Merge branch 'x86-mm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[cascardo/linux.git] / drivers / dma / edma.c
1 /*
2  * TI EDMA DMA engine driver
3  *
4  * Copyright 2012 Texas Instruments
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation version 2.
9  *
10  * This program is distributed "as is" WITHOUT ANY WARRANTY of any
11  * kind, whether express or implied; without even the implied warranty
12  * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  */
15
16 #include <linux/dmaengine.h>
17 #include <linux/dma-mapping.h>
18 #include <linux/edma.h>
19 #include <linux/err.h>
20 #include <linux/init.h>
21 #include <linux/interrupt.h>
22 #include <linux/list.h>
23 #include <linux/module.h>
24 #include <linux/platform_device.h>
25 #include <linux/slab.h>
26 #include <linux/spinlock.h>
27 #include <linux/of.h>
28 #include <linux/of_dma.h>
29 #include <linux/of_irq.h>
30 #include <linux/of_address.h>
31 #include <linux/of_device.h>
32 #include <linux/pm_runtime.h>
33
34 #include <linux/platform_data/edma.h>
35
36 #include "dmaengine.h"
37 #include "virt-dma.h"
38
39 /* Offsets matching "struct edmacc_param" */
40 #define PARM_OPT                0x00
41 #define PARM_SRC                0x04
42 #define PARM_A_B_CNT            0x08
43 #define PARM_DST                0x0c
44 #define PARM_SRC_DST_BIDX       0x10
45 #define PARM_LINK_BCNTRLD       0x14
46 #define PARM_SRC_DST_CIDX       0x18
47 #define PARM_CCNT               0x1c
48
49 #define PARM_SIZE               0x20
50
51 /* Offsets for EDMA CC global channel registers and their shadows */
52 #define SH_ER                   0x00    /* 64 bits */
53 #define SH_ECR                  0x08    /* 64 bits */
54 #define SH_ESR                  0x10    /* 64 bits */
55 #define SH_CER                  0x18    /* 64 bits */
56 #define SH_EER                  0x20    /* 64 bits */
57 #define SH_EECR                 0x28    /* 64 bits */
58 #define SH_EESR                 0x30    /* 64 bits */
59 #define SH_SER                  0x38    /* 64 bits */
60 #define SH_SECR                 0x40    /* 64 bits */
61 #define SH_IER                  0x50    /* 64 bits */
62 #define SH_IECR                 0x58    /* 64 bits */
63 #define SH_IESR                 0x60    /* 64 bits */
64 #define SH_IPR                  0x68    /* 64 bits */
65 #define SH_ICR                  0x70    /* 64 bits */
66 #define SH_IEVAL                0x78
67 #define SH_QER                  0x80
68 #define SH_QEER                 0x84
69 #define SH_QEECR                0x88
70 #define SH_QEESR                0x8c
71 #define SH_QSER                 0x90
72 #define SH_QSECR                0x94
73 #define SH_SIZE                 0x200
74
75 /* Offsets for EDMA CC global registers */
76 #define EDMA_REV                0x0000
77 #define EDMA_CCCFG              0x0004
78 #define EDMA_QCHMAP             0x0200  /* 8 registers */
79 #define EDMA_DMAQNUM            0x0240  /* 8 registers (4 on OMAP-L1xx) */
80 #define EDMA_QDMAQNUM           0x0260
81 #define EDMA_QUETCMAP           0x0280
82 #define EDMA_QUEPRI             0x0284
83 #define EDMA_EMR                0x0300  /* 64 bits */
84 #define EDMA_EMCR               0x0308  /* 64 bits */
85 #define EDMA_QEMR               0x0310
86 #define EDMA_QEMCR              0x0314
87 #define EDMA_CCERR              0x0318
88 #define EDMA_CCERRCLR           0x031c
89 #define EDMA_EEVAL              0x0320
90 #define EDMA_DRAE               0x0340  /* 4 x 64 bits*/
91 #define EDMA_QRAE               0x0380  /* 4 registers */
92 #define EDMA_QUEEVTENTRY        0x0400  /* 2 x 16 registers */
93 #define EDMA_QSTAT              0x0600  /* 2 registers */
94 #define EDMA_QWMTHRA            0x0620
95 #define EDMA_QWMTHRB            0x0624
96 #define EDMA_CCSTAT             0x0640
97
98 #define EDMA_M                  0x1000  /* global channel registers */
99 #define EDMA_ECR                0x1008
100 #define EDMA_ECRH               0x100C
101 #define EDMA_SHADOW0            0x2000  /* 4 shadow regions */
102 #define EDMA_PARM               0x4000  /* PaRAM entries */
103
104 #define PARM_OFFSET(param_no)   (EDMA_PARM + ((param_no) << 5))
105
106 #define EDMA_DCHMAP             0x0100  /* 64 registers */
107
108 /* CCCFG register */
109 #define GET_NUM_DMACH(x)        (x & 0x7) /* bits 0-2 */
110 #define GET_NUM_QDMACH(x)       ((x & 0x70) >> 4) /* bits 4-6 */
111 #define GET_NUM_PAENTRY(x)      ((x & 0x7000) >> 12) /* bits 12-14 */
112 #define GET_NUM_EVQUE(x)        ((x & 0x70000) >> 16) /* bits 16-18 */
113 #define GET_NUM_REGN(x)         ((x & 0x300000) >> 20) /* bits 20-21 */
114 #define CHMAP_EXIST             BIT(24)
115
116 /* CCSTAT register */
117 #define EDMA_CCSTAT_ACTV        BIT(4)
118
119 /*
120  * Max of 20 segments per channel to conserve PaRAM slots
121  * Also note that MAX_NR_SG should be atleast the no.of periods
122  * that are required for ASoC, otherwise DMA prep calls will
123  * fail. Today davinci-pcm is the only user of this driver and
124  * requires atleast 17 slots, so we setup the default to 20.
125  */
126 #define MAX_NR_SG               20
127 #define EDMA_MAX_SLOTS          MAX_NR_SG
128 #define EDMA_DESCRIPTORS        16
129
130 #define EDMA_CHANNEL_ANY                -1      /* for edma_alloc_channel() */
131 #define EDMA_SLOT_ANY                   -1      /* for edma_alloc_slot() */
132 #define EDMA_CONT_PARAMS_ANY             1001
133 #define EDMA_CONT_PARAMS_FIXED_EXACT     1002
134 #define EDMA_CONT_PARAMS_FIXED_NOT_EXACT 1003
135
136 /* PaRAM slots are laid out like this */
137 struct edmacc_param {
138         u32 opt;
139         u32 src;
140         u32 a_b_cnt;
141         u32 dst;
142         u32 src_dst_bidx;
143         u32 link_bcntrld;
144         u32 src_dst_cidx;
145         u32 ccnt;
146 } __packed;
147
148 /* fields in edmacc_param.opt */
149 #define SAM             BIT(0)
150 #define DAM             BIT(1)
151 #define SYNCDIM         BIT(2)
152 #define STATIC          BIT(3)
153 #define EDMA_FWID       (0x07 << 8)
154 #define TCCMODE         BIT(11)
155 #define EDMA_TCC(t)     ((t) << 12)
156 #define TCINTEN         BIT(20)
157 #define ITCINTEN        BIT(21)
158 #define TCCHEN          BIT(22)
159 #define ITCCHEN         BIT(23)
160
161 struct edma_pset {
162         u32                             len;
163         dma_addr_t                      addr;
164         struct edmacc_param             param;
165 };
166
167 struct edma_desc {
168         struct virt_dma_desc            vdesc;
169         struct list_head                node;
170         enum dma_transfer_direction     direction;
171         int                             cyclic;
172         int                             absync;
173         int                             pset_nr;
174         struct edma_chan                *echan;
175         int                             processed;
176
177         /*
178          * The following 4 elements are used for residue accounting.
179          *
180          * - processed_stat: the number of SG elements we have traversed
181          * so far to cover accounting. This is updated directly to processed
182          * during edma_callback and is always <= processed, because processed
183          * refers to the number of pending transfer (programmed to EDMA
184          * controller), where as processed_stat tracks number of transfers
185          * accounted for so far.
186          *
187          * - residue: The amount of bytes we have left to transfer for this desc
188          *
189          * - residue_stat: The residue in bytes of data we have covered
190          * so far for accounting. This is updated directly to residue
191          * during callbacks to keep it current.
192          *
193          * - sg_len: Tracks the length of the current intermediate transfer,
194          * this is required to update the residue during intermediate transfer
195          * completion callback.
196          */
197         int                             processed_stat;
198         u32                             sg_len;
199         u32                             residue;
200         u32                             residue_stat;
201
202         struct edma_pset                pset[0];
203 };
204
205 struct edma_cc;
206
207 struct edma_tc {
208         struct device_node              *node;
209         u16                             id;
210 };
211
212 struct edma_chan {
213         struct virt_dma_chan            vchan;
214         struct list_head                node;
215         struct edma_desc                *edesc;
216         struct edma_cc                  *ecc;
217         struct edma_tc                  *tc;
218         int                             ch_num;
219         bool                            alloced;
220         bool                            hw_triggered;
221         int                             slot[EDMA_MAX_SLOTS];
222         int                             missed;
223         struct dma_slave_config         cfg;
224 };
225
226 struct edma_cc {
227         struct device                   *dev;
228         struct edma_soc_info            *info;
229         void __iomem                    *base;
230         int                             id;
231         bool                            legacy_mode;
232
233         /* eDMA3 resource information */
234         unsigned                        num_channels;
235         unsigned                        num_qchannels;
236         unsigned                        num_region;
237         unsigned                        num_slots;
238         unsigned                        num_tc;
239         bool                            chmap_exist;
240         enum dma_event_q                default_queue;
241
242         /*
243          * The slot_inuse bit for each PaRAM slot is clear unless the slot is
244          * in use by Linux or if it is allocated to be used by DSP.
