmis << 1) | 1;
3364
3365                 /* Clear interrupt # */
3366                 pcicr = (pcicr << 1) | 1;
3367
3368                 /* Set channel to physical mode */
3369                 prmseo[i % 2] = prmseo[i % 2] << 2;
3370                 prmseo[i % 2] |= 1;
3371
3372         }
3373
3374         writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3375         writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3376         writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3377         writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3378
3379         /* Write which interrupt to enable */
3380         writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en);
3381
3382         /* Write which interrupt to clear */
3383         writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear);
3384
3385         /* These are __initdata and cannot be accessed after init */
3386         base->gen_dmac.init_reg = NULL;
3387         base->gen_dmac.init_reg_size = 0;
3388 }
3389
3390 static int __init d40_lcla_allocate(struct d40_base *base)
3391 {
3392         struct d40_lcla_pool *pool = &base->lcla_pool;
3393         unsigned long *page_list;
3394         int i, j;
3395         int ret = 0;
3396
3397         /*
3398          * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3399          * To full fill this hardware requirement without wasting 256 kb
3400          * we allocate pages until we get an aligned one.
3401          */
3402         page_list = kmalloc(sizeof(unsigned long) * MAX_LCLA_ALLOC_ATTEMPTS,
3403                             GFP_KERNEL);
3404
3405         if (!page_list) {
3406                 ret = -ENOMEM;
3407                 goto failure;
3408         }
3409
3410         /* Calculating how many pages that are required */
3411         base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3412
3413         for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3414                 page_list[i] = __get_free_pages(GFP_KERNEL,
3415                                                 base->lcla_pool.pages);
3416                 if (!page_list[i]) {
3417
3418                         d40_err(base->dev, "Failed to allocate %d pages.\n",
3419                                 base->lcla_pool.pages);
3420
3421                         for (j = 0; j < i; j++)
3422                                 free_pages(page_list[j], base->lcla_pool.pages);
3423                         goto failure;
3424                 }
3425
3426                 if ((virt_to_phys((void *)page_list[i]) &
3427                      (LCLA_ALIGNMENT - 1)) == 0)
3428                         break;
3429         }
3430
3431         for (j = 0; j < i; j++)
3432                 free_pages(page_list[j], base->lcla_pool.pages);
3433
3434         if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3435                 base->lcla_pool.base = (void *)page_list[i];
3436         } else {
3437                 /*
3438                  * After many attempts and no succees with finding the correct
3439                  * alignment, try with allocating a big buffer.
3440                  */
3441                 dev_warn(base->dev,
3442                          "[%s] Failed to get %d pages @ 18 bit align.\n",
3443                          __func__, base->lcla_pool.pages);
3444                 base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3445                                                          base->num_phy_chans +
3446                                                          LCLA_ALIGNMENT,
3447                                                          GFP_KERNEL);
3448                 if (!base->lcla_pool.base_unaligned) {
3449                         ret = -ENOMEM;
3450                         goto failure;
3451                 }
3452
3453                 base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3454                                                  LCLA_ALIGNMENT);
3455         }
3456
3457         pool->dma_addr = dma_map_single(base->dev, pool->base,
3458                                         SZ_1K * base->num_phy_chans,
3459                                         DMA_TO_DEVICE);
3460         if (dma_mapping_error(base->dev, pool->dma_addr)) {
3461                 pool->dma_addr = 0;
3462                 ret = -ENOMEM;
3463                 goto failure;
3464         }
3465
3466         writel(virt_to_phys(base->lcla_pool.base),
3467                base->virtbase + D40_DREG_LCLA);
3468 failure:
3469         kfree(page_list);
3470         return ret;
3471 }
3472
3473 static int __init d40_probe(struct platform_device *pdev)
3474 {
3475         int err;
3476         int ret = -ENOENT;
3477         struct d40_base *base;
3478         struct resource *res = NULL;
3479         int num_reserved_chans;
3480         u32 val;
3481
3482         base = d40_hw_detect_init(pdev);
3483
3484         if (!base)
3485                 goto failure;
3486
3487         num_reserved_chans = d40_phy_res_init(base);
3488
3489         platform_set_drvdata(pdev, base);
3490
3491         spin_lock_init(&base->interrupt_lock);
3492         spin_lock_init(&base->execmd_lock);
3493
3494         /* Get IO for logical channel parameter address */
3495         res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
3496         if (!res) {
3497                 ret = -ENOENT;
3498                 d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
3499                 goto failure;
3500         }
3501         base->lcpa_size = resource_size(res);
3502         base->phy_lcpa = res->start;
3503
3504         if (request_mem_region(res->start, resource_size(res),
3505                                D40_NAME " I/O lcpa") == NULL) {
3506                 ret = -EBUSY;
3507                 d40_err(&pdev->dev,
3508                         "Failed to request LCPA region 0x%x-0x%x\n",
3509                         res->start, res->end);
3510                 goto failure;
3511         }
3512
3513         /* We make use of ESRAM memory for this. */
3514         val = readl(base->virtbase + D40_DREG_LCPA);
3515         if (res->start != val && val != 0) {
3516                 dev_warn(&pdev->dev,
3517                          "[%s] Mismatch LCPA dma 0x%x, def 0x%x\n",
