?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
sh_eth: fix EESIPR values for SH77{34|63}
[linux/fpc-iii.git] / drivers / dma / edma.c
blob3879f80a4815cb27ba4329e29510752e34331ba4
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 unsigned int ccint;
243 unsigned int ccerrint;
244
245 /*
246 * The slot_inuse bit for each PaRAM slot is clear unless the slot is
247 * in use by Linux or if it is allocated to be used by DSP.
248 */
249 unsigned long *slot_inuse;
250
251 struct dma_device dma_slave;
252 struct dma_device *dma_memcpy;
253 struct edma_chan *slave_chans;
254 struct edma_tc *tc_list;
255 int dummy_slot;
256 };
257
258 /* dummy param set used to (re)initialize parameter RAM slots */
259 static const struct edmacc_param dummy_paramset = {
260 .link_bcntrld = 0xffff,
261 .ccnt = 1,
262 };
263
264 #define EDMA_BINDING_LEGACY 0
265 #define EDMA_BINDING_TPCC 1
266 static const u32 edma_binding_type[] = {
267 [EDMA_BINDING_LEGACY] = EDMA_BINDING_LEGACY,
268 [EDMA_BINDING_TPCC] = EDMA_BINDING_TPCC,
269 };
270
271 static const struct of_device_id edma_of_ids[] = {
272 {
273 .compatible = "ti,edma3",
274 .data = &edma_binding_type[EDMA_BINDING_LEGACY],
275 },
276 {
277 .compatible = "ti,edma3-tpcc",
278 .data = &edma_binding_type[EDMA_BINDING_TPCC],
279 },
280 {}
281 };
282 MODULE_DEVICE_TABLE(of, edma_of_ids);
283
284 static const struct of_device_id edma_tptc_of_ids[] = {
285 { .compatible = "ti,edma3-tptc", },
286 {}
287 };
288 MODULE_DEVICE_TABLE(of, edma_tptc_of_ids);
289
290 static inline unsigned int edma_read(struct edma_cc *ecc, int offset)
291 {
292 return (unsigned int)__raw_readl(ecc->base + offset);
293 }
294
295 static inline void edma_write(struct edma_cc *ecc, int offset, int val)
296 {
297 __raw_writel(val, ecc->base + offset);
298 }
299
300 static inline void edma_modify(struct edma_cc *ecc, int offset, unsigned and,
301 unsigned or)
302 {
303 unsigned val = edma_read(ecc, offset);
304
305 val &= and;
306 val |= or;
307 edma_write(ecc, offset, val);
308 }
309
310 static inline void edma_and(struct edma_cc *ecc, int offset, unsigned and)
311 {
312 unsigned val = edma_read(ecc, offset);
313
314 val &= and;
315 edma_write(ecc, offset, val);
316 }
317
318 static inline void edma_or(struct edma_cc *ecc, int offset, unsigned or)
319 {
320 unsigned val = edma_read(ecc, offset);
321
322 val |= or;
323 edma_write(ecc, offset, val);
324 }
325
326 static inline unsigned int edma_read_array(struct edma_cc *ecc, int offset,
327 int i)
328 {
329 return edma_read(ecc, offset + (i << 2));
330 }
331
332 static inline void edma_write_array(struct edma_cc *ecc, int offset, int i,
333 unsigned val)
334 {
335 edma_write(ecc, offset + (i << 2), val);
336 }
337
338 static inline void edma_modify_array(struct edma_cc *ecc, int offset, int i,
339 unsigned and, unsigned or)
340 {
341 edma_modify(ecc, offset + (i << 2), and, or);
342 }
343
344 static inline void edma_or_array(struct edma_cc *ecc, int offset, int i,
345 unsigned or)
346 {
347 edma_or(ecc, offset + (i << 2), or);
348 }
349
350 static inline void edma_or_array2(struct edma_cc *ecc, int offset, int i, int j,
351 unsigned or)
352 {
353 edma_or(ecc, offset + ((i * 2 + j) << 2), or);
354 }
355
356 static inline void edma_write_array2(struct edma_cc *ecc, int offset, int i,
357 int j, unsigned val)
358 {
359 edma_write(ecc, offset + ((i * 2 + j) << 2), val);
360 }
361
362 static inline unsigned int edma_shadow0_read(struct edma_cc *ecc, int offset)
363 {
364 return edma_read(ecc, EDMA_SHADOW0 + offset);
365 }
366
367 static inline unsigned int edma_shadow0_read_array(struct edma_cc *ecc,
368 int offset, int i)
369 {
370 return edma_read(ecc, EDMA_SHADOW0 + offset + (i << 2));
371 }
372
373 static inline void edma_shadow0_write(struct edma_cc *ecc, int offset,
374 unsigned val)
375 {
376 edma_write(ecc, EDMA_SHADOW0 + offset, val);
377 }
378
379 static inline void edma_shadow0_write_array(struct edma_cc *ecc, int offset,
380 int i, unsigned val)
381 {
382 edma_write(ecc, EDMA_SHADOW0 + offset + (i << 2), val);
383 }
384
385 static inline unsigned int edma_param_read(struct edma_cc *ecc, int offset,
386 int param_no)
387 {
388 return edma_read(ecc, EDMA_PARM + offset + (param_no << 5));
389 }
390
391 static inline void edma_param_write(struct edma_cc *ecc, int offset,
392 int param_no, unsigned val)
393 {
394 edma_write(ecc, EDMA_PARM + offset + (param_no << 5), val);
395 }
396
397 static inline void edma_param_modify(struct edma_cc *ecc, int offset,
398 int param_no, unsigned and, unsigned or)
399 {
400 edma_modify(ecc, EDMA_PARM + offset + (param_no << 5), and, or);
401 }
402
403 static inline void edma_param_and(struct edma_cc *ecc, int offset, int param_no,
404 unsigned and)
405 {
406 edma_and(ecc, EDMA_PARM + offset + (param_no << 5), and);
407 }
408
409 static inline void edma_param_or(struct edma_cc *ecc, int offset, int param_no,
410 unsigned or)
411 {
412 edma_or(ecc, EDMA_PARM + offset + (param_no << 5), or);
413 }
414
415 static inline void edma_set_bits(int offset, int len, unsigned long *p)