245          */
246         unsigned long *slot_inuse;
247
248         struct dma_device               dma_slave;
249         struct dma_device               *dma_memcpy;
250         struct edma_chan                *slave_chans;
251         struct edma_tc                  *tc_list;
252         int                             dummy_slot;
253 };
254
255 /* dummy param set used to (re)initialize parameter RAM slots */
256 static const struct edmacc_param dummy_paramset = {
257         .link_bcntrld = 0xffff,
258         .ccnt = 1,
259 };
260
261 #define EDMA_BINDING_LEGACY     0
262 #define EDMA_BINDING_TPCC       1
263 static const struct of_device_id edma_of_ids[] = {
264         {
265                 .compatible = "ti,edma3",
266                 .data = (void *)EDMA_BINDING_LEGACY,
267         },
268         {
269                 .compatible = "ti,edma3-tpcc",
270                 .data = (void *)EDMA_BINDING_TPCC,
271         },
272         {}
273 };
274
275 static const struct of_device_id edma_tptc_of_ids[] = {
276         { .compatible = "ti,edma3-tptc", },
277         {}
278 };
279
280 static inline unsigned int edma_read(struct edma_cc *ecc, int offset)
281 {
282         return (unsigned int)__raw_readl(ecc->base + offset);
283 }
284
285 static inline void edma_write(struct edma_cc *ecc, int offset, int val)
286 {
287         __raw_writel(val, ecc->base + offset);
288 }
289
290 static inline void edma_modify(struct edma_cc *ecc, int offset, unsigned and,
291                                unsigned or)
292 {
293         unsigned val = edma_read(ecc, offset);
294
295         val &= and;
296         val |= or;
297         edma_write(ecc, offset, val);
298 }
299
300 static inline void edma_and(struct edma_cc *ecc, int offset, unsigned and)
301 {
302         unsigned val = edma_read(ecc, offset);
303
304         val &= and;
305         edma_write(ecc, offset, val);
306 }
307
308 static inline void edma_or(struct edma_cc *ecc, int offset, unsigned or)
309 {
310         unsigned val = edma_read(ecc, offset);
311
312         val |= or;
313         edma_write(ecc, offset, val);
314 }
315
316 static inline unsigned int edma_read_array(struct edma_cc *ecc, int offset,
317                                            int i)
318 {
319         return edma_read(ecc, offset + (i << 2));
320 }
321
322 static inline void edma_write_array(struct edma_cc *ecc, int offset, int i,
323                                     unsigned val)
324 {
325         edma_write(ecc, offset + (i << 2), val);
326 }
327
328 static inline void edma_modify_array(struct edma_cc *ecc, int offset, int i,
329                                      unsigned and, unsigned or)
330 {
331         edma_modify(ecc, offset + (i << 2), and, or);
332 }
333
334 static inline void edma_or_array(struct edma_cc *ecc, int offset, int i,
335                                  unsigned or)
336 {
337         edma_or(ecc, offset + (i << 2), or);
338 }
339
340 static inline void edma_or_array2(struct edma_cc *ecc, int offset, int i, int j,
341                                   unsigned or)
342 {
343         edma_or(ecc, offset + ((i * 2 + j) << 2), or);
344 }
345
346 static inline void edma_write_array2(struct edma_cc *ecc, int offset, int i,
347                                      int j, unsigned val)
348 {
349         edma_write(ecc, offset + ((i * 2 + j) << 2), val);
350 }
351
352 static inline unsigned int edma_shadow0_read(struct edma_cc *ecc, int offset)
353 {
354         return edma_read(ecc, EDMA_SHADOW0 + offset);
355 }
356
357 static inline unsigned int edma_shadow0_read_array(struct edma_cc *ecc,
358                                                    int offset, int i)
359 {
360         return edma_read(ecc, EDMA_SHADOW0 + offset + (i << 2));
361 }
362
363 static inline void edma_shadow0_write(struct edma_cc *ecc, int offset,
364                                       unsigned val)
365 {
366         edma_write(ecc, EDMA_SHADOW0 + offset, val);
367 }
368
369 static inline void edma_shadow0_write_array(struct edma_cc *ecc, int offset,
370                                             int i, unsigned val)
371 {
372         edma_write(ecc, EDMA_SHADOW0 + offset + (i << 2), val);
373 }
374
375 static inline unsigned int edma_param_read(struct edma_cc *ecc, int offset,
376                                            int param_no)
377 {
378         return edma_read(ecc, EDMA_PARM + offset + (param_no << 5));
379 }
380
381 static inline void edma_param_write(struct edma_cc *ecc, int offset,
382                                     int param_no, unsigned val)
383 {
384         edma_write(ecc, EDMA_PARM + offset + (param_no << 5), val);
385 }
386
387 static inline void edma_param_modify(struct edma_cc *ecc, int offset,
388                                      int param_no, unsigned and, unsigned or)
389 {
390         edma_modify(ecc, EDMA_PARM + offset + (param_no << 5), and, or);
391 }
392
393 static inline void edma_param_and(struct edma_cc *ecc, int offset, int param_no,
394                                   unsigned and)
395 {
396         edma_and(ecc, EDMA_PARM + offset + (param_no << 5), and);
397 }
398
399 static inline void edma_param_or(struct edma_cc *ecc, int offset, int param_no,
400                                  unsigned or)
401 {
402         edma_or(ecc, EDMA_PARM + offset + (param_no << 5), or);
403 }
404
405 static inline void set_bits(int offset, int len, unsigned long *p)
406 {
407         for (; len > 0; len--)
408                 set_bit(offset + (len - 1), p);
409 }
410
411 static inline void clear_bits(int offset, int len, unsigned long *p)
412 {
413         for (; len > 0; len--)
414                 clear_bit(offset + (len - 1), p);
415 }
416
417 static void edma_assign_priority_to_queue(struct edma_cc *ecc, int queue_no,
418                                           int priority)
419 {
420         int bit = queue_no * 4;
421
422         edma_modify(ecc, EDMA_QUEPRI, ~(0x7 << bit), ((priority & 0x7) << bit));
423 }
424
425 static void edma_set_chmap(struct edma_chan *echan, int slot)
426 {
427         struct edma_cc *ecc = echan->ecc;
428         int channel = EDMA_CHAN_SLOT(echan->ch_num);
429
430         if (ecc->chmap_exist) {
431                 slot = EDMA_CHAN_SLOT(slot);
432                 edma_write_array(ecc, EDMA_DCHMAP, channel, (slot << 5));
433         }
434 }
435
436 static void edma_setup_interrupt(struct edma_chan *echan, bool enable)
437 {
438         struct edma_cc *ecc = echan->ecc;
439         int channel = EDMA_CHAN_SLOT(echan->ch_num);
440
441         if (enable) {
442                 edma_shadow0_write_array(ecc, SH_ICR, channel >> 5,
443                                          BIT(channel & 0x1f));
444                 edma_shadow0_write_array(ecc, SH_IESR, channel >> 5,
445                                          BIT(channel & 0x1f));
446         } else {
447                 edma_shadow0_write_array(ecc, SH_IECR, channel >> 5,
448                                          BIT(channel & 0x1f));
449         }
450 }
451
452 /*
453  * paRAM slot management functions
454  */
455 static void edma_write_slot(struct edma_cc *ecc, unsigned slot,
456                             const struct edmacc_param *param)
457 {
458         slot = EDMA_CHAN_SLOT(slot);
459         if (slot >= ecc->num_slots)
460                 return;
461         memcpy_toio(ecc->base + PARM_OFFSET(slot), param, PARM_SIZE);
462 }
463
464 static void edma_read_slot(struct edma_cc *ecc, unsigned slot,
465                            struct edmacc_param *param)
466 {
467         slot = EDMA_CHAN_SLOT(slot);
468         if (slot >= ecc->num_slots)
469                 return;
470         memcpy_fromio(param, ecc->base + PARM_OFFSET(slot), PARM_SIZE);
471 }
472
473 /**
474  * edma_alloc_slot - allocate DMA parameter RAM
475  * @ecc: pointer to edma_cc struct
476  * @slot: specific slot to allocate; negative for "any unused slot"
477  *
478  * This allocates a parameter RAM slot, initializing it to hold a
479  * dummy transfer.  Slots allocated using this routine have not been
480  * mapped to a hardware DMA channel, and will normally be used by
481  * linking to them from a slot associated with a DMA channel.
482  *
483  * Normal use is to pass EDMA_SLOT_ANY as the @slot, but specific
484  * slots may be allocated on behalf of DSP firmware.
485  *
486  * Returns the number of the slot, else negative errno.
487  */
488 static int edma_alloc_slot(struct edma_cc *ecc, int slot)
489 {
490         if (slot >= 0) {
491                 slot = EDMA_CHAN_SLOT(slot);
492                 /* Requesting entry paRAM slot for a HW triggered channel. */
493                 if (ecc->chmap_exist && slot < ecc->num_channels)
494                         slot = EDMA_SLOT_ANY;
495         }
496
497         if (slot < 0) {
498                 if (ecc->chmap_exist)
499                         slot = 0;
500                 else
501                         slot = ecc->num_channels;
502                 for (;;) {
503                         slot = find_next_zero_bit(ecc->slot_inuse,
504                                                   ecc->num_slots,
505                                                   slot);
506                         if (slot == ecc->num_slots)
507                                 return -ENOMEM;
508                         if (!test_and_set_bit(slot, ecc->slot_inuse))
509                                 break;
510                 }
511         } else if (slot >= ecc->num_slots) {
512                 return -EINVAL;
513         } else if (test_and_set_bit(slot, ecc->slot_inuse)) {
514                 return -EBUSY;
515         }
516
517         edma_write_slot(ecc, slot, &dummy_paramset);
518
519         return EDMA_CTLR_CHAN(ecc->id, slot);
520 }
521
522 static void edma_free_slot(struct edma_cc *ecc, unsigned slot)
523 {
524         slot = EDMA_CHAN_SLOT(slot);
525         if (slot >= ecc->num_slots)
526                 return;
527
528         edma_write_slot(ecc, slot, &dummy_paramset);
529         clear_bit(slot, ecc->slot_inuse);
530 }
531
532 /**
533  * edma_link - link one parameter RAM slot to another
534  * @ecc: pointer to edma_cc struct
535  * @from: parameter RAM slot originating the link
536  * @to: parameter RAM slot which is the link target
537  *
538  * The originating slot should not be part of any active DMA transfer.
539  */
540 static void edma_link(struct edma_cc *ecc, unsigned from, unsigned to)
541 {
542         if (unlikely(EDMA_CTLR(from) != EDMA_CTLR(to)))
543                 dev_warn(ecc->dev, "Ignoring eDMA instance for linking\n");
544
545         from = EDMA_CHAN_SLOT(from);
546         to = EDMA_CHAN_SLOT(to);
547         if (from >= ecc->num_slots || to >= ecc->num_slots)
548                 return;
549
550         edma_param_modify(ecc, PARM_LINK_BCNTRLD, from, 0xffff0000,
551                           PARM_OFFSET(to));
552 }
553
554 /**
555  * edma_get_position - returns the current transfer point
556  * @ecc: pointer to edma_cc struct
557  * @slot: parameter RAM slot being examined
558  * @dst:  true selects the dest position, false the source
559  *
560  * Returns the position of the current active slot
561  */
562 static dma_addr_t edma_get_position(struct edma_cc *ecc, unsigned slot,
563                                     bool dst)
564 {
565         u32 offs;
566
567         slot = EDMA_CHAN_SLOT(slot);
568         offs = PARM_OFFSET(slot);
569         offs += dst ? PARM_DST : PARM_SRC;
570
571         return edma_read(ecc, offs);
572 }
573
574 /*
575  * Channels with event associations will be triggered by their hardware
576  * events, and channels without such associations will be triggered by
577  * software.  (At this writing there is no interface for using software
578  * triggers except with channels that don't support hardware triggers.)
579  */
580 static void edma_start(struct edma_chan *echan)
581 {
582         struct edma_cc *ecc = echan->ecc;
583         int channel = EDMA_CHAN_SLOT(echan->ch_num);
584         int j = (channel >> 5);
585         unsigned int mask = BIT(channel & 0x1f);
586
587         if (!echan->hw_triggered) {
588                 /* EDMA channels without event association */
589                 dev_dbg(ecc->dev, "ESR%d %08x\n", j,
590                         edma_shadow0_read_array(ecc, SH_ESR, j));
591                 edma_shadow0_write_array(ecc, SH_ESR, j, mask);
592         } else {
593                 /* EDMA channel with event association */
594                 dev_dbg(ecc->dev, "ER%d %08x\n", j,
595                         edma_shadow0_read_array(ecc, SH_ER, j));
596                 /* Clear any pending event or error */
597                 edma_write_array(ecc, EDMA_ECR, j, mask);
598                 edma_write_array(ecc, EDMA_EMCR, j, mask);
599                 /* Clear any SER */
600                 edma_shadow0_write_array(ecc, SH_SECR, j, mask);
601                 edma_shadow0_write_array(ecc, SH_EESR, j, mask);
602                 dev_dbg(ecc->dev, "EER%d %08x\n", j,
603                         edma_shadow0_read_array(ecc, SH_EER, j));
604         }
605 }
606
607 static void edma_stop(struct edma_chan *echan)
608 {
609         struct edma_cc *ecc = echan->ecc;
610         int channel = EDMA_CHAN_SLOT(echan->ch_num);
611         int j = (channel >> 5);
612         unsigned int mask = BIT(channel & 0x1f);
613
614         edma_shadow0_write_array(ecc, SH_EECR, j, mask);
615         edma_shadow0_write_array(ecc, SH_ECR, j, mask);
616         edma_shadow0_write_array(ecc, SH_SECR, j, mask);
617         edma_write_array(ecc, EDMA_EMCR, j, mask);
618
619         /* clear possibly pending completion interrupt */
620         edma_shadow0_write_array(ecc, SH_ICR, j, mask);
621
622         dev_dbg(ecc->dev, "EER%d %08x\n", j,
623                 edma_shadow0_read_array(ecc, SH_EER, j));
624
625         /* REVISIT:  consider guarding against inappropriate event
626          * chaining by overwriting with dummy_paramset.