3518                          __func__, val, res->start);
3519         } else
3520                 writel(res->start, base->virtbase + D40_DREG_LCPA);
3521
3522         base->lcpa_base = ioremap(res->start, resource_size(res));
3523         if (!base->lcpa_base) {
3524                 ret = -ENOMEM;
3525                 d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
3526                 goto failure;
3527         }
3528         /* If lcla has to be located in ESRAM we don't need to allocate */
3529         if (base->plat_data->use_esram_lcla) {
3530                 res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3531                                                         "lcla_esram");
3532                 if (!res) {
3533                         ret = -ENOENT;
3534                         d40_err(&pdev->dev,
3535                                 "No \"lcla_esram\" memory resource\n");
3536                         goto failure;
3537                 }
3538                 base->lcla_pool.base = ioremap(res->start,
3539                                                 resource_size(res));
3540                 if (!base->lcla_pool.base) {
3541                         ret = -ENOMEM;
3542                         d40_err(&pdev->dev, "Failed to ioremap LCLA region\n");
3543                         goto failure;
3544                 }
3545                 writel(res->start, base->virtbase + D40_DREG_LCLA);
3546
3547         } else {
3548                 ret = d40_lcla_allocate(base);
3549                 if (ret) {
3550                         d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
3551                         goto failure;
3552                 }
3553         }
3554
3555         spin_lock_init(&base->lcla_pool.lock);
3556
3557         base->irq = platform_get_irq(pdev, 0);
3558
3559         ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3560         if (ret) {
3561                 d40_err(&pdev->dev, "No IRQ defined\n");
3562                 goto failure;
3563         }
3564
3565         pm_runtime_irq_safe(base->dev);
3566         pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3567         pm_runtime_use_autosuspend(base->dev);
3568         pm_runtime_enable(base->dev);
3569         pm_runtime_resume(base->dev);
3570
3571         if (base->plat_data->use_esram_lcla) {
3572
3573                 base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3574                 if (IS_ERR(base->lcpa_regulator)) {
3575                         d40_err(&pdev->dev, "Failed to get lcpa_regulator\n");
3576                         base->lcpa_regulator = NULL;
3577                         goto failure;
3578                 }
3579
3580                 ret = regulator_enable(base->lcpa_regulator);
3581                 if (ret) {
3582                         d40_err(&pdev->dev,
3583                                 "Failed to enable lcpa_regulator\n");
3584                         regulator_put(base->lcpa_regulator);
3585                         base->lcpa_regulator = NULL;
3586                         goto failure;
3587                 }
3588         }
3589
3590         base->initialized = true;
3591         err = d40_dmaengine_init(base, num_reserved_chans);
3592         if (err)
3593                 goto failure;
3594
3595         base->dev->dma_parms = &base->dma_parms;
3596         err = dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
3597         if (err) {
3598                 d40_err(&pdev->dev, "Failed to set dma max seg size\n");
3599                 goto failure;
3600         }
3601
3602         d40_hw_init(base);
3603
3604         dev_info(base->dev, "initialized\n");
3605         return 0;
3606
3607 failure:
3608         if (base) {
3609                 if (base->desc_slab)
3610                         kmem_cache_destroy(base->desc_slab);
3611                 if (base->virtbase)
3612                         iounmap(base->virtbase);
3613
3614                 if (base->lcla_pool.base && base->plat_data->use_esram_lcla) {
3615                         iounmap(base->lcla_pool.base);
3616                         base->lcla_pool.base = NULL;
3617                 }
3618
3619                 if (base->lcla_pool.dma_addr)
3620                         dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3621                                          SZ_1K * base->num_phy_chans,
3622                                          DMA_TO_DEVICE);
3623
3624                 if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3625                         free_pages((unsigned long)base->lcla_pool.base,
3626                                    base->lcla_pool.pages);
3627
3628                 kfree(base->lcla_pool.base_unaligned);
3629
3630                 if (base->phy_lcpa)
3631                         release_mem_region(base->phy_lcpa,
3632                                            base->lcpa_size);
3633                 if (base->phy_start)
3634                         release_mem_region(base->phy_start,
3635                                            base->phy_size);
3636                 if (base->clk) {
3637                         clk_disable_unprepare(base->clk);
3638                         clk_put(base->clk);
3639                 }
3640
3641                 if (base->lcpa_regulator) {
3642                         regulator_disable(base->lcpa_regulator);
3643                         regulator_put(base->lcpa_regulator);
3644                 }
3645
3646                 kfree(base->lcla_pool.alloc_map);
3647                 kfree(base->lookup_log_chans);
3648                 kfree(base->lookup_phy_chans);
3649                 kfree(base->phy_res);
3650                 kfree(base);
3651         }
3652
3653         d40_err(&pdev->dev, "probe failed\n");
3654         return ret;
3655 }
3656
3657 static struct platform_driver d40_driver = {
3658         .driver = {
3659                 .owner = THIS_MODULE,
3660                 .name  = D40_NAME,
3661                 .pm = DMA40_PM_OPS,
3662         },
3663 };
3664
3665 static int __init stedma40_init(void)
3666 {
3667         return platform_driver_probe(&d40_driver, d40_probe);
3668 }
3669 subsys_initcall(stedma40_init);