416 {
417 for (; len > 0; len--)
418 set_bit(offset + (len - 1), p);
419 }
420
421 static void edma_assign_priority_to_queue(struct edma_cc *ecc, int queue_no,
422 int priority)
423 {
424 int bit = queue_no * 4;
425
426 edma_modify(ecc, EDMA_QUEPRI, ~(0x7 << bit), ((priority & 0x7) << bit));
427 }
428
429 static void edma_set_chmap(struct edma_chan *echan, int slot)
430 {
431 struct edma_cc *ecc = echan->ecc;
432 int channel = EDMA_CHAN_SLOT(echan->ch_num);
433
434 if (ecc->chmap_exist) {
435 slot = EDMA_CHAN_SLOT(slot);
436 edma_write_array(ecc, EDMA_DCHMAP, channel, (slot << 5));
437 }
438 }
439
440 static void edma_setup_interrupt(struct edma_chan *echan, bool enable)
441 {
442 struct edma_cc *ecc = echan->ecc;
443 int channel = EDMA_CHAN_SLOT(echan->ch_num);
444
445 if (enable) {
446 edma_shadow0_write_array(ecc, SH_ICR, channel >> 5,
447 BIT(channel & 0x1f));
448 edma_shadow0_write_array(ecc, SH_IESR, channel >> 5,
449 BIT(channel & 0x1f));
450 } else {
451 edma_shadow0_write_array(ecc, SH_IECR, channel >> 5,
452 BIT(channel & 0x1f));
453 }
454 }
455
456 /*
457 * paRAM slot management functions
458 */
459 static void edma_write_slot(struct edma_cc *ecc, unsigned slot,
460 const struct edmacc_param *param)
461 {
462 slot = EDMA_CHAN_SLOT(slot);
463 if (slot >= ecc->num_slots)
464 return;
465 memcpy_toio(ecc->base + PARM_OFFSET(slot), param, PARM_SIZE);
466 }
467
468 static int edma_read_slot(struct edma_cc *ecc, unsigned slot,
469 struct edmacc_param *param)
470 {
471 slot = EDMA_CHAN_SLOT(slot);
472 if (slot >= ecc->num_slots)
473 return -EINVAL;
474 memcpy_fromio(param, ecc->base + PARM_OFFSET(slot), PARM_SIZE);
475
476 return 0;
477 }
478
479 /**
480 * edma_alloc_slot - allocate DMA parameter RAM
481 * @ecc: pointer to edma_cc struct
482 * @slot: specific slot to allocate; negative for "any unused slot"
483 *
484 * This allocates a parameter RAM slot, initializing it to hold a
485 * dummy transfer. Slots allocated using this routine have not been
486 * mapped to a hardware DMA channel, and will normally be used by
487 * linking to them from a slot associated with a DMA channel.
488 *
489 * Normal use is to pass EDMA_SLOT_ANY as the @slot, but specific
490 * slots may be allocated on behalf of DSP firmware.
491 *
492 * Returns the number of the slot, else negative errno.
493 */
494 static int edma_alloc_slot(struct edma_cc *ecc, int slot)
495 {
496 if (slot >= 0) {
497 slot = EDMA_CHAN_SLOT(slot);
498 /* Requesting entry paRAM slot for a HW triggered channel. */
499 if (ecc->chmap_exist && slot < ecc->num_channels)
500 slot = EDMA_SLOT_ANY;
501 }
502
503 if (slot < 0) {
504 if (ecc->chmap_exist)
505 slot = 0;
506 else
507 slot = ecc->num_channels;
508 for (;;) {
509 slot = find_next_zero_bit(ecc->slot_inuse,
510 ecc->num_slots,
511 slot);
512 if (slot == ecc->num_slots)
513 return -ENOMEM;
514 if (!test_and_set_bit(slot, ecc->slot_inuse))
515 break;
516 }
517 } else if (slot >= ecc->num_slots) {
518 return -EINVAL;
519 } else if (test_and_set_bit(slot, ecc->slot_inuse)) {
520 return -EBUSY;
521 }
522
523 edma_write_slot(ecc, slot, &dummy_paramset);
524
525 return EDMA_CTLR_CHAN(ecc->id, slot);
526 }
527
528 static void edma_free_slot(struct edma_cc *ecc, unsigned slot)
529 {
530 slot = EDMA_CHAN_SLOT(slot);
531 if (slot >= ecc->num_slots)
532 return;
533
534 edma_write_slot(ecc, slot, &dummy_paramset);
535 clear_bit(slot, ecc->slot_inuse);
536 }
537
538 /**
539 * edma_link - link one parameter RAM slot to another
540 * @ecc: pointer to edma_cc struct
541 * @from: parameter RAM slot originating the link
542 * @to: parameter RAM slot which is the link target
543 *
544 * The originating slot should not be part of any active DMA transfer.
545 */
546 static void edma_link(struct edma_cc *ecc, unsigned from, unsigned to)
547 {
548 if (unlikely(EDMA_CTLR(from) != EDMA_CTLR(to)))
549 dev_warn(ecc->dev, "Ignoring eDMA instance for linking\n");
550
551 from = EDMA_CHAN_SLOT(from);
552 to = EDMA_CHAN_SLOT(to);
553 if (from >= ecc->num_slots || to >= ecc->num_slots)
554 return;
555
556 edma_param_modify(ecc, PARM_LINK_BCNTRLD, from, 0xffff0000,
557 PARM_OFFSET(to));
558 }
559
560 /**
561 * edma_get_position - returns the current transfer point
562 * @ecc: pointer to edma_cc struct
563 * @slot: parameter RAM slot being examined
564 * @dst: true selects the dest position, false the source
565 *
566 * Returns the position of the current active slot
567 */
568 static dma_addr_t edma_get_position(struct edma_cc *ecc, unsigned slot,
569 bool dst)
570 {
571 u32 offs;
572
573 slot = EDMA_CHAN_SLOT(slot);
574 offs = PARM_OFFSET(slot);
575 offs += dst ? PARM_DST : PARM_SRC;
576
577 return edma_read(ecc, offs);
578 }
579
580 /*
581 * Channels with event associations will be triggered by their hardware
582 * events, and channels without such associations will be triggered by
583 * software. (At this writing there is no interface for using software
584 * triggers except with channels that don't support hardware triggers.)