627          */
628 }
629
630 /*
631  * Temporarily disable EDMA hardware events on the specified channel,
632  * preventing them from triggering new transfers
633  */
634 static void edma_pause(struct edma_chan *echan)
635 {
636         int channel = EDMA_CHAN_SLOT(echan->ch_num);
637         unsigned int mask = BIT(channel & 0x1f);
638
639         edma_shadow0_write_array(echan->ecc, SH_EECR, channel >> 5, mask);
640 }
641
642 /* Re-enable EDMA hardware events on the specified channel.  */
643 static void edma_resume(struct edma_chan *echan)
644 {
645         int channel = EDMA_CHAN_SLOT(echan->ch_num);
646         unsigned int mask = BIT(channel & 0x1f);
647
648         edma_shadow0_write_array(echan->ecc, SH_EESR, channel >> 5, mask);
649 }
650
651 static void edma_trigger_channel(struct edma_chan *echan)
652 {
653         struct edma_cc *ecc = echan->ecc;
654         int channel = EDMA_CHAN_SLOT(echan->ch_num);
655         unsigned int mask = BIT(channel & 0x1f);
656
657         edma_shadow0_write_array(ecc, SH_ESR, (channel >> 5), mask);
658
659         dev_dbg(ecc->dev, "ESR%d %08x\n", (channel >> 5),
660                 edma_shadow0_read_array(ecc, SH_ESR, (channel >> 5)));
661 }
662
663 static void edma_clean_channel(struct edma_chan *echan)
664 {
665         struct edma_cc *ecc = echan->ecc;
666         int channel = EDMA_CHAN_SLOT(echan->ch_num);
667         int j = (channel >> 5);
668         unsigned int mask = BIT(channel & 0x1f);
669
670         dev_dbg(ecc->dev, "EMR%d %08x\n", j, edma_read_array(ecc, EDMA_EMR, j));
671         edma_shadow0_write_array(ecc, SH_ECR, j, mask);
672         /* Clear the corresponding EMR bits */
673         edma_write_array(ecc, EDMA_EMCR, j, mask);
674         /* Clear any SER */
675         edma_shadow0_write_array(ecc, SH_SECR, j, mask);
676         edma_write(ecc, EDMA_CCERRCLR, BIT(16) | BIT(1) | BIT(0));
677 }
678
679 /* Move channel to a specific event queue */
680 static void edma_assign_channel_eventq(struct edma_chan *echan,
681                                        enum dma_event_q eventq_no)
682 {
683         struct edma_cc *ecc = echan->ecc;
684         int channel = EDMA_CHAN_SLOT(echan->ch_num);
685         int bit = (channel & 0x7) * 4;
686
687         /* default to low priority queue */
688         if (eventq_no == EVENTQ_DEFAULT)
689                 eventq_no = ecc->default_queue;
690         if (eventq_no >= ecc->num_tc)
691                 return;
692
693         eventq_no &= 7;
694         edma_modify_array(ecc, EDMA_DMAQNUM, (channel >> 3), ~(0x7 << bit),
695                           eventq_no << bit);
696 }
697
698 static int edma_alloc_channel(struct edma_chan *echan,
699                               enum dma_event_q eventq_no)
700 {
701         struct edma_cc *ecc = echan->ecc;
702         int channel = EDMA_CHAN_SLOT(echan->ch_num);
703
704         /* ensure access through shadow region 0 */
705         edma_or_array2(ecc, EDMA_DRAE, 0, channel >> 5, BIT(channel & 0x1f));
706
707         /* ensure no events are pending */
708         edma_stop(echan);
709
710         edma_setup_interrupt(echan, true);
711
712         edma_assign_channel_eventq(echan, eventq_no);
713
714         return 0;
715 }
716
717 static void edma_free_channel(struct edma_chan *echan)
718 {
719         /* ensure no events are pending */
720         edma_stop(echan);
721         /* REVISIT should probably take out of shadow region 0 */
722         edma_setup_interrupt(echan, false);
723 }
724
725 static inline struct edma_cc *to_edma_cc(struct dma_device *d)
726 {
727         return container_of(d, struct edma_cc, dma_slave);
728 }
729
730 static inline struct edma_chan *to_edma_chan(struct dma_chan *c)
731 {
732         return container_of(c, struct edma_chan, vchan.chan);
733 }
734
735 static inline struct edma_desc *to_edma_desc(struct dma_async_tx_descriptor *tx)
736 {
737         return container_of(tx, struct edma_desc, vdesc.tx);
738 }
739
740 static void edma_desc_free(struct virt_dma_desc *vdesc)
741 {
742         kfree(container_of(vdesc, struct edma_desc, vdesc));
743 }
744
745 /* Dispatch a queued descriptor to the controller (caller holds lock) */
746 static void edma_execute(struct edma_chan *echan)
747 {
748         struct edma_cc *ecc = echan->ecc;
749         struct virt_dma_desc *vdesc;
750         struct edma_desc *edesc;
751         struct device *dev = echan->vchan.chan.device->dev;
752         int i, j, left, nslots;
753
754         if (!echan->edesc) {
755                 /* Setup is needed for the first transfer */
756                 vdesc = vchan_next_desc(&echan->vchan);
757                 if (!vdesc)
758                         return;
759                 list_del(&vdesc->node);
760                 echan->edesc = to_edma_desc(&vdesc->tx);
761         }
762
763         edesc = echan->edesc;
764
765         /* Find out how many left */
766         left = edesc->pset_nr - edesc->processed;
767         nslots = min(MAX_NR_SG, left);
768         edesc->sg_len = 0;
769
770         /* Write descriptor PaRAM set(s) */
771         for (i = 0; i < nslots; i++) {
772                 j = i + edesc->processed;
773                 edma_write_slot(ecc, echan->slot[i], &edesc->pset[j].param);
774                 edesc->sg_len += edesc->pset[j].len;
775                 dev_vdbg(dev,
776                          "\n pset[%d]:\n"
777                          "  chnum\t%d\n"
778                          "  slot\t%d\n"
779                          "  opt\t%08x\n"
780                          "  src\t%08x\n"
781                          "  dst\t%08x\n"
782                          "  abcnt\t%08x\n"
783                          "  ccnt\t%08x\n"
784                          "  bidx\t%08x\n"
785                          "  cidx\t%08x\n"
786                          "  lkrld\t%08x\n",
787                          j, echan->ch_num, echan->slot[i],
788                          edesc->pset[j].param.opt,
789                          edesc->pset[j].param.src,
790                          edesc->pset[j].param.dst,
791                          edesc->pset[j].param.a_b_cnt,
792                          edesc->pset[j].param.ccnt,
793                          edesc->pset[j].param.src_dst_bidx,
794                          edesc->pset[j].param.src_dst_cidx,
795                          edesc->pset[j].param.link_bcntrld);
796                 /* Link to the previous slot if not the last set */
797                 if (i != (nslots - 1))
798                         edma_link(ecc, echan->slot[i], echan->slot[i + 1]);
799         }
800
801         edesc->processed += nslots;
802
803         /*
804          * If this is either the last set in a set of SG-list transactions
805          * then setup a link to the dummy slot, this results in all future
806          * events being absorbed and that's OK because we're done
807          */
808         if (edesc->processed == edesc->pset_nr) {
809                 if (edesc->cyclic)
810                         edma_link(ecc, echan->slot[nslots - 1], echan->slot[1]);
811                 else
812                         edma_link(ecc, echan->slot[nslots - 1],
813                                   echan->ecc->dummy_slot);
814         }
815
816         if (echan->missed) {
817                 /*
818                  * This happens due to setup times between intermediate
819                  * transfers in long SG lists which have to be broken up into
820                  * transfers of MAX_NR_SG
821                  */
822                 dev_dbg(dev, "missed event on channel %d\n", echan->ch_num);
823                 edma_clean_channel(echan);
824                 edma_stop(echan);
825                 edma_start(echan);
826                 edma_trigger_channel(echan);
827                 echan->missed = 0;
828         } else if (edesc->processed <= MAX_NR_SG) {
829                 dev_dbg(dev, "first transfer starting on channel %d\n",
830                         echan->ch_num);
831                 edma_start(echan);
832         } else {
833                 dev_dbg(dev, "chan: %d: completed %d elements, resuming\n",
834                         echan->ch_num, edesc->processed);
835                 edma_resume(echan);
836         }
837 }
838
839 static int edma_terminate_all(struct dma_chan *chan)
840 {
841         struct edma_chan *echan = to_edma_chan(chan);
842         unsigned long flags;
843         LIST_HEAD(head);
844
845         spin_lock_irqsave(&echan->vchan.lock, flags);
846
847         /*
848          * Stop DMA activity: we assume the callback will not be called
849          * after edma_dma() returns (even if it does, it will see
850          * echan->edesc is NULL and exit.)
851          */
852         if (echan->edesc) {
853                 edma_stop(echan);
854                 /* Move the cyclic channel back to default queue */
855                 if (!echan->tc && echan->edesc->cyclic)
856                         edma_assign_channel_eventq(echan, EVENTQ_DEFAULT);
857                 /*
858                  * free the running request descriptor
859                  * since it is not in any of the vdesc lists
860                  */
861                 edma_desc_free(&echan->edesc->vdesc);
862                 echan->edesc = NULL;
863         }
864
865         vchan_get_all_descriptors(&echan->vchan, &head);
866         spin_unlock_irqrestore(&echan->vchan.lock, flags);
867         vchan_dma_desc_free_list(&echan->vchan, &head);
868
869         return 0;
870 }
871
872 static int edma_slave_config(struct dma_chan *chan,
873         struct dma_slave_config *cfg)
874 {
875         struct edma_chan *echan = to_edma_chan(chan);
876
877         if (cfg->src_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES ||
878             cfg->dst_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES)
879                 return -EINVAL;
880
881         memcpy(&echan->cfg, cfg, sizeof(echan->cfg));
882
883         return 0;
884 }
885
886 static int edma_dma_pause(struct dma_chan *chan)
887 {
888         struct edma_chan *echan = to_edma_chan(chan);
889
890         if (!echan->edesc)
891                 return -EINVAL;
892
893         edma_pause(echan);
894         return 0;
895 }
896
897 static int edma_dma_resume(struct dma_chan *chan)
898 {
899         struct edma_chan *echan = to_edma_chan(chan);
900
901         edma_resume(echan);
902         return 0;
903 }
904
905 /*
906  * A PaRAM set configuration abstraction used by other modes
907  * @chan: Channel who's PaRAM set we're configuring
908  * @pset: PaRAM set to initialize and setup.
909  * @src_addr: Source address of the DMA
910  * @dst_addr: Destination address of the DMA
911  * @burst: In units of dev_width, how much to send
912  * @dev_width: How much is the dev_width
913  * @dma_length: Total length of the DMA transfer
914  * @direction: Direction of the transfer
915  */
916 static int edma_config_pset(struct dma_chan *chan, struct edma_pset *epset,
917                             dma_addr_t src_addr, dma_addr_t dst_addr, u32 burst,
918                             unsigned int acnt, unsigned int dma_length,
919                             enum dma_transfer_direction direction)
920 {
921         struct edma_chan *echan = to_edma_chan(chan);
922         struct device *dev = chan->device->dev;
923         struct edmacc_param *param = &epset->param;
924         int bcnt, ccnt, cidx;
925         int src_bidx, dst_bidx, src_cidx, dst_cidx;
926         int absync;
927
928         /* src/dst_maxburst == 0 is the same case as src/dst_maxburst == 1 */
929         if (!burst)
930                 burst = 1;
931         /*
932          * If the maxburst is equal to the fifo width, use
933          * A-synced transfers. This allows for large contiguous
934          * buffer transfers using only one PaRAM set.
935          */
936         if (burst == 1) {
937                 /*
938                  * For the A-sync case, bcnt and ccnt are the remainder
939                  * and quotient respectively of the division of:
940                  * (dma_length / acnt) by (SZ_64K -1). This is so
941                  * that in case bcnt over flows, we have ccnt to use.
942                  * Note: In A-sync tranfer only, bcntrld is used, but it
943                  * only applies for sg_dma_len(sg) >= SZ_64K.
944                  * In this case, the best way adopted is- bccnt for the
945                  * first frame will be the remainder below. Then for
946                  * every successive frame, bcnt will be SZ_64K-1. This
947                  * is assured as bcntrld = 0xffff in end of function.
948                  */
949                 absync = false;
950                 ccnt = dma_length / acnt / (SZ_64K - 1);
951                 bcnt = dma_length / acnt - ccnt * (SZ_64K - 1);
952                 /*
953                  * If bcnt is non-zero, we have a remainder and hence an
954                  * extra frame to transfer, so increment ccnt.
955                  */
956                 if (bcnt)
957                         ccnt++;
958                 else
959                         bcnt = SZ_64K - 1;
960                 cidx = acnt;
961         } else {
962                 /*
963                  * If maxburst is greater than the fifo address_width,
964                  * use AB-synced transfers where A count is the fifo
965                  * address_width and B count is the maxburst. In this
966                  * case, we are limited to transfers of C count frames
967                  * of (address_width * maxburst) where C count is limited
968                  * to SZ_64K-1. This places an upper bound on the length
969                  * of an SG segment that can be handled.
970                  */
971                 absync = true;
972                 bcnt = burst;
973                 ccnt = dma_length / (acnt * bcnt);
974                 if (ccnt > (SZ_64K - 1)) {
975                         dev_err(dev, "Exceeded max SG segment size\n");
976                         return -EINVAL;
977                 }
978                 cidx = acnt * bcnt;
979         }
980
981         epset->len = dma_length;
982
983         if (direction == DMA_MEM_TO_DEV) {
984                 src_bidx = acnt;
985                 src_cidx = cidx;
986                 dst_bidx = 0;
987                 dst_cidx = 0;
988                 epset->addr = src_addr;
989         } else if (direction == DMA_DEV_TO_MEM)  {
990                 src_bidx = 0;
991                 src_cidx = 0;
992                 dst_bidx = acnt;
993                 dst_cidx = cidx;
994                 epset->addr = dst_addr;
995         } else if (direction == DMA_MEM_TO_MEM)  {
996                 src_bidx = acnt;
997                 src_cidx = cidx;
998                 dst_bidx = acnt;
999                 dst_cidx = cidx;
1000         } else {
1001                 dev_err(dev, "%s: direction not implemented yet\n", __func__);
1002                 return -EINVAL;
1003         }
1004
1005         param->opt = EDMA_TCC(EDMA_CHAN_SLOT(echan->ch_num));
1006         /* Configure A or AB synchronized transfers */
1007         if (absync)
1008                 param->opt |= SYNCDIM;
1009
1010         param->src = src_addr;
1011         param->dst = dst_addr;
1012
1013         param->src_dst_bidx = (dst_bidx << 16) | src_bidx;
1014         param->src_dst_cidx = (dst_cidx << 16) | src_cidx;
1015
1016         param->a_b_cnt = bcnt << 16 | acnt;
1017         param->ccnt = ccnt;
1018         /*
1019          * Only time when (bcntrld) auto reload is required is for
1020          * A-sync case, and in this case, a requirement of reload value
1021          * of SZ_64K-1 only is assured. 'link' is initially set to NULL
1022          * and then later will be populated by edma_execute.
1023          */
1024         param->link_bcntrld = 0xffffffff;
1025         return absync;
1026 }
1027
1028 static struct dma_async_tx_descriptor *edma_prep_slave_sg(
1029         struct dma_chan *chan, struct scatterlist *sgl,
1030         unsigned int sg_len, enum dma_transfer_direction direction,
1031         unsigned long tx_flags, void *context)
1032 {
1033         struct edma_chan *echan = to_edma_chan(chan);
1034         struct device *dev = chan->device->dev;
1035         struct edma_desc *edesc;
1036         dma_addr_t src_addr = 0, dst_addr = 0;
1037         enum dma_slave_buswidth dev_width;
1038         u32 burst;
1039         struct scatterlist *sg;
1040         int i, nslots, ret;
1041
1042         if (unlikely(!echan || !sgl || !sg_len))
1043                 return NULL;
1044
1045         if (direction == DMA_DEV_TO_MEM) {
1046                 src_addr = echan->cfg.src_addr;
1047                 dev_width = echan->cfg.src_addr_width;
1048                 burst = echan->cfg.src_maxburst;
1049         } else if (direction == DMA_MEM_TO_DEV) {
1050                 dst_addr = echan->cfg.dst_addr;
1051                 dev_width = echan->cfg.dst_addr_width;
1052                 burst = echan->cfg.dst_maxburst;
1053         } else {
1054                 dev_err(dev, "%s: bad direction: %d\n", __func__, direction);
1055                 return NULL;
1056         }
1057
1058         if (dev_width == DMA_SLAVE_BUSWIDTH_UNDEFINED) {
1059                 dev_err(dev, "%s: Undefined slave buswidth\n", __func__);
1060                 return NULL;
1061         }
1062
1063         edesc = kzalloc(sizeof(*edesc) + sg_len * sizeof(edesc->pset[0]),
1064                         GFP_ATOMIC);
1065         if (!edesc) {
1066                 dev_err(dev, "%s: Failed to allocate a descriptor\n", __func__);
1067                 return NULL;
1068         }
1069
1070         edesc->pset_nr = sg_len;
1071         edesc->residue = 0;
1072         edesc->direction = direction;
1073         edesc->echan = echan;
1074
1075         /* Allocate a PaRAM slot, if needed */
1076         nslots = min_t(unsigned, MAX_NR_SG, sg_len);
1077
1078         for (i = 0; i < nslots; i++) {
1079                 if (echan->slot[i] < 0) {
1080                         echan->slot[i] =
1081                                 edma_alloc_slot(echan->ecc, EDMA_SLOT_ANY);
1082                         if (echan->slot[i] < 0) {
1083                                 kfree(edesc);
1084                                 dev_err(dev, "%s: Failed to allocate slot\n",
1085                                         __func__);
1086                                 return NULL;
1087                         }
1088                 }
1089         }
1090
1091         /* Configure PaRAM sets for each SG */
1092         for_each_sg(sgl, sg, sg_len, i) {
1093                 /* Get address for each SG */
1094                 if (direction == DMA_DEV_TO_MEM)
1095                         dst_addr = sg_dma_address(sg);
1096                 else
1097                         src_addr = sg_dma_address(sg);
1098
1099                 ret = edma_config_pset(chan, &edesc->pset[i], src_addr,
1100                                        dst_addr, burst, dev_width,
1101                                        sg_dma_len(sg), direction);
1102                 if (ret < 0) {
1103                         kfree(edesc);
1104                         return NULL;
1105                 }
1106
1107                 edesc->absync = ret;
1108                 edesc->residue += sg_dma_len(sg);
1109
1110                 /* If this is the last in a current SG set of transactions,
1111                    enable interrupts so that next set is processed */
1112                 if (!((i+1) % MAX_NR_SG))
1113                         edesc->pset[i].param.opt |= TCINTEN;
1114
1115                 /* If this is the last set, enable completion interrupt flag */
1116                 if (i == sg_len - 1)
1117                         edesc->pset[i].param.opt |= TCINTEN;
1118         }
1119         edesc->residue_stat = edesc->residue;
1120
1121         return vchan_tx_prep(&echan->vchan, &edesc->vdesc, tx_flags);
1122 }
1123
1124 static struct dma_async_tx_descriptor *edma_prep_dma_memcpy(
1125         struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
1126         size_t len, unsigned long tx_flags)
1127 {
1128         int ret, nslots;
1129         struct edma_desc *edesc;
1130         struct device *dev = chan->device->dev;
1131         struct edma_chan *echan = to_edma_chan(chan);
1132         unsigned int width, pset_len;
1133
1134         if (unlikely(!echan || !len))
1135                 return NULL;
1136
1137         if (len < SZ_64K) {
1138                 /*
1139                  * Transfer size less than 64K can be handled with one paRAM
1140                  * slot and with one burst.
1141                  * ACNT = length
1142                  */
1143                 width = len;
1144                 pset_len = len;
1145                 nslots = 1;
1146         } else {
1147                 /*
1148                  * Transfer size bigger than 64K will be handled with maximum of
1149                  * two paRAM slots.
1150                  * slot1: (full_length / 32767) times 32767 bytes bursts.
1151                  *        ACNT = 32767, length1: (full_length / 32767) * 32767
1152                  * slot2: the remaining amount of data after slot1.
1153                  *        ACNT = full_length - length1, length2 = ACNT
1154                  *
1155                  * When the full_length is multibple of 32767 one slot can be
1156                  * used to complete the transfer.
1157                  */
1158                 width = SZ_32K - 1;
1159                 pset_len = rounddown(len, width);
1160                 /* One slot is enough for lengths multiple of (SZ_32K -1) */
1161                 if (unlikely(pset_len == len))
1162                         nslots = 1;
1163                 else
1164                         nslots = 2;
1165         }
1166
1167         edesc = kzalloc(sizeof(*edesc) + nslots * sizeof(edesc->pset[0]),
1168                         GFP_ATOMIC);
1169         if (!edesc) {
1170                 dev_dbg(dev, "Failed to allocate a descriptor\n");
1171                 return NULL;
1172         }
1173
1174         edesc->pset_nr = nslots;
1175         edesc->residue = edesc->residue_stat = len;
1176         edesc->direction = DMA_MEM_TO_MEM;
1177         edesc->echan = echan;
1178
1179         ret = edma_config_pset(chan, &edesc->pset[0], src, dest, 1,
1180                                width, pset_len, DMA_MEM_TO_MEM);
1181         if (ret < 0) {
1182                 kfree(edesc);
1183                 return NULL;
1184         }
1185
1186         edesc->absync = ret;
1187
1188         edesc->pset[0].param.opt |= ITCCHEN;
1189         if (nslots == 1) {
1190                 /* Enable transfer complete interrupt */
1191                 edesc->pset[0].param.opt |= TCINTEN;
1192         } else {
1193                 /* Enable transfer complete chaining for the first slot */
1194                 edesc->pset[0].param.opt |= TCCHEN;
1195
1196                 if (echan->slot[1] < 0) {
1197                         echan->slot[1] = edma_alloc_slot(echan->ecc,
1198                                                          EDMA_SLOT_ANY);
1199                         if (echan->slot[1] < 0) {
1200                                 kfree(edesc);
1201                                 dev_err(dev, "%s: Failed to allocate slot\n",
1202                                         __func__);
1203                                 return NULL;
1204                         }
1205                 }
1206                 dest += pset_len;
1207                 src += pset_len;
1208                 pset_len = width = len % (SZ_32K - 1);
1209
1210                 ret = edma_config_pset(chan, &edesc->pset[1], src, dest, 1,
1211                                        width, pset_len, DMA_MEM_TO_MEM);
1212                 if (ret < 0) {
1213                         kfree(edesc);
1214                         return NULL;
1215                 }
1216
1217                 edesc->pset[1].param.opt |= ITCCHEN;
1218                 edesc->pset[1].param.opt |= TCINTEN;
1219         }
1220
1221         return vchan_tx_prep(&echan->vchan, &edesc->vdesc, tx_flags);
1222 }
1223
1224 static struct dma_async_tx_descriptor *edma_prep_dma_cyclic(
1225         struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len,
1226         size_t period_len, enum dma_transfer_direction direction,
1227         unsigned long tx_flags)
1228 {
1229         struct edma_chan *echan = to_edma_chan(chan);
1230         struct device *dev = chan->device->dev;
1231         struct edma_desc *edesc;
1232         dma_addr_t src_addr, dst_addr;
1233         enum dma_slave_buswidth dev_width;
1234         u32 burst;
1235         int i, ret, nslots;
1236
1237         if (unlikely(!echan || !buf_len || !period_len))
1238                 return NULL;
1239
1240         if (direction == DMA_DEV_TO_MEM) {
1241                 src_addr = echan->cfg.src_addr;
1242                 dst_addr = buf_addr;
1243                 dev_width = echan->cfg.src_addr_width;
1244                 burst = echan->cfg.src_maxburst;
1245         } else if (direction == DMA_MEM_TO_DEV) {
1246                 src_addr = buf_addr;
1247                 dst_addr = echan->cfg.dst_addr;
1248                 dev_width = echan->cfg.dst_addr_width;
1249                 burst = echan->cfg.dst_maxburst;
1250         } else {
1251                 dev_err(dev, "%s: bad direction: %d\n", __func__, direction);
1252                 return NULL;
1253         }
1254
1255         if (dev_width == DMA_SLAVE_BUSWIDTH_UNDEFINED) {
1256                 dev_err(dev, "%s: Undefined slave buswidth\n", __func__);
1257                 return NULL;
1258         }
1259
1260         if (unlikely(buf_len % period_len)) {
1261                 dev_err(dev, "Period should be multiple of Buffer length\n");
1262                 return NULL;
1263         }
1264
1265         nslots = (buf_len / period_len) + 1;
1266
1267         /*
1268          * Cyclic DMA users such as audio cannot tolerate delays introduced
1269          * by cases where the number of periods is more than the maximum
1270          * number of SGs the EDMA driver can handle at a time. For DMA types
1271          * such as Slave SGs, such delays are tolerable and synchronized,
1272          * but the synchronization is difficult to achieve with Cyclic and
1273          * cannot be guaranteed, so we error out early.