585 */
586 static void edma_start(struct edma_chan *echan)
587 {
588 struct edma_cc *ecc = echan->ecc;
589 int channel = EDMA_CHAN_SLOT(echan->ch_num);
590 int j = (channel >> 5);
591 unsigned int mask = BIT(channel & 0x1f);
592
593 if (!echan->hw_triggered) {
594 /* EDMA channels without event association */
595 dev_dbg(ecc->dev, "ESR%d %08x\n", j,
596 edma_shadow0_read_array(ecc, SH_ESR, j));
597 edma_shadow0_write_array(ecc, SH_ESR, j, mask);
598 } else {
599 /* EDMA channel with event association */
600 dev_dbg(ecc->dev, "ER%d %08x\n", j,
601 edma_shadow0_read_array(ecc, SH_ER, j));
602 /* Clear any pending event or error */
603 edma_write_array(ecc, EDMA_ECR, j, mask);
604 edma_write_array(ecc, EDMA_EMCR, j, mask);
605 /* Clear any SER */
606 edma_shadow0_write_array(ecc, SH_SECR, j, mask);
607 edma_shadow0_write_array(ecc, SH_EESR, j, mask);
608 dev_dbg(ecc->dev, "EER%d %08x\n", j,
609 edma_shadow0_read_array(ecc, SH_EER, j));
610 }
611 }
612
613 static void edma_stop(struct edma_chan *echan)
614 {
615 struct edma_cc *ecc = echan->ecc;
616 int channel = EDMA_CHAN_SLOT(echan->ch_num);
617 int j = (channel >> 5);
618 unsigned int mask = BIT(channel & 0x1f);
619
620 edma_shadow0_write_array(ecc, SH_EECR, j, mask);
621 edma_shadow0_write_array(ecc, SH_ECR, j, mask);
622 edma_shadow0_write_array(ecc, SH_SECR, j, mask);
623 edma_write_array(ecc, EDMA_EMCR, j, mask);
624
625 /* clear possibly pending completion interrupt */
626 edma_shadow0_write_array(ecc, SH_ICR, j, mask);
627
628 dev_dbg(ecc->dev, "EER%d %08x\n", j,
629 edma_shadow0_read_array(ecc, SH_EER, j));
630
631 /* REVISIT: consider guarding against inappropriate event
632 * chaining by overwriting with dummy_paramset.
633 */
634 }
635
636 /*
637 * Temporarily disable EDMA hardware events on the specified channel,
638 * preventing them from triggering new transfers
639 */
640 static void edma_pause(struct edma_chan *echan)
641 {
642 int channel = EDMA_CHAN_SLOT(echan->ch_num);
643 unsigned int mask = BIT(channel & 0x1f);
644
645 edma_shadow0_write_array(echan->ecc, SH_EECR, channel >> 5, mask);
646 }
647
648 /* Re-enable EDMA hardware events on the specified channel. */
649 static void edma_resume(struct edma_chan *echan)
650 {
651 int channel = EDMA_CHAN_SLOT(echan->ch_num);
652 unsigned int mask = BIT(channel & 0x1f);
653
654 edma_shadow0_write_array(echan->ecc, SH_EESR, channel >> 5, mask);
655 }
656
657 static void edma_trigger_channel(struct edma_chan *echan)
658 {
659 struct edma_cc *ecc = echan->ecc;
660 int channel = EDMA_CHAN_SLOT(echan->ch_num);
661 unsigned int mask = BIT(channel & 0x1f);
662
663 edma_shadow0_write_array(ecc, SH_ESR, (channel >> 5), mask);
664
665 dev_dbg(ecc->dev, "ESR%d %08x\n", (channel >> 5),
666 edma_shadow0_read_array(ecc, SH_ESR, (channel >> 5)));
667 }
668
669 static void edma_clean_channel(struct edma_chan *echan)
670 {
671 struct edma_cc *ecc = echan->ecc;
672 int channel = EDMA_CHAN_SLOT(echan->ch_num);
673 int j = (channel >> 5);
674 unsigned int mask = BIT(channel & 0x1f);
675
676 dev_dbg(ecc->dev, "EMR%d %08x\n", j, edma_read_array(ecc, EDMA_EMR, j));
677 edma_shadow0_write_array(ecc, SH_ECR, j, mask);
678 /* Clear the corresponding EMR bits */
679 edma_write_array(ecc, EDMA_EMCR, j, mask);
680 /* Clear any SER */
681 edma_shadow0_write_array(ecc, SH_SECR, j, mask);
682 edma_write(ecc, EDMA_CCERRCLR, BIT(16) | BIT(1) | BIT(0));
683 }
684
685 /* Move channel to a specific event queue */
686 static void edma_assign_channel_eventq(struct edma_chan *echan,
687 enum dma_event_q eventq_no)
688 {
689 struct edma_cc *ecc = echan->ecc;
690 int channel = EDMA_CHAN_SLOT(echan->ch_num);
691 int bit = (channel & 0x7) * 4;
692
693 /* default to low priority queue */
694 if (eventq_no == EVENTQ_DEFAULT)
695 eventq_no = ecc->default_queue;
696 if (eventq_no >= ecc->num_tc)
697 return;
698
699 eventq_no &= 7;
700 edma_modify_array(ecc, EDMA_DMAQNUM, (channel >> 3), ~(0x7 << bit),
701 eventq_no << bit);
702 }
703
704 static int edma_alloc_channel(struct edma_chan *echan,
705 enum dma_event_q eventq_no)
706 {
707 struct edma_cc *ecc = echan->ecc;
708 int channel = EDMA_CHAN_SLOT(echan->ch_num);
709
710 /* ensure access through shadow region 0 */
711 edma_or_array2(ecc, EDMA_DRAE, 0, channel >> 5, BIT(channel & 0x1f));
712
713 /* ensure no events are pending */
714 edma_stop(echan);
715
716 edma_setup_interrupt(echan, true);
717
718 edma_assign_channel_eventq(echan, eventq_no);
719
720 return 0;
721 }
722
723 static void edma_free_channel(struct edma_chan *echan)
724 {
725 /* ensure no events are pending */
726 edma_stop(echan);
727 /* REVISIT should probably take out of shadow region 0 */
728 edma_setup_interrupt(echan, false);
729 }
730
731 static inline struct edma_cc *to_edma_cc(struct dma_device *d)
732 {
733 return container_of(d, struct edma_cc, dma_slave);
734 }
735
736 static inline struct edma_chan *to_edma_chan(struct dma_chan *c)
737 {
738 return container_of(c, struct edma_chan, vchan.chan);
739 }
740
741 static inline struct edma_desc *to_edma_desc(struct dma_async_tx_descriptor *tx)
742 {
743 return container_of(tx, struct edma_desc, vdesc.tx);
744 }
745
746 static void edma_desc_free(struct virt_dma_desc *vdesc)
747 {
748 kfree(container_of(vdesc, struct edma_desc, vdesc));
749 }
750
751 /* Dispatch a queued descriptor to the controller (caller holds lock) */
752 static void edma_execute(struct edma_chan *echan)
753 {
754 struct edma_cc *ecc = echan->ecc;
755 struct virt_dma_desc *vdesc;
756 struct edma_desc *edesc;
757 struct device *dev = echan->vchan.chan.device->dev;