1274          */
1275         if (nslots > MAX_NR_SG)
1276                 return NULL;
1277
1278         edesc = kzalloc(sizeof(*edesc) + nslots * sizeof(edesc->pset[0]),
1279                         GFP_ATOMIC);
1280         if (!edesc) {
1281                 dev_err(dev, "%s: Failed to allocate a descriptor\n", __func__);
1282                 return NULL;
1283         }
1284
1285         edesc->cyclic = 1;
1286         edesc->pset_nr = nslots;
1287         edesc->residue = edesc->residue_stat = buf_len;
1288         edesc->direction = direction;
1289         edesc->echan = echan;
1290
1291         dev_dbg(dev, "%s: channel=%d nslots=%d period_len=%zu buf_len=%zu\n",
1292                 __func__, echan->ch_num, nslots, period_len, buf_len);
1293
1294         for (i = 0; i < nslots; i++) {
1295                 /* Allocate a PaRAM slot, if needed */
1296                 if (echan->slot[i] < 0) {
1297                         echan->slot[i] =
1298                                 edma_alloc_slot(echan->ecc, EDMA_SLOT_ANY);
1299                         if (echan->slot[i] < 0) {
1300                                 kfree(edesc);
1301                                 dev_err(dev, "%s: Failed to allocate slot\n",
1302                                         __func__);
1303                                 return NULL;
1304                         }
1305                 }
1306
1307                 if (i == nslots - 1) {
1308                         memcpy(&edesc->pset[i], &edesc->pset[0],
1309                                sizeof(edesc->pset[0]));
1310                         break;
1311                 }
1312
1313                 ret = edma_config_pset(chan, &edesc->pset[i], src_addr,
1314                                        dst_addr, burst, dev_width, period_len,
1315                                        direction);
1316                 if (ret < 0) {
1317                         kfree(edesc);
1318                         return NULL;
1319                 }
1320
1321                 if (direction == DMA_DEV_TO_MEM)
1322                         dst_addr += period_len;
1323                 else
1324                         src_addr += period_len;
1325
1326                 dev_vdbg(dev, "%s: Configure period %d of buf:\n", __func__, i);
1327                 dev_vdbg(dev,
1328                         "\n pset[%d]:\n"
1329                         "  chnum\t%d\n"
1330                         "  slot\t%d\n"
1331                         "  opt\t%08x\n"
1332                         "  src\t%08x\n"
1333                         "  dst\t%08x\n"
1334                         "  abcnt\t%08x\n"
1335                         "  ccnt\t%08x\n"
1336                         "  bidx\t%08x\n"
1337                         "  cidx\t%08x\n"
1338                         "  lkrld\t%08x\n",
1339                         i, echan->ch_num, echan->slot[i],
1340                         edesc->pset[i].param.opt,
1341                         edesc->pset[i].param.src,
1342                         edesc->pset[i].param.dst,
1343                         edesc->pset[i].param.a_b_cnt,
1344                         edesc->pset[i].param.ccnt,
1345                         edesc->pset[i].param.src_dst_bidx,
1346                         edesc->pset[i].param.src_dst_cidx,
1347                         edesc->pset[i].param.link_bcntrld);
1348
1349                 edesc->absync = ret;
1350
1351                 /*
1352                  * Enable period interrupt only if it is requested
1353                  */
1354                 if (tx_flags & DMA_PREP_INTERRUPT)
1355                         edesc->pset[i].param.opt |= TCINTEN;
1356         }
1357
1358         /* Place the cyclic channel to highest priority queue */
1359         if (!echan->tc)
1360                 edma_assign_channel_eventq(echan, EVENTQ_0);
1361
1362         return vchan_tx_prep(&echan->vchan, &edesc->vdesc, tx_flags);
1363 }
1364
1365 static void edma_completion_handler(struct edma_chan *echan)
1366 {
1367         struct device *dev = echan->vchan.chan.device->dev;
1368         struct edma_desc *edesc = echan->edesc;
1369
1370         if (!edesc)
1371                 return;
1372
1373         spin_lock(&echan->vchan.lock);
1374         if (edesc->cyclic) {
1375                 vchan_cyclic_callback(&edesc->vdesc);
1376                 spin_unlock(&echan->vchan.lock);
1377                 return;
1378         } else if (edesc->processed == edesc->pset_nr) {
1379                 edesc->residue = 0;
1380                 edma_stop(echan);
1381                 vchan_cookie_complete(&edesc->vdesc);
1382                 echan->edesc = NULL;
1383
1384                 dev_dbg(dev, "Transfer completed on channel %d\n",
1385                         echan->ch_num);
1386         } else {
1387                 dev_dbg(dev, "Sub transfer completed on channel %d\n",
1388                         echan->ch_num);
1389
1390                 edma_pause(echan);
1391
1392                 /* Update statistics for tx_status */
1393                 edesc->residue -= edesc->sg_len;
1394                 edesc->residue_stat = edesc->residue;
1395                 edesc->processed_stat = edesc->processed;
1396         }
1397         edma_execute(echan);
1398
1399         spin_unlock(&echan->vchan.lock);
1400 }
1401
1402 /* eDMA interrupt handler */
1403 static irqreturn_t dma_irq_handler(int irq, void *data)
1404 {
1405         struct edma_cc *ecc = data;
1406         int ctlr;
1407         u32 sh_ier;
1408         u32 sh_ipr;
1409         u32 bank;
1410
1411         ctlr = ecc->id;
1412         if (ctlr < 0)
1413                 return IRQ_NONE;
1414
1415         dev_vdbg(ecc->dev, "dma_irq_handler\n");
1416
1417         sh_ipr = edma_shadow0_read_array(ecc, SH_IPR, 0);
1418         if (!sh_ipr) {
1419                 sh_ipr = edma_shadow0_read_array(ecc, SH_IPR, 1);
1420                 if (!sh_ipr)
1421                         return IRQ_NONE;
1422                 sh_ier = edma_shadow0_read_array(ecc, SH_IER, 1);
1423                 bank = 1;
1424         } else {
1425                 sh_ier = edma_shadow0_read_array(ecc, SH_IER, 0);
1426                 bank = 0;
1427         }
1428
1429         do {
1430                 u32 slot;
1431                 u32 channel;
1432
1433                 slot = __ffs(sh_ipr);
1434                 sh_ipr &= ~(BIT(slot));
1435
1436                 if (sh_ier & BIT(slot)) {
1437                         channel = (bank << 5) | slot;
1438                         /* Clear the corresponding IPR bits */
1439                         edma_shadow0_write_array(ecc, SH_ICR, bank, BIT(slot));
1440                         edma_completion_handler(&ecc->slave_chans[channel]);
1441                 }
1442         } while (sh_ipr);
1443
1444         edma_shadow0_write(ecc, SH_IEVAL, 1);
1445         return IRQ_HANDLED;
1446 }
1447
1448 static void edma_error_handler(struct edma_chan *echan)
1449 {
1450         struct edma_cc *ecc = echan->ecc;
1451         struct device *dev = echan->vchan.chan.device->dev;
1452         struct edmacc_param p;
1453
1454         if (!echan->edesc)
1455                 return;
1456
1457         spin_lock(&echan->vchan.lock);
1458
1459         edma_read_slot(ecc, echan->slot[0], &p);
1460         /*
1461          * Issue later based on missed flag which will be sure
1462          * to happen as:
1463          * (1) we finished transmitting an intermediate slot and
1464          *     edma_execute is coming up.
1465          * (2) or we finished current transfer and issue will
1466          *     call edma_execute.
1467          *
1468          * Important note: issuing can be dangerous here and
1469          * lead to some nasty recursion when we are in a NULL
1470          * slot. So we avoid doing so and set the missed flag.
1471          */
1472         if (p.a_b_cnt == 0 && p.ccnt == 0) {
1473                 dev_dbg(dev, "Error on null slot, setting miss\n");
1474                 echan->missed = 1;
1475         } else {
1476                 /*
1477                  * The slot is already programmed but the event got
1478                  * missed, so its safe to issue it here.
1479                  */
1480                 dev_dbg(dev, "Missed event, TRIGGERING\n");
1481                 edma_clean_channel(echan);
1482                 edma_stop(echan);
1483                 edma_start(echan);
1484                 edma_trigger_channel(echan);
1485         }
1486         spin_unlock(&echan->vchan.lock);
1487 }
1488
1489 static inline bool edma_error_pending(struct edma_cc *ecc)
1490 {
1491         if (edma_read_array(ecc, EDMA_EMR, 0) ||
1492             edma_read_array(ecc, EDMA_EMR, 1) ||
1493             edma_read(ecc, EDMA_QEMR) || edma_read(ecc, EDMA_CCERR))
1494                 return true;
1495
1496         return false;
1497 }
1498
1499 /* eDMA error interrupt handler */
1500 static irqreturn_t dma_ccerr_handler(int irq, void *data)
1501 {
1502         struct edma_cc *ecc = data;
1503         int i, j;
1504         int ctlr;
1505         unsigned int cnt = 0;
1506         unsigned int val;
1507
1508         ctlr = ecc->id;
1509         if (ctlr < 0)
1510                 return IRQ_NONE;
1511
1512         dev_vdbg(ecc->dev, "dma_ccerr_handler\n");
1513
1514         if (!edma_error_pending(ecc))
1515                 return IRQ_NONE;
1516
1517         while (1) {
1518                 /* Event missed register(s) */
1519                 for (j = 0; j < 2; j++) {
1520                         unsigned long emr;
1521
1522                         val = edma_read_array(ecc, EDMA_EMR, j);
1523                         if (!val)
1524                                 continue;
1525
1526                         dev_dbg(ecc->dev, "EMR%d 0x%08x\n", j, val);
1527                         emr = val;
1528                         for (i = find_next_bit(&emr, 32, 0); i < 32;
1529                              i = find_next_bit(&emr, 32, i + 1)) {
1530                                 int k = (j << 5) + i;
1531
1532                                 /* Clear the corresponding EMR bits */
1533                                 edma_write_array(ecc, EDMA_EMCR, j, BIT(i));
1534                                 /* Clear any SER */
1535                                 edma_shadow0_write_array(ecc, SH_SECR, j,
1536                                                          BIT(i));
1537                                 edma_error_handler(&ecc->slave_chans[k]);
1538                         }
1539                 }
1540
1541                 val = edma_read(ecc, EDMA_QEMR);
1542                 if (val) {
1543                         dev_dbg(ecc->dev, "QEMR 0x%02x\n", val);
1544                         /* Not reported, just clear the interrupt reason. */
1545                         edma_write(ecc, EDMA_QEMCR, val);
1546                         edma_shadow0_write(ecc, SH_QSECR, val);
1547                 }
1548
1549                 val = edma_read(ecc, EDMA_CCERR);
1550                 if (val) {
1551                         dev_warn(ecc->dev, "CCERR 0x%08x\n", val);
1552                         /* Not reported, just clear the interrupt reason. */
1553                         edma_write(ecc, EDMA_CCERRCLR, val);
1554                 }
1555
1556                 if (!edma_error_pending(ecc))
1557                         break;
1558                 cnt++;
1559                 if (cnt > 10)
1560                         break;
1561         }
1562         edma_write(ecc, EDMA_EEVAL, 1);
1563         return IRQ_HANDLED;
1564 }
1565
1566 static void edma_tc_set_pm_state(struct edma_tc *tc, bool enable)
1567 {
1568         struct platform_device *tc_pdev;
1569         int ret;
1570
1571         if (!IS_ENABLED(CONFIG_OF) || !tc)
1572                 return;
1573
1574         tc_pdev = of_find_device_by_node(tc->node);
1575         if (!tc_pdev) {
1576                 pr_err("%s: TPTC device is not found\n", __func__);
1577                 return;
1578         }
1579         if (!pm_runtime_enabled(&tc_pdev->dev))
1580                 pm_runtime_enable(&tc_pdev->dev);
1581
1582         if (enable)
1583                 ret = pm_runtime_get_sync(&tc_pdev->dev);
1584         else
1585                 ret = pm_runtime_put_sync(&tc_pdev->dev);
1586
1587         if (ret < 0)
1588                 pr_err("%s: pm_runtime_%s_sync() failed for %s\n", __func__,
1589                        enable ? "get" : "put", dev_name(&tc_pdev->dev));
1590 }
1591
1592 /* Alloc channel resources */
1593 static int edma_alloc_chan_resources(struct dma_chan *chan)
1594 {
1595         struct edma_chan *echan = to_edma_chan(chan);
1596         struct edma_cc *ecc = echan->ecc;
1597         struct device *dev = ecc->dev;
1598         enum dma_event_q eventq_no = EVENTQ_DEFAULT;
1599         int ret;
1600
1601         if (echan->tc) {
1602                 eventq_no = echan->tc->id;
1603         } else if (ecc->tc_list) {
1604                 /* memcpy channel */
1605                 echan->tc = &ecc->tc_list[ecc->info->default_queue];
1606                 eventq_no = echan->tc->id;
1607         }
1608
1609         ret = edma_alloc_channel(echan, eventq_no);
1610         if (ret)
1611                 return ret;
1612
1613         echan->slot[0] = edma_alloc_slot(ecc, echan->ch_num);
1614         if (echan->slot[0] < 0) {
1615                 dev_err(dev, "Entry slot allocation failed for channel %u\n",
1616                         EDMA_CHAN_SLOT(echan->ch_num));
1617                 goto err_slot;
1618         }
1619
1620         /* Set up channel -> slot mapping for the entry slot */
1621         edma_set_chmap(echan, echan->slot[0]);
1622         echan->alloced = true;
1623
1624         dev_dbg(dev, "Got eDMA channel %d for virt channel %d (%s trigger)\n",
1625                 EDMA_CHAN_SLOT(echan->ch_num), chan->chan_id,
1626                 echan->hw_triggered ? "HW" : "SW");
1627
1628         edma_tc_set_pm_state(echan->tc, true);
1629
1630         return 0;
1631
1632 err_slot:
1633         edma_free_channel(echan);
1634         return ret;
1635 }
1636
1637 /* Free channel resources */
1638 static void edma_free_chan_resources(struct dma_chan *chan)
1639 {
1640         struct edma_chan *echan = to_edma_chan(chan);
1641         struct device *dev = echan->ecc->dev;
1642         int i;
1643
1644         /* Terminate transfers */
1645         edma_stop(echan);
1646
1647         vchan_free_chan_resources(&echan->vchan);
1648
1649         /* Free EDMA PaRAM slots */
1650         for (i = 0; i < EDMA_MAX_SLOTS; i++) {
1651                 if (echan->slot[i] >= 0) {
1652                         edma_free_slot(echan->ecc, echan->slot[i]);
1653                         echan->slot[i] = -1;
1654                 }
1655         }
1656
1657         /* Set entry slot to the dummy slot */
1658         edma_set_chmap(echan, echan->ecc->dummy_slot);
1659
1660         /* Free EDMA channel */
1661         if (echan->alloced) {
1662                 edma_free_channel(echan);
1663                 echan->alloced = false;
1664         }
1665
1666         edma_tc_set_pm_state(echan->tc, false);
1667         echan->tc = NULL;
1668         echan->hw_triggered = false;
1669
1670         dev_dbg(dev, "Free eDMA channel %d for virt channel %d\n",
1671                 EDMA_CHAN_SLOT(echan->ch_num), chan->chan_id);
1672 }
1673
1674 /* Send pending descriptor to hardware */
1675 static void edma_issue_pending(struct dma_chan *chan)
1676 {
1677         struct edma_chan *echan = to_edma_chan(chan);
1678         unsigned long flags;
1679
1680         spin_lock_irqsave(&echan->vchan.lock, flags);
1681         if (vchan_issue_pending(&echan->vchan) && !echan->edesc)
1682                 edma_execute(echan);
1683         spin_unlock_irqrestore(&echan->vchan.lock, flags);
1684 }
1685
1686 /*
1687  * This limit exists to avoid a possible infinite loop when waiting for proof
1688  * that a particular transfer is completed. This limit can be hit if there
1689  * are large bursts to/from slow devices or the CPU is never able to catch
1690  * the DMA hardware idle. On an AM335x transfering 48 bytes from the UART
1691  * RX-FIFO, as many as 55 loops have been seen.
1692  */
1693 #define EDMA_MAX_TR_WAIT_LOOPS 1000
1694
1695 static u32 edma_residue(struct edma_desc *edesc)
1696 {
1697         bool dst = edesc->direction == DMA_DEV_TO_MEM;
1698         int loop_count = EDMA_MAX_TR_WAIT_LOOPS;
1699         struct edma_chan *echan = edesc->echan;
1700         struct edma_pset *pset = edesc->pset;
1701         dma_addr_t done, pos;
1702         int i;
1703
1704         /*
1705          * We always read the dst/src position from the first RamPar
1706          * pset. That's the one which is active now.
1707          */
1708         pos = edma_get_position(echan->ecc, echan->slot[0], dst);
1709
1710         /*
1711          * "pos" may represent a transfer request that is still being
1712          * processed by the EDMACC or EDMATC. We will busy wait until
1713          * any one of the situations occurs:
1714          *   1. the DMA hardware is idle
1715          *   2. a new transfer request is setup
1716          *   3. we hit the loop limit
1717          */
1718         while (edma_read(echan->ecc, EDMA_CCSTAT) & EDMA_CCSTAT_ACTV) {
1719                 /* check if a new transfer request is setup */
1720                 if (edma_get_position(echan->ecc,
1721                                       echan->slot[0], dst) != pos) {
1722                         break;
1723                 }
1724
1725                 if (!--loop_count) {
1726                         dev_dbg_ratelimited(echan->vchan.chan.device->dev,
1727                                 "%s: timeout waiting for PaRAM update\n",
1728                                 __func__);
1729                         break;
1730                 }
1731
1732                 cpu_relax();
1733         }
1734
1735         /*
1736          * Cyclic is simple. Just subtract pset[0].addr from pos.
1737          *
1738          * We never update edesc->residue in the cyclic case, so we
1739          * can tell the remaining room to the end of the circular
1740          * buffer.
1741          */
1742         if (edesc->cyclic) {
1743                 done = pos - pset->addr;
1744                 edesc->residue_stat = edesc->residue - done;
1745                 return edesc->residue_stat;
1746         }
1747
1748         /*
1749          * For SG operation we catch up with the last processed
1750          * status.
1751          */
1752         pset += edesc->processed_stat;
1753
1754         for (i = edesc->processed_stat; i < edesc->processed; i++, pset++) {
1755                 /*
1756                  * If we are inside this pset address range, we know
1757                  * this is the active one. Get the current delta and
1758                  * stop walking the psets.
1759                  */
1760                 if (pos >= pset->addr && pos < pset->addr + pset->len)
1761                         return edesc->residue_stat - (pos - pset->addr);
1762
1763                 /* Otherwise mark it done and update residue_stat. */
1764                 edesc->processed_stat++;
1765                 edesc->residue_stat -= pset->len;
1766         }
1767         return edesc->residue_stat;
1768 }
1769
1770 /* Check request completion status */
1771 static enum dma_status edma_tx_status(struct dma_chan *chan,
1772                                       dma_cookie_t cookie,
1773                                       struct dma_tx_state *txstate)
1774 {
1775         struct edma_chan *echan = to_edma_chan(chan);
1776         struct virt_dma_desc *vdesc;
1777         enum dma_status ret;
1778         unsigned long flags;
1779
1780         ret = dma_cookie_status(chan, cookie, txstate);
1781         if (ret == DMA_COMPLETE || !txstate)
1782                 return ret;
1783
1784         spin_lock_irqsave(&echan->vchan.lock, flags);
1785         if (echan->edesc && echan->edesc->vdesc.tx.cookie == cookie)
1786                 txstate->residue = edma_residue(echan->edesc);
1787         else if ((vdesc = vchan_find_desc(&echan->vchan, cookie)))
1788                 txstate->residue = to_edma_desc(&vdesc->tx)->residue;
1789         spin_unlock_irqrestore(&echan->vchan.lock, flags);
1790
1791         return ret;
1792 }
1793
1794 static bool edma_is_memcpy_channel(int ch_num, s32 *memcpy_channels)
1795 {
1796         if (!memcpy_channels)
1797                 return false;
1798         while (*memcpy_channels != -1) {
1799                 if (*memcpy_channels == ch_num)
1800                         return true;
1801                 memcpy_channels++;
1802         }
1803         return false;
1804 }
1805
1806 #define EDMA_DMA_BUSWIDTHS      (BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
1807                                  BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
1808                                  BIT(DMA_SLAVE_BUSWIDTH_3_BYTES) | \
1809                                  BIT(DMA_SLAVE_BUSWIDTH_4_BYTES))
1810
1811 static void edma_dma_init(struct edma_cc *ecc, bool legacy_mode)
1812 {
1813         struct dma_device *s_ddev = &ecc->dma_slave;
1814         struct dma_device *m_ddev = NULL;
1815         s32 *memcpy_channels = ecc->info->memcpy_channels;
1816         int i, j;
1817
1818         dma_cap_zero(s_ddev->cap_mask);
1819         dma_cap_set(DMA_SLAVE, s_ddev->cap_mask);
1820         dma_cap_set(DMA_CYCLIC, s_ddev->cap_mask);
1821         if (ecc->legacy_mode && !memcpy_channels) {
1822                 dev_warn(ecc->dev,
1823                          "Legacy memcpy is enabled, things might not work\n");
1824
1825                 dma_cap_set(DMA_MEMCPY, s_ddev->cap_mask);
1826                 s_ddev->device_prep_dma_memcpy = edma_prep_dma_memcpy;
1827                 s_ddev->directions = BIT(DMA_MEM_TO_MEM);
1828         }
1829
1830         s_ddev->device_prep_slave_sg = edma_prep_slave_sg;
1831         s_ddev->device_prep_dma_cyclic = edma_prep_dma_cyclic;
1832         s_ddev->device_alloc_chan_resources = edma_alloc_chan_resources;
1833         s_ddev->device_free_chan_resources = edma_free_chan_resources;
1834         s_ddev->device_issue_pending = edma_issue_pending;
1835         s_ddev->device_tx_status = edma_tx_status;
1836         s_ddev->device_config = edma_slave_config;
1837         s_ddev->device_pause = edma_dma_pause;
1838         s_ddev->device_resume = edma_dma_resume;
1839         s_ddev->device_terminate_all = edma_terminate_all;
1840
1841         s_ddev->src_addr_widths = EDMA_DMA_BUSWIDTHS;
1842         s_ddev->dst_addr_widths = EDMA_DMA_BUSWIDTHS;
1843         s_ddev->directions |= (BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV));
1844         s_ddev->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
1845
1846         s_ddev->dev = ecc->dev;
1847         INIT_LIST_HEAD(&s_ddev->channels);
1848
1849         if (memcpy_channels) {
1850                 m_ddev = devm_kzalloc(ecc->dev, sizeof(*m_ddev), GFP_KERNEL);
1851                 ecc->dma_memcpy = m_ddev;
1852
1853                 dma_cap_zero(m_ddev->cap_mask);
1854                 dma_cap_set(DMA_MEMCPY, m_ddev->cap_mask);
1855
1856                 m_ddev->device_prep_dma_memcpy = edma_prep_dma_memcpy;
1857                 m_ddev->device_alloc_chan_resources = edma_alloc_chan_resources;
1858                 m_ddev->device_free_chan_resources = edma_free_chan_resources;
1859                 m_ddev->device_issue_pending = edma_issue_pending;
1860                 m_ddev->device_tx_status = edma_tx_status;
1861                 m_ddev->device_config = edma_slave_config;
1862                 m_ddev->device_pause = edma_dma_pause;
1863                 m_ddev->device_resume = edma_dma_resume;
1864                 m_ddev->device_terminate_all = edma_terminate_all;
1865
1866                 m_ddev->src_addr_widths = EDMA_DMA_BUSWIDTHS;
1867                 m_ddev->dst_addr_widths = EDMA_DMA_BUSWIDTHS;
1868                 m_ddev->directions = BIT(DMA_MEM_TO_MEM);
1869                 m_ddev->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
1870
1871                 m_ddev->dev = ecc->dev;
1872                 INIT_LIST_HEAD(&m_ddev->channels);
1873         } else if (!ecc->legacy_mode) {
1874                 dev_info(ecc->dev, "memcpy is disabled\n");
1875         }
1876
1877         for (i = 0; i < ecc->num_channels; i++) {
1878                 struct edma_chan *echan = &ecc->slave_chans[i];
1879                 echan->ch_num = EDMA_CTLR_CHAN(ecc->id, i);
1880                 echan->ecc = ecc;
1881                 echan->vchan.desc_free = edma_desc_free;
1882
1883                 if (m_ddev && edma_is_memcpy_channel(i, memcpy_channels))
1884                         vchan_init(&echan->vchan, m_ddev);
1885                 else
1886                         vchan_init(&echan->vchan, s_ddev);
1887
1888                 INIT_LIST_HEAD(&echan->node);
1889                 for (j = 0; j < EDMA_MAX_SLOTS; j++)
1890                         echan->slot[j] = -1;
1891         }
1892 }
1893
1894 static int edma_setup_from_hw(struct device *dev, struct edma_soc_info *pdata,
1895                               struct edma_cc *ecc)
1896 {
1897         int i;
1898         u32 value, cccfg;
1899         s8 (*queue_priority_map)[2];
1900
1901         /* Decode the eDMA3 configuration from CCCFG register */
1902         cccfg = edma_read(ecc, EDMA_CCCFG);
1903
1904         value = GET_NUM_REGN(cccfg);
1905         ecc->num_region = BIT(value);
1906
1907         value = GET_NUM_DMACH(cccfg);
1908         ecc->num_channels = BIT(value + 1);
1909
1910         value = GET_NUM_QDMACH(cccfg);
1911         ecc->num_qchannels = value * 2;
1912
1913         value = GET_NUM_PAENTRY(cccfg);
1914         ecc->num_slots = BIT(value + 4);
1915
1916         value = GET_NUM_EVQUE(cccfg);
1917         ecc->num_tc = value + 1;
1918
1919         ecc->chmap_exist = (cccfg & CHMAP_EXIST) ? true : false;
1920
1921         dev_dbg(dev, "eDMA3 CC HW configuration (cccfg: 0x%08x):\n", cccfg);
1922         dev_dbg(dev, "num_region: %u\n", ecc->num_region);
1923         dev_dbg(dev, "num_channels: %u\n", ecc->num_channels);
1924         dev_dbg(dev, "num_qchannels: %u\n", ecc->num_qchannels);
1925         dev_dbg(dev, "num_slots: %u\n", ecc->num_slots);
1926         dev_dbg(dev, "num_tc: %u\n", ecc->num_tc);
1927         dev_dbg(dev, "chmap_exist: %s\n", ecc->chmap_exist ? "yes" : "no");
1928
1929         /* Nothing need to be done if queue priority is provided */
1930         if (pdata->queue_priority_mapping)
1931                 return 0;
1932
1933         /*
1934          * Configure TC/queue priority as follows:
1935          * Q0 - priority 0
1936          * Q1 - priority 1
1937          * Q2 - priority 2
1938          * ...