758 int i, j, left, nslots;
759
760 if (!echan->edesc) {
761 /* Setup is needed for the first transfer */
762 vdesc = vchan_next_desc(&echan->vchan);
763 if (!vdesc)
764 return;
765 list_del(&vdesc->node);
766 echan->edesc = to_edma_desc(&vdesc->tx);
767 }
768
769 edesc = echan->edesc;
770
771 /* Find out how many left */
772 left = edesc->pset_nr - edesc->processed;
773 nslots = min(MAX_NR_SG, left);
774 edesc->sg_len = 0;
775
776 /* Write descriptor PaRAM set(s) */
777 for (i = 0; i < nslots; i++) {
778 j = i + edesc->processed;
779 edma_write_slot(ecc, echan->slot[i], &edesc->pset[j].param);
780 edesc->sg_len += edesc->pset[j].len;
781 dev_vdbg(dev,
782 "\n pset[%d]:\n"
783 " chnum\t%d\n"
784 " slot\t%d\n"
785 " opt\t%08x\n"
786 " src\t%08x\n"
787 " dst\t%08x\n"
788 " abcnt\t%08x\n"
789 " ccnt\t%08x\n"
790 " bidx\t%08x\n"
791 " cidx\t%08x\n"
792 " lkrld\t%08x\n",
793 j, echan->ch_num, echan->slot[i],
794 edesc->pset[j].param.opt,
795 edesc->pset[j].param.src,
796 edesc->pset[j].param.dst,
797 edesc->pset[j].param.a_b_cnt,
798 edesc->pset[j].param.ccnt,
799 edesc->pset[j].param.src_dst_bidx,
800 edesc->pset[j].param.src_dst_cidx,
801 edesc->pset[j].param.link_bcntrld);
802 /* Link to the previous slot if not the last set */
803 if (i != (nslots - 1))
804 edma_link(ecc, echan->slot[i], echan->slot[i + 1]);
805 }
806
807 edesc->processed += nslots;
808
809 /*
810 * If this is either the last set in a set of SG-list transactions
811 * then setup a link to the dummy slot, this results in all future
812 * events being absorbed and that's OK because we're done
813 */
814 if (edesc->processed == edesc->pset_nr) {
815 if (edesc->cyclic)
816 edma_link(ecc, echan->slot[nslots - 1], echan->slot[1]);
817 else
818 edma_link(ecc, echan->slot[nslots - 1],
819 echan->ecc->dummy_slot);
820 }
821
822 if (echan->missed) {
823 /*
824 * This happens due to setup times between intermediate
825 * transfers in long SG lists which have to be broken up into
826 * transfers of MAX_NR_SG
827 */
828 dev_dbg(dev, "missed event on channel %d\n", echan->ch_num);
829 edma_clean_channel(echan);
830 edma_stop(echan);
831 edma_start(echan);
832 edma_trigger_channel(echan);
833 echan->missed = 0;
834 } else if (edesc->processed <= MAX_NR_SG) {
835 dev_dbg(dev, "first transfer starting on channel %d\n",
836 echan->ch_num);
837 edma_start(echan);
838 } else {
839 dev_dbg(dev, "chan: %d: completed %d elements, resuming\n",
840 echan->ch_num, edesc->processed);
841 edma_resume(echan);
842 }
843 }
844
845 static int edma_terminate_all(struct dma_chan *chan)
846 {
847 struct edma_chan *echan = to_edma_chan(chan);
848 unsigned long flags;
849 LIST_HEAD(head);
850
851 spin_lock_irqsave(&echan->vchan.lock, flags);
852
853 /*
854 * Stop DMA activity: we assume the callback will not be called
855 * after edma_dma() returns (even if it does, it will see
856 * echan->edesc is NULL and exit.)
857 */
858 if (echan->edesc) {
859 edma_stop(echan);
860 /* Move the cyclic channel back to default queue */
861 if (!echan->tc && echan->edesc->cyclic)
862 edma_assign_channel_eventq(echan, EVENTQ_DEFAULT);
863 /*
864 * free the running request descriptor
865 * since it is not in any of the vdesc lists
866 */
867 edma_desc_free(&echan->edesc->vdesc);
868 echan->edesc = NULL;
869 }
870
871 vchan_get_all_descriptors(&echan->vchan, &head);
872 spin_unlock_irqrestore(&echan->vchan.lock, flags);
873 vchan_dma_desc_free_list(&echan->vchan, &head);
874
875 return 0;
876 }
877
878 static void edma_synchronize(struct dma_chan *chan)
879 {
880 struct edma_chan *echan = to_edma_chan(chan);
881
882 vchan_synchronize(&echan->vchan);
883 }
884
885 static int edma_slave_config(struct dma_chan *chan,
886 struct dma_slave_config *cfg)
887 {
888 struct edma_chan *echan = to_edma_chan(chan);
889
890 if (cfg->src_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES ||
891 cfg->dst_addr_width == DMA_SLAVE_BUSWIDTH_8_BYTES)
892 return -EINVAL;
893
894 memcpy(&echan->cfg, cfg, sizeof(echan->cfg));
895
896 return 0;
897 }
898
899 static int edma_dma_pause(struct dma_chan *chan)
900 {
901 struct edma_chan *echan = to_edma_chan(chan);
902
903 if (!echan->edesc)
904 return -EINVAL;
905
906 edma_pause(echan);
907 return 0;
908 }
909
910 static int edma_dma_resume(struct dma_chan *chan)
911 {
912 struct edma_chan *echan = to_edma_chan(chan);
913
914 edma_resume(echan);
915 return 0;
916 }
917
918 /*
919 * A PaRAM set configuration abstraction used by other modes
920 * @chan: Channel who's PaRAM set we're configuring
921 * @pset: PaRAM set to initialize and setup.
922 * @src_addr: Source address of the DMA
923 * @dst_addr: Destination address of the DMA
924 * @burst: In units of dev_width, how much to send
925 * @dev_width: How much is the dev_width
926 * @dma_length: Total length of the DMA transfer
927 * @direction: Direction of the transfer
928 */
929 static int edma_config_pset(struct dma_chan *chan, struct edma_pset *epset,
930 dma_addr_t src_addr, dma_addr_t dst_addr, u32 burst,
931 unsigned int acnt, unsigned int dma_length,
932 enum dma_transfer_direction direction)
933 {
934 struct edma_chan *echan = to_edma_chan(chan);
935 struct device *dev = chan->device->dev;
936 struct edmacc_param *param = &epset->param;
937 int bcnt, ccnt, cidx;
938 int src_bidx, dst_bidx, src_cidx, dst_cidx;
939 int absync;
940
941 /* src/dst_maxburst == 0 is the same case as src/dst_maxburst == 1 */
942 if (!burst)
943 burst = 1;
944 /*
945 * If the maxburst is equal to the fifo width, use
946 * A-synced transfers. This allows for large contiguous
947 * buffer transfers using only one PaRAM set.