1939          * The meaning of priority numbers: 0 highest priority, 7 lowest
1940          * priority. So Q0 is the highest priority queue and the last queue has
1941          * the lowest priority.
1942          */
1943         queue_priority_map = devm_kcalloc(dev, ecc->num_tc + 1, sizeof(s8),
1944                                           GFP_KERNEL);
1945         if (!queue_priority_map)
1946                 return -ENOMEM;
1947
1948         for (i = 0; i < ecc->num_tc; i++) {
1949                 queue_priority_map[i][0] = i;
1950                 queue_priority_map[i][1] = i;
1951         }
1952         queue_priority_map[i][0] = -1;
1953         queue_priority_map[i][1] = -1;
1954
1955         pdata->queue_priority_mapping = queue_priority_map;
1956         /* Default queue has the lowest priority */
1957         pdata->default_queue = i - 1;
1958
1959         return 0;
1960 }
1961
1962 #if IS_ENABLED(CONFIG_OF)
1963 static int edma_xbar_event_map(struct device *dev, struct edma_soc_info *pdata,
1964                                size_t sz)
1965 {
1966         const char pname[] = "ti,edma-xbar-event-map";
1967         struct resource res;
1968         void __iomem *xbar;
1969         s16 (*xbar_chans)[2];
1970         size_t nelm = sz / sizeof(s16);
1971         u32 shift, offset, mux;
1972         int ret, i;
1973
1974         xbar_chans = devm_kcalloc(dev, nelm + 2, sizeof(s16), GFP_KERNEL);
1975         if (!xbar_chans)
1976                 return -ENOMEM;
1977
1978         ret = of_address_to_resource(dev->of_node, 1, &res);
1979         if (ret)
1980                 return -ENOMEM;
1981
1982         xbar = devm_ioremap(dev, res.start, resource_size(&res));
1983         if (!xbar)
1984                 return -ENOMEM;
1985
1986         ret = of_property_read_u16_array(dev->of_node, pname, (u16 *)xbar_chans,
1987                                          nelm);
1988         if (ret)
1989                 return -EIO;
1990
1991         /* Invalidate last entry for the other user of this mess */
1992         nelm >>= 1;
1993         xbar_chans[nelm][0] = -1;
1994         xbar_chans[nelm][1] = -1;
1995
1996         for (i = 0; i < nelm; i++) {
1997                 shift = (xbar_chans[i][1] & 0x03) << 3;
1998                 offset = xbar_chans[i][1] & 0xfffffffc;
1999                 mux = readl(xbar + offset);
2000                 mux &= ~(0xff << shift);
2001                 mux |= xbar_chans[i][0] << shift;
2002                 writel(mux, (xbar + offset));
2003         }
2004
2005         pdata->xbar_chans = (const s16 (*)[2]) xbar_chans;
2006         return 0;
2007 }
2008
2009 static struct edma_soc_info *edma_setup_info_from_dt(struct device *dev,
2010                                                      bool legacy_mode)
2011 {
2012         struct edma_soc_info *info;
2013         struct property *prop;
2014         size_t sz;
2015         int ret;
2016
2017         info = devm_kzalloc(dev, sizeof(struct edma_soc_info), GFP_KERNEL);
2018         if (!info)
2019                 return ERR_PTR(-ENOMEM);
2020
2021         if (legacy_mode) {
2022                 prop = of_find_property(dev->of_node, "ti,edma-xbar-event-map",
2023                                         &sz);
2024                 if (prop) {
2025                         ret = edma_xbar_event_map(dev, info, sz);
2026                         if (ret)
2027                                 return ERR_PTR(ret);
2028                 }
2029                 return info;
2030         }
2031
2032         /* Get the list of channels allocated to be used for memcpy */
2033         prop = of_find_property(dev->of_node, "ti,edma-memcpy-channels", &sz);
2034         if (prop) {
2035                 const char pname[] = "ti,edma-memcpy-channels";
2036                 size_t nelm = sz / sizeof(s32);
2037                 s32 *memcpy_ch;
2038
2039                 memcpy_ch = devm_kcalloc(dev, nelm + 1, sizeof(s32),
2040                                          GFP_KERNEL);
2041                 if (!memcpy_ch)
2042                         return ERR_PTR(-ENOMEM);
2043
2044                 ret = of_property_read_u32_array(dev->of_node, pname,
2045                                                  (u32 *)memcpy_ch, nelm);
2046                 if (ret)
2047                         return ERR_PTR(ret);
2048
2049                 memcpy_ch[nelm] = -1;
2050                 info->memcpy_channels = memcpy_ch;
2051         }
2052
2053         prop = of_find_property(dev->of_node, "ti,edma-reserved-slot-ranges",
2054                                 &sz);
2055         if (prop) {
2056                 const char pname[] = "ti,edma-reserved-slot-ranges";
2057                 u32 (*tmp)[2];
2058                 s16 (*rsv_slots)[2];
2059                 size_t nelm = sz / sizeof(*tmp);
2060                 struct edma_rsv_info *rsv_info;
2061                 int i;
2062
2063                 if (!nelm)
2064                         return info;
2065
2066                 tmp = kcalloc(nelm, sizeof(*tmp), GFP_KERNEL);
2067                 if (!tmp)
2068                         return ERR_PTR(-ENOMEM);
2069
2070                 rsv_info = devm_kzalloc(dev, sizeof(*rsv_info), GFP_KERNEL);
2071                 if (!rsv_info) {
2072                         kfree(tmp);
2073                         return ERR_PTR(-ENOMEM);
2074                 }
2075
2076                 rsv_slots = devm_kcalloc(dev, nelm + 1, sizeof(*rsv_slots),
2077                                          GFP_KERNEL);
2078                 if (!rsv_slots) {
2079                         kfree(tmp);
2080                         return ERR_PTR(-ENOMEM);
2081                 }
2082
2083                 ret = of_property_read_u32_array(dev->of_node, pname,
2084                                                  (u32 *)tmp, nelm * 2);
2085                 if (ret) {
2086                         kfree(tmp);
2087                         return ERR_PTR(ret);
2088                 }
2089
2090                 for (i = 0; i < nelm; i++) {
2091                         rsv_slots[i][0] = tmp[i][0];
2092                         rsv_slots[i][1] = tmp[i][1];
2093                 }
2094                 rsv_slots[nelm][0] = -1;
2095                 rsv_slots[nelm][1] = -1;
2096
2097                 info->rsv = rsv_info;
2098                 info->rsv->rsv_slots = (const s16 (*)[2])rsv_slots;
2099
2100                 kfree(tmp);
2101         }
2102
2103         return info;
2104 }
2105
2106 static struct dma_chan *of_edma_xlate(struct of_phandle_args *dma_spec,
2107                                       struct of_dma *ofdma)
2108 {
2109         struct edma_cc *ecc = ofdma->of_dma_data;
2110         struct dma_chan *chan = NULL;
2111         struct edma_chan *echan;
2112         int i;
2113
2114         if (!ecc || dma_spec->args_count < 1)
2115                 return NULL;
2116
2117         for (i = 0; i < ecc->num_channels; i++) {
2118                 echan = &ecc->slave_chans[i];
2119                 if (echan->ch_num == dma_spec->args[0]) {
2120                         chan = &echan->vchan.chan;
2121                         break;
2122                 }
2123         }
2124
2125         if (!chan)
2126                 return NULL;
2127
2128         if (echan->ecc->legacy_mode && dma_spec->args_count == 1)
2129                 goto out;
2130
2131         if (!echan->ecc->legacy_mode && dma_spec->args_count == 2 &&
2132             dma_spec->args[1] < echan->ecc->num_tc) {
2133                 echan->tc = &echan->ecc->tc_list[dma_spec->args[1]];
2134                 goto out;
2135         }
2136
2137         return NULL;
2138 out:
2139         /* The channel is going to be used as HW synchronized */
2140         echan->hw_triggered = true;
2141         return dma_get_slave_channel(chan);
2142 }
2143 #else
2144 static struct edma_soc_info *edma_setup_info_from_dt(struct device *dev,
2145                                                      bool legacy_mode)
2146 {
2147         return ERR_PTR(-EINVAL);
2148 }
2149
2150 static struct dma_chan *of_edma_xlate(struct of_phandle_args *dma_spec,
2151                                       struct of_dma *ofdma)
2152 {
2153         return NULL;
2154 }
2155 #endif
2156
2157 static int edma_probe(struct platform_device *pdev)
2158 {
2159         struct edma_soc_info    *info = pdev->dev.platform_data;
2160         s8                      (*queue_priority_mapping)[2];
2161         int                     i, off, ln;
2162         const s16               (*rsv_slots)[2];
2163         const s16               (*xbar_chans)[2];
2164         int                     irq;
2165         char                    *irq_name;
2166         struct resource         *mem;
2167         struct device_node      *node = pdev->dev.of_node;
2168         struct device           *dev = &pdev->dev;
2169         struct edma_cc          *ecc;
2170         bool                    legacy_mode = true;
2171         int ret;
2172
2173         if (node) {
2174                 const struct of_device_id *match;
2175
2176                 match = of_match_node(edma_of_ids, node);
2177                 if (match && (u32)match->data == EDMA_BINDING_TPCC)
2178                         legacy_mode = false;
2179
2180                 info = edma_setup_info_from_dt(dev, legacy_mode);
2181                 if (IS_ERR(info)) {
2182                         dev_err(dev, "failed to get DT data\n");
2183                         return PTR_ERR(info);
2184                 }
2185         }
2186
2187         if (!info)
2188                 return -ENODEV;
2189
2190         pm_runtime_enable(dev);
2191         ret = pm_runtime_get_sync(dev);
2192         if (ret < 0) {
2193                 dev_err(dev, "pm_runtime_get_sync() failed\n");
2194                 return ret;
2195         }
2196
2197         ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
2198         if (ret)
2199                 return ret;
2200
2201         ecc = devm_kzalloc(dev, sizeof(*ecc), GFP_KERNEL);
2202         if (!ecc) {
2203                 dev_err(dev, "Can't allocate controller\n");
2204                 return -ENOMEM;
2205         }
2206
2207         ecc->dev = dev;
2208         ecc->id = pdev->id;
2209         ecc->legacy_mode = legacy_mode;
2210         /* When booting with DT the pdev->id is -1 */
2211         if (ecc->id < 0)
2212                 ecc->id = 0;
2213
2214         mem = platform_get_resource_byname(pdev, IORESOURCE_MEM, "edma3_cc");
2215         if (!mem) {
2216                 dev_dbg(dev, "mem resource not found, using index 0\n");
2217                 mem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
2218                 if (!mem) {
2219                         dev_err(dev, "no mem resource?\n");
2220                         return -ENODEV;
2221                 }
2222         }
2223         ecc->base = devm_ioremap_resource(dev, mem);
2224         if (IS_ERR(ecc->base))
2225                 return PTR_ERR(ecc->base);
2226
2227         platform_set_drvdata(pdev, ecc);
2228
2229         /* Get eDMA3 configuration from IP */
2230         ret = edma_setup_from_hw(dev, info, ecc);
2231         if (ret)
2232                 return ret;
2233
2234         /* Allocate memory based on the information we got from the IP */
2235         ecc->slave_chans = devm_kcalloc(dev, ecc->num_channels,
2236                                         sizeof(*ecc->slave_chans), GFP_KERNEL);
2237         if (!ecc->slave_chans)
2238                 return -ENOMEM;
2239
2240         ecc->slot_inuse = devm_kcalloc(dev, BITS_TO_LONGS(ecc->num_slots),
2241                                        sizeof(unsigned long), GFP_KERNEL);
2242         if (!ecc->slot_inuse)
2243                 return -ENOMEM;
2244
2245         ecc->default_queue = info->default_queue;
2246
2247         for (i = 0; i < ecc->num_slots; i++)
2248                 edma_write_slot(ecc, i, &dummy_paramset);
2249
2250         if (info->rsv) {
2251                 /* Set the reserved slots in inuse list */
2252                 rsv_slots = info->rsv->rsv_slots;
2253                 if (rsv_slots) {
2254                         for (i = 0; rsv_slots[i][0] != -1; i++) {
2255                                 off = rsv_slots[i][0];
2256                                 ln = rsv_slots[i][1];
2257                                 set_bits(off, ln, ecc->slot_inuse);
2258                         }
2259                 }
2260         }
2261
2262         /* Clear the xbar mapped channels in unused list */
2263         xbar_chans = info->xbar_chans;
2264         if (xbar_chans) {
2265                 for (i = 0; xbar_chans[i][1] != -1; i++) {
2266                         off = xbar_chans[i][1];
2267                 }
2268         }
2269
2270         irq = platform_get_irq_byname(pdev, "edma3_ccint");
2271         if (irq < 0 && node)
2272                 irq = irq_of_parse_and_map(node, 0);
2273
2274         if (irq >= 0) {
2275                 irq_name = devm_kasprintf(dev, GFP_KERNEL, "%s_ccint",
2276                                           dev_name(dev));
2277                 ret = devm_request_irq(dev, irq, dma_irq_handler, 0, irq_name,
2278                                        ecc);
2279                 if (ret) {
2280                         dev_err(dev, "CCINT (%d) failed --> %d\n", irq, ret);
2281                         return ret;
2282                 }
2283         }
2284
2285         irq = platform_get_irq_byname(pdev, "edma3_ccerrint");
2286         if (irq < 0 && node)
2287                 irq = irq_of_parse_and_map(node, 2);
2288
2289         if (irq >= 0) {
2290                 irq_name = devm_kasprintf(dev, GFP_KERNEL, "%s_ccerrint",
2291                                           dev_name(dev));
2292                 ret = devm_request_irq(dev, irq, dma_ccerr_handler, 0, irq_name,
2293                                        ecc);
2294                 if (ret) {
2295                         dev_err(dev, "CCERRINT (%d) failed --> %d\n", irq, ret);
2296                         return ret;
2297                 }
2298         }
2299
2300         ecc->dummy_slot = edma_alloc_slot(ecc, EDMA_SLOT_ANY);
2301         if (ecc->dummy_slot < 0) {
2302                 dev_err(dev, "Can't allocate PaRAM dummy slot\n");
2303                 return ecc->dummy_slot;
2304         }
2305
2306         queue_priority_mapping = info->queue_priority_mapping;
2307
2308         if (!ecc->legacy_mode) {
2309                 int lowest_priority = 0;
2310                 struct of_phandle_args tc_args;
2311
2312                 ecc->tc_list = devm_kcalloc(dev, ecc->num_tc,
2313                                             sizeof(*ecc->tc_list), GFP_KERNEL);
2314                 if (!ecc->tc_list)
2315                         return -ENOMEM;
2316
2317                 for (i = 0;; i++) {
2318                         ret = of_parse_phandle_with_fixed_args(node, "ti,tptcs",
2319                                                                1, i, &tc_args);
2320                         if (ret || i == ecc->num_tc)
2321                                 break;
2322
2323                         ecc->tc_list[i].node = tc_args.np;
2324                         ecc->tc_list[i].id = i;
2325                         queue_priority_mapping[i][1] = tc_args.args[0];
2326                         if (queue_priority_mapping[i][1] > lowest_priority) {
2327                                 lowest_priority = queue_priority_mapping[i][1];
2328                                 info->default_queue = i;
2329                         }
2330                 }
2331         }
2332
2333         /* Event queue priority mapping */
2334         for (i = 0; queue_priority_mapping[i][0] != -1; i++)
2335                 edma_assign_priority_to_queue(ecc, queue_priority_mapping[i][0],
2336                                               queue_priority_mapping[i][1]);
2337
2338         for (i = 0; i < ecc->num_region; i++) {
2339                 edma_write_array2(ecc, EDMA_DRAE, i, 0, 0x0);
2340                 edma_write_array2(ecc, EDMA_DRAE, i, 1, 0x0);
2341                 edma_write_array(ecc, EDMA_QRAE, i, 0x0);
2342         }
2343         ecc->info = info;
2344
2345         /* Init the dma device and channels */
2346         edma_dma_init(ecc, legacy_mode);
2347
2348         for (i = 0; i < ecc->num_channels; i++) {
2349                 /* Assign all channels to the default queue */
2350                 edma_assign_channel_eventq(&ecc->slave_chans[i],
2351                                            info->default_queue);
2352                 /* Set entry slot to the dummy slot */
2353                 edma_set_chmap(&ecc->slave_chans[i], ecc->dummy_slot);
2354         }
2355
2356         ecc->dma_slave.filter.map = info->slave_map;
2357         ecc->dma_slave.filter.mapcnt = info->slavecnt;
2358         ecc->dma_slave.filter.fn = edma_filter_fn;
2359
2360         ret = dma_async_device_register(&ecc->dma_slave);
2361         if (ret) {
2362                 dev_err(dev, "slave ddev registration failed (%d)\n", ret);
2363                 goto err_reg1;
2364         }
2365
2366         if (ecc->dma_memcpy) {
2367                 ret = dma_async_device_register(ecc->dma_memcpy);
2368                 if (ret) {
2369                         dev_err(dev, "memcpy ddev registration failed (%d)\n",
2370                                 ret);
2371                         dma_async_device_unregister(&ecc->dma_slave);
2372                         goto err_reg1;
2373                 }
2374         }
2375
2376         if (node)
2377                 of_dma_controller_register(node, of_edma_xlate, ecc);
2378
2379         dev_info(dev, "TI EDMA DMA engine driver\n");
2380
2381         return 0;
2382
2383 err_reg1:
2384         edma_free_slot(ecc, ecc->dummy_slot);
2385         return ret;
2386 }
2387
2388 static int edma_remove(struct platform_device *pdev)
2389 {
2390         struct device *dev = &pdev->dev;
2391         struct edma_cc *ecc = dev_get_drvdata(dev);
2392
2393         if (dev->of_node)
2394                 of_dma_controller_free(dev->of_node);
2395         dma_async_device_unregister(&ecc->dma_slave);
2396         if (ecc->dma_memcpy)
2397                 dma_async_device_unregister(ecc->dma_memcpy);
2398         edma_free_slot(ecc, ecc->dummy_slot);
2399
2400         return 0;
2401 }
2402
2403 #ifdef CONFIG_PM_SLEEP
2404 static int edma_pm_suspend(struct device *dev)
2405 {
2406         struct edma_cc *ecc = dev_get_drvdata(dev);
2407         struct edma_chan *echan = ecc->slave_chans;
2408         int i;
2409
2410         for (i = 0; i < ecc->num_channels; i++) {
2411                 if (echan[i].alloced) {
2412                         edma_setup_interrupt(&echan[i], false);
2413                         edma_tc_set_pm_state(echan[i].tc, false);
2414                 }
2415         }
2416
2417         return 0;
2418 }
2419
2420 static int edma_pm_resume(struct device *dev)
2421 {
2422         struct edma_cc *ecc = dev_get_drvdata(dev);
2423         struct edma_chan *echan = ecc->slave_chans;
2424         int i;
2425         s8 (*queue_priority_mapping)[2];
2426
2427         queue_priority_mapping = ecc->info->queue_priority_mapping;
2428
2429         /* Event queue priority mapping */
2430         for (i = 0; queue_priority_mapping[i][0] != -1; i++)
2431                 edma_assign_priority_to_queue(ecc, queue_priority_mapping[i][0],
2432                                               queue_priority_mapping[i][1]);
2433
2434         for (i = 0; i < ecc->num_channels; i++) {
2435                 if (echan[i].alloced) {
2436                         /* ensure access through shadow region 0 */
2437                         edma_or_array2(ecc, EDMA_DRAE, 0, i >> 5,
2438                                        BIT(i & 0x1f));
2439
2440                         edma_setup_interrupt(&echan[i], true);
2441
2442                         /* Set up channel -> slot mapping for the entry slot */
2443                         edma_set_chmap(&echan[i], echan[i].slot[0]);
2444
2445                         edma_tc_set_pm_state(echan[i].tc, true);
2446                 }
2447         }
2448
2449         return 0;
2450 }
2451 #endif
2452
2453 static const struct dev_pm_ops edma_pm_ops = {
2454         SET_LATE_SYSTEM_SLEEP_PM_OPS(edma_pm_suspend, edma_pm_resume)
2455 };
2456
2457 static struct platform_driver edma_driver = {
2458         .probe          = edma_probe,
2459         .remove         = edma_remove,
2460         .driver = {
2461                 .name   = "edma",
2462                 .pm     = &edma_pm_ops,
2463                 .of_match_table = edma_of_ids,
2464         },
2465 };
2466
2467 static int edma_tptc_probe(struct platform_device *pdev)
2468 {
2469         return 0;
2470 }
2471
2472 static struct platform_driver edma_tptc_driver = {
2473         .probe          = edma_tptc_probe,
2474         .driver = {
2475                 .name   = "edma3-tptc",
2476                 .of_match_table = edma_tptc_of_ids,
2477         },
2478 };
2479
2480 bool edma_filter_fn(struct dma_chan *chan, void *param)
2481 {
2482         bool match = false;
2483
2484         if (chan->device->dev->driver == &edma_driver.driver) {
2485                 struct edma_chan *echan = to_edma_chan(chan);
2486                 unsigned ch_req = *(unsigned *)param;
2487                 if (ch_req == echan->ch_num) {
2488                         /* The channel is going to be used as HW synchronized */
2489                         echan->hw_triggered = true;
2490                         match = true;
2491                 }
2492         }
2493         return match;
2494 }
2495 EXPORT_SYMBOL(edma_filter_fn);
2496
2497 static int edma_init(void)
2498 {
2499         int ret;
2500
2501         ret = platform_driver_register(&edma_tptc_driver);
2502         if (ret)
2503                 return ret;
2504
2505         return platform_driver_register(&edma_driver);
2506 }
2507 subsys_initcall(edma_init);
2508
2509 static void __exit edma_exit(void)
2510 {
2511         platform_driver_unregister(&edma_driver);
2512         platform_driver_unregister(&edma_tptc_driver);
2513 }
2514 module_exit(edma_exit);
2515
2516 MODULE_AUTHOR("Matt Porter <matt.porter@linaro.org>");
2517 MODULE_DESCRIPTION("TI EDMA DMA engine driver");
2518 MODULE_LICENSE("GPL v2");