948 */
949 if (burst == 1) {
950 /*
951 * For the A-sync case, bcnt and ccnt are the remainder
952 * and quotient respectively of the division of:
953 * (dma_length / acnt) by (SZ_64K -1). This is so
954 * that in case bcnt over flows, we have ccnt to use.
955 * Note: In A-sync tranfer only, bcntrld is used, but it
956 * only applies for sg_dma_len(sg) >= SZ_64K.
957 * In this case, the best way adopted is- bccnt for the
958 * first frame will be the remainder below. Then for
959 * every successive frame, bcnt will be SZ_64K-1. This
960 * is assured as bcntrld = 0xffff in end of function.
961 */
962 absync = false;
963 ccnt = dma_length / acnt / (SZ_64K - 1);
964 bcnt = dma_length / acnt - ccnt * (SZ_64K - 1);
965 /*
966 * If bcnt is non-zero, we have a remainder and hence an
967 * extra frame to transfer, so increment ccnt.
968 */
969 if (bcnt)
970 ccnt++;
971 else
972 bcnt = SZ_64K - 1;
973 cidx = acnt;
974 } else {
975 /*
976 * If maxburst is greater than the fifo address_width,
977 * use AB-synced transfers where A count is the fifo
978 * address_width and B count is the maxburst. In this
979 * case, we are limited to transfers of C count frames
980 * of (address_width * maxburst) where C count is limited
981 * to SZ_64K-1. This places an upper bound on the length
982 * of an SG segment that can be handled.
983 */
984 absync = true;
985 bcnt = burst;
986 ccnt = dma_length / (acnt * bcnt);
987 if (ccnt > (SZ_64K - 1)) {
988 dev_err(dev, "Exceeded max SG segment size\n");
989 return -EINVAL;
990 }
991 cidx = acnt * bcnt;
992 }
993
994 epset->len = dma_length;
995
996 if (direction == DMA_MEM_TO_DEV) {
997 src_bidx = acnt;
998 src_cidx = cidx;
999 dst_bidx = 0;
1000 dst_cidx = 0;
1001 epset->addr = src_addr;
1002 } else if (direction == DMA_DEV_TO_MEM) {
1003 src_bidx = 0;
1004 src_cidx = 0;
1005 dst_bidx = acnt;
1006 dst_cidx = cidx;
1007 epset->addr = dst_addr;
1008 } else if (direction == DMA_MEM_TO_MEM) {
1009 src_bidx = acnt;
1010 src_cidx = cidx;
1011 dst_bidx = acnt;
1012 dst_cidx = cidx;
1013 } else {
1014 dev_err(dev, "%s: direction not implemented yet\n", __func__);
1015 return -EINVAL;
1016 }
1017
1018 param->opt = EDMA_TCC(EDMA_CHAN_SLOT(echan->ch_num));
1019 /* Configure A or AB synchronized transfers */
1020 if (absync)
1021 param->opt |= SYNCDIM;
1022
1023 param->src = src_addr;
1024 param->dst = dst_addr;
1025
1026 param->src_dst_bidx = (dst_bidx << 16) | src_bidx;
1027 param->src_dst_cidx = (dst_cidx << 16) | src_cidx;
1028
1029 param->a_b_cnt = bcnt << 16 | acnt;
1030 param->ccnt = ccnt;
1031 /*
1032 * Only time when (bcntrld) auto reload is required is for
1033 * A-sync case, and in this case, a requirement of reload value
1034 * of SZ_64K-1 only is assured. 'link' is initially set to NULL
1035 * and then later will be populated by edma_execute.
1036 */
1037 param->link_bcntrld = 0xffffffff;
1038 return absync;
1039 }
1040
1041 static struct dma_async_tx_descriptor *edma_prep_slave_sg(
1042 struct dma_chan *chan, struct scatterlist *sgl,
1043 unsigned int sg_len, enum dma_transfer_direction direction,
1044 unsigned long tx_flags, void *context)
1045 {
1046 struct edma_chan *echan = to_edma_chan(chan);
1047 struct device *dev = chan->device->dev;
1048 struct edma_desc *edesc;
1049 dma_addr_t src_addr = 0, dst_addr = 0;
1050 enum dma_slave_buswidth dev_width;
1051 u32 burst;
1052 struct scatterlist *sg;
1053 int i, nslots, ret;
1054
1055 if (unlikely(!echan || !sgl || !sg_len))
1056 return NULL;
1057
1058 if (direction == DMA_DEV_TO_MEM) {
1059 src_addr = echan->cfg.src_addr;
1060 dev_width = echan->cfg.src_addr_width;
1061 burst = echan->cfg.src_maxburst;
1062 } else if (direction == DMA_MEM_TO_DEV) {
1063 dst_addr = echan->cfg.dst_addr;
1064 dev_width = echan->cfg.dst_addr_width;
1065 burst = echan->cfg.dst_maxburst;
1066 } else {
1067 dev_err(dev, "%s: bad direction: %d\n", __func__, direction);
1068 return NULL;
1069 }
1070
1071 if (dev_width == DMA_SLAVE_BUSWIDTH_UNDEFINED) {
1072 dev_err(dev, "%s: Undefined slave buswidth\n", __func__);
1073 return NULL;
1074 }
1075
1076 edesc = kzalloc(sizeof(*edesc) + sg_len * sizeof(edesc->pset[0]),
1077 GFP_ATOMIC);
1078 if (!edesc)
1079 return NULL;
1080
1081 edesc->pset_nr = sg_len;
1082 edesc->residue = 0;
1083 edesc->direction = direction;
1084 edesc->echan = echan;
1085
1086 /* Allocate a PaRAM slot, if needed */
1087 nslots = min_t(unsigned, MAX_NR_SG, sg_len);
1088
1089 for (i = 0; i < nslots; i++) {
1090 if (echan->slot[i] < 0) {
1091 echan->slot[i] =
1092 edma_alloc_slot(echan->ecc, EDMA_SLOT_ANY);
1093 if (echan->slot[i] < 0) {
1094 kfree(edesc);
1095 dev_err(dev, "%s: Failed to allocate slot\n",
1096 __func__);
1097 return NULL;
1098 }
1099 }
1100 }
1101
1102 /* Configure PaRAM sets for each SG */
1103 for_each_sg(sgl, sg, sg_len, i) {
1104 /* Get address for each SG */
1105 if (direction == DMA_DEV_TO_MEM)
1106 dst_addr = sg_dma_address(sg);
1107 else
1108 src_addr = sg_dma_address(sg);
1109
1110 ret = edma_config_pset(chan, &edesc->pset[i], src_addr,
1111 dst_addr, burst, dev_width,
1112 sg_dma_len(sg), direction);
1113 if (ret < 0) {
1114 kfree(edesc);
1115 return NULL;
1116 }
1117
1118 edesc->absync = ret;
1119 edesc->residue += sg_dma_len(sg);
1120
1121 if (i == sg_len - 1)
1122 /* Enable completion interrupt */
1123 edesc->pset[i].param.opt |= TCINTEN;
1124 else if (!((i+1) % MAX_NR_SG))
1125 /*
1126 * Enable early completion interrupt for the
1127 * intermediateset. In this case the driver will be
1128 * notified when the paRAM set is submitted to TC. This
1129 * will allow more time to set up the next set of slots.
1130 */
1131 edesc->pset[i].param.opt |= (TCINTEN | TCCMODE);
1132 }
1133 edesc->residue_stat = edesc->residue;
1134
1135 return vchan_tx_prep(&echan->vchan, &edesc->vdesc, tx_flags);
1136 }
1137
1138 static struct dma_async_tx_descriptor *edma_prep_dma_memcpy(
1139 struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
1140 size_t len, unsigned long tx_flags)
1141 {
1142 int ret, nslots;
1143 struct edma_desc *edesc;
1144 struct device *dev = chan->device->dev;
1145 struct edma_chan *echan = to_edma_chan(chan);
1146 unsigned int width, pset_len;
1147
1148 if (unlikely(!echan || !len))
1149 return NULL;
1150
1151 if (len < SZ_64K) {
1152 /*
1153 * Transfer size less than 64K can be handled with one paRAM
1154 * slot and with one burst.
1155 * ACNT = length
1156 */
1157 width = len;
1158 pset_len = len;
1159 nslots = 1;
1160 } else {
1161 /*
1162 * Transfer size bigger than 64K will be handled with maximum of
1163 * two paRAM slots.
1164 * slot1: (full_length / 32767) times 32767 bytes bursts.
1165 * ACNT = 32767, length1: (full_length / 32767) * 32767
1166 * slot2: the remaining amount of data after slot1.
1167 * ACNT = full_length - length1, length2 = ACNT
1168 *
1169 * When the full_length is multibple of 32767 one slot can be
1170 * used to complete the transfer.
1171 */
1172 width = SZ_32K - 1;
1173 pset_len = rounddown(len, width);
1174 /* One slot is enough for lengths multiple of (SZ_32K -1) */
1175 if (unlikely(pset_len == len))
1176 nslots = 1;
1177 else
1178 nslots = 2;
1179 }
1180
1181 edesc = kzalloc(sizeof(*edesc) + nslots * sizeof(edesc->pset[0]),
1182 GFP_ATOMIC);
1183 if (!edesc)
1184 return NULL;
1185
1186 edesc->pset_nr = nslots;
1187 edesc->residue = edesc->residue_stat = len;
1188 edesc->direction = DMA_MEM_TO_MEM;
1189 edesc->echan = echan;
1190
1191 ret = edma_config_pset(chan, &edesc->pset[0], src, dest, 1,
1192 width, pset_len, DMA_MEM_TO_MEM);
1193 if (ret < 0) {
1194 kfree(edesc);
1195 return NULL;
1196 }
1197
1198 edesc->absync = ret;
1199
1200 edesc->pset[0].param.opt |= ITCCHEN;
1201 if (nslots == 1) {
1202 /* Enable transfer complete interrupt */
1203 edesc->pset[0].param.opt |= TCINTEN;
1204 } else {
1205 /* Enable transfer complete chaining for the first slot */
1206 edesc->pset[0].param.opt |= TCCHEN;
1207
1208 if (echan->slot[1] < 0) {
1209 echan->slot[1] = edma_alloc_slot(echan->ecc,
1210 EDMA_SLOT_ANY);
1211 if (echan->slot[1] < 0) {
1212 kfree(edesc);
1213 dev_err(dev, "%s: Failed to allocate slot\n",
1214 __func__);
1215 return NULL;
1216 }
1217 }
1218 dest += pset_len;
1219 src += pset_len;
1220 pset_len = width = len % (SZ_32K - 1);
1221
1222 ret = edma_config_pset(chan, &edesc->pset[1], src, dest, 1,
1223 width, pset_len, DMA_MEM_TO_MEM);
1224 if (ret < 0) {
1225 kfree(edesc);
1226 return NULL;
1227 }
1228
1229 edesc->pset[1].param.opt |= ITCCHEN;
1230 edesc->pset[1].param.opt |= TCINTEN;
1231 }
1232
1233 return vchan_tx_prep(&echan->vchan, &edesc->vdesc, tx_flags);
1234 }
1235
1236 static struct dma_async_tx_descriptor *edma_prep_dma_cyclic(
1237 struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len,
1238 size_t period_len, enum dma_transfer_direction direction,
1239 unsigned long tx_flags)
1240 {
1241 struct edma_chan *echan = to_edma_chan(chan);
1242 struct device *dev = chan->device->dev;
1243 struct edma_desc *edesc;
1244 dma_addr_t src_addr, dst_addr;
1245 enum dma_slave_buswidth dev_width;
1246 bool use_intermediate = false;
1247 u32 burst;
1248 int i, ret, nslots;
1249
1250 if (unlikely(!echan || !buf_len || !period_len))
1251 return NULL;
1252
1253 if (direction == DMA_DEV_TO_MEM) {
1254 src_addr = echan->cfg.src_addr;
1255 dst_addr = buf_addr;
1256 dev_width = echan->cfg.src_addr_width;
1257 burst = echan->cfg.src_maxburst;
1258 } else if (direction == DMA_MEM_TO_DEV) {
1259 src_addr = buf_addr;
1260 dst_addr = echan->cfg.dst_addr;
1261 dev_width = echan->cfg.dst_addr_width;
1262 burst = echan->cfg.dst_maxburst;
1263 } else {
1264 dev_err(dev, "%s: bad direction: %d\n", __func__, direction);
1265 return NULL;
1266 }
1267
1268 if (dev_width == DMA_SLAVE_BUSWIDTH_UNDEFINED) {
1269 dev_err(dev, "%s: Undefined slave buswidth\n", __func__);
1270 return NULL;
1271 }
1272
1273 if (unlikely(buf_len % period_len)) {
1274 dev_err(dev, "Period should be multiple of Buffer length\n");
1275 return NULL;
1276 }
1277
1278 nslots = (buf_len / period_len) + 1;
1279
1280 /*
1281 * Cyclic DMA users such as audio cannot tolerate delays introduced
1282 * by cases where the number of periods is more than the maximum
1283 * number of SGs the EDMA driver can handle at a time. For DMA types
1284 * such as Slave SGs, such delays are tolerable and synchronized,
1285 * but the synchronization is difficult to achieve with Cyclic and
1286 * cannot be guaranteed, so we error out early.
1287 */
1288 if (nslots > MAX_NR_SG) {
1289 /*
1290 * If the burst and period sizes are the same, we can put
1291 * the full buffer into a single period and activate
1292 * intermediate interrupts. This will produce interrupts
1293 * after each burst, which is also after each desired period.
1294 */
1295 if (burst == period_len) {
1296 period_len = buf_len;
1297 nslots = 2;
1298 use_intermediate = true;
1299 } else {
1300 return NULL;
1301 }
1302 }
1303
1304 edesc = kzalloc(sizeof(*edesc) + nslots * sizeof(edesc->pset[0]),
1305 GFP_ATOMIC);
1306 if (!edesc)
1307 return NULL;
1308
1309 edesc->cyclic = 1;
1310 edesc->pset_nr = nslots;
1311 edesc->residue = edesc->residue_stat = buf_len;
1312 edesc->direction = direction;
1313 edesc->echan = echan;
1314
1315 dev_dbg(dev, "%s: channel=%d nslots=%d period_len=%zu buf_len=%zu\n",
1316 __func__, echan->ch_num, nslots, period_len, buf_len);
1317
1318 for (i = 0; i < nslots; i++) {
1319 /* Allocate a PaRAM slot, if needed */
1320 if (echan->slot[i] < 0) {
1321 echan->slot[i] =
1322 edma_alloc_slot(echan->ecc, EDMA_SLOT_ANY);
1323 if (echan->slot[i] < 0) {
1324 kfree(edesc);
1325 dev_err(dev, "%s: Failed to allocate slot\n",
1326 __func__);
1327 return NULL;
1328 }
1329 }
1330
1331 if (i == nslots - 1) {
1332 memcpy(&edesc->pset[i], &edesc->pset[0],
1333 sizeof(edesc->pset[0]));
1334 break;
1335 }
1336
1337 ret = edma_config_pset(chan, &edesc->pset[i], src_addr,
1338 dst_addr, burst, dev_width, period_len,
1339 direction);
1340 if (ret < 0) {
1341 kfree(edesc);
1342 return NULL;
1343 }
1344
1345 if (direction == DMA_DEV_TO_MEM)
1346 dst_addr += period_len;
1347 else
1348 src_addr += period_len;
1349
1350 dev_vdbg(dev, "%s: Configure period %d of buf:\n", __func__, i);
1351 dev_vdbg(dev,
1352 "\n pset[%d]:\n"
1353 " chnum\t%d\n"
1354 " slot\t%d\n"
1355 " opt\t%08x\n"
1356 " src\t%08x\n"
1357 " dst\t%08x\n"
1358 " abcnt\t%08x\n"
1359 " ccnt\t%08x\n"
1360 " bidx\t%08x\n"
1361 " cidx\t%08x\n"
1362 " lkrld\t%08x\n",
1363 i, echan->ch_num, echan->slot[i],
1364 edesc->pset[i].param.opt,
1365 edesc->pset[i].param.src,
1366 edesc->pset[i].param.dst,
1367 edesc->pset[i].param.a_b_cnt,
1368 edesc->pset[i].param.ccnt,
1369 edesc->pset[i].param.src_dst_bidx,
1370 edesc->pset[i].param.src_dst_cidx,
1371 edesc->pset[i].param.link_bcntrld);
1372
1373 edesc->absync = ret;
1374
1375 /*
1376 * Enable period interrupt only if it is requested
1377 */
1378 if (tx_flags & DMA_PREP_INTERRUPT) {
1379 edesc->pset[i].param.opt |= TCINTEN;
1380
1381 /* Also enable intermediate interrupts if necessary */
1382 if (use_intermediate)
1383 edesc->pset[i].param.opt |= ITCINTEN;
1384 }
1385 }
1386
1387 /* Place the cyclic channel to highest priority queue */
1388 if (!echan->tc)
1389 edma_assign_channel_eventq(echan, EVENTQ_0);
1390
1391 return vchan_tx_prep(&echan->vchan, &edesc->vdesc, tx_flags);
1392 }
1393
1394 static void edma_completion_handler(struct edma_chan *echan)
1395 {
1396 struct device *dev = echan->vchan.chan.device->dev;
1397 struct edma_desc *edesc;
1398
1399 spin_lock(&echan->vchan.lock);
1400 edesc = echan->edesc;
1401 if (edesc) {
1402 if (edesc->cyclic) {
1403 vchan_cyclic_callback(&edesc->vdesc);
1404 spin_unlock(&echan->vchan.lock);
1405 return;
1406 } else if (edesc->processed == edesc->pset_nr) {
1407 edesc->residue = 0;
1408 edma_stop(echan);
1409 vchan_cookie_complete(&edesc->vdesc);
1410 echan->edesc = NULL;
1411
1412 dev_dbg(dev, "Transfer completed on channel %d\n",
1413 echan->ch_num);
1414 } else {
1415 dev_dbg(dev, "Sub transfer completed on channel %d\n",
1416 echan->ch_num);
1417
1418 edma_pause(echan);
1419
1420 /* Update statistics for tx_status */
1421 edesc->residue -= edesc->sg_len;
1422 edesc->residue_stat = edesc->residue;
1423 edesc->processed_stat = edesc->processed;
1424 }
1425 edma_execute(echan);
1426 }
1427
1428 spin_unlock(&echan->vchan.lock);
1429 }
1430
1431 /* eDMA interrupt handler */
1432 static irqreturn_t dma_irq_handler(int irq, void *data)
1433 {
1434 struct edma_cc *ecc = data;
1435 int ctlr;
1436 u32 sh_ier;
1437 u32 sh_ipr;
1438 u32 bank;
1439
1440 ctlr = ecc->id;
1441 if (ctlr < 0)
1442 return IRQ_NONE;
1443
1444 dev_vdbg(ecc->dev, "dma_irq_handler\n");
1445
1446 sh_ipr = edma_shadow0_read_array(ecc, SH_IPR, 0);
1447 if (!sh_ipr) {
1448 sh_ipr = edma_shadow0_read_array(ecc, SH_IPR, 1);
1449 if (!sh_ipr)
1450 return IRQ_NONE;
1451 sh_ier = edma_shadow0_read_array(ecc, SH_IER, 1);
1452 bank = 1;
1453 } else {
1454 sh_ier = edma_shadow0_read_array(ecc, SH_IER, 0);
1455 bank = 0;
1456 }
1457
1458 do {
1459 u32 slot;
1460 u32 channel;
1461
1462 slot = __ffs(sh_ipr);
1463 sh_ipr &= ~(BIT(slot));
1464
1465 if (sh_ier & BIT(slot)) {
1466 channel = (bank << 5) | slot;
1467 /* Clear the corresponding IPR bits */
1468 edma_shadow0_write_array(ecc, SH_ICR, bank, BIT(slot));
1469 edma_completion_handler(&ecc->slave_chans[channel]);
1470 }
1471 } while (sh_ipr);
1472
1473 edma_shadow0_write(ecc, SH_IEVAL, 1);
1474 return IRQ_HANDLED;
1475 }
1476
1477 static void edma_error_handler(struct edma_chan *echan)
1478 {
1479 struct edma_cc *ecc = echan->ecc;
1480 struct device *dev = echan->vchan.chan.device->dev;
1481 struct edmacc_param p;
1482 int err;
1483
1484 if (!echan->edesc)
1485 return;
1486
1487 spin_lock(&echan->vchan.lock);
1488
1489 err = edma_read_slot(ecc, echan->slot[0], &p);
1490
1491 /*
1492 * Issue later based on missed flag which will be sure
1493 * to happen as:
1494 * (1) we finished transmitting an intermediate slot and
1495 * edma_execute is coming up.
1496 * (2) or we finished current transfer and issue will
1497 * call edma_execute.
1498 *
1499 * Important note: issuing can be dangerous here and
1500 * lead to some nasty recursion when we are in a NULL
1501 * slot. So we avoid doing so and set the missed flag.
1502 */
1503 if (err || (p.a_b_cnt == 0 && p.ccnt == 0)) {
1504 dev_dbg(dev, "Error on null slot, setting miss\n");
1505 echan->missed = 1;
1506 } else {
1507 /*
1508 * The slot is already programmed but the event got
1509 * missed, so its safe to issue it here.
1510 */
1511 dev_dbg(dev, "Missed event, TRIGGERING\n");
1512 edma_clean_channel(echan);
1513 edma_stop(echan);
1514 edma_start(echan);
1515 edma_trigger_channel(echan);
1516 }
1517 spin_unlock(&echan->vchan.lock);
1518 }
1519
1520 static inline bool edma_error_pending(struct edma_cc *ecc)
1521 {
1522 if (edma_read_array(ecc, EDMA_EMR, 0) ||
1523 edma_read_array(ecc, EDMA_EMR, 1) ||
1524 edma_read(ecc, EDMA_QEMR) || edma_read(ecc, EDMA_CCERR))
1525 return true;
1526
1527 return false;
1528 }
1529
1530 /* eDMA error interrupt handler */
1531 static irqreturn_t dma_ccerr_handler(int irq, void *data)
1532 {
1533 struct edma_cc *ecc = data;
1534 int i, j;
1535 int ctlr;
1536 unsigned int cnt = 0;
1537 unsigned int val;
1538
1539 ctlr = ecc->id;
1540 if (ctlr < 0)
1541 return IRQ_NONE;
1542
1543 dev_vdbg(ecc->dev, "dma_ccerr_handler\n");
1544
1545 if (!edma_error_pending(ecc)) {
1546 /*
1547 * The registers indicate no pending error event but the irq
1548 * handler has been called.
1549 * Ask eDMA to re-evaluate the error registers.
1550 */
1551 dev_err(ecc->dev, "%s: Error interrupt without error event!\n",
1552 __func__);
1553 edma_write(ecc, EDMA_EEVAL, 1);
1554 return IRQ_NONE;
1555 }
1556
1557 while (1) {
1558 /* Event missed register(s) */
1559 for (j = 0; j < 2; j++) {
1560 unsigned long emr;
1561
1562 val = edma_read_array(ecc, EDMA_EMR, j);
1563 if (!val)
1564 continue;
1565
1566 dev_dbg(ecc->dev, "EMR%d 0x%08x\n", j, val);
1567 emr = val;
1568 for (i = find_next_bit(&emr, 32, 0); i < 32;
1569 i = find_next_bit(&emr, 32, i + 1)) {
1570 int k = (j << 5) + i;
1571
1572 /* Clear the corresponding EMR bits */
1573 edma_write_array(ecc, EDMA_EMCR, j, BIT(i));
1574 /* Clear any SER */
1575 edma_shadow0_write_array(ecc, SH_SECR, j,
1576 BIT(i));
1577 edma_error_handler(&ecc->slave_chans[k]);
1578 }
1579 }
1580
1581 val = edma_read(ecc, EDMA_QEMR);
1582 if (val) {
1583 dev_dbg(ecc->dev, "QEMR 0x%02x\n", val);
1584 /* Not reported, just clear the interrupt reason. */
1585 edma_write(ecc, EDMA_QEMCR, val);
1586 edma_shadow0_write(ecc, SH_QSECR, val);
1587 }
1588
1589 val = edma_read(ecc, EDMA_CCERR);
1590 if (val) {
1591 dev_warn(ecc->dev, "CCERR 0x%08x\n", val);
1592 /* Not reported, just clear the interrupt reason. */
1593 edma_write(ecc, EDMA_CCERRCLR, val);
1594 }
1595
1596 if (!edma_error_pending(ecc))
1597 break;
1598 cnt++;
1599 if (cnt > 10)
1600 break;
1601 }
1602 edma_write(ecc, EDMA_EEVAL, 1);
1603 return IRQ_HANDLED;
1604 }
1605
1606 /* Alloc channel resources */
1607 static int edma_alloc_chan_resources(struct dma_chan *chan)
1608 {
1609 struct edma_chan *echan = to_edma_chan(chan);
1610 struct edma_cc *ecc = echan->ecc;
1611 struct device *dev = ecc->dev;
1612 enum dma_event_q eventq_no = EVENTQ_DEFAULT;
1613 int ret;
1614
1615 if (echan->tc) {
1616 eventq_no = echan->tc->id;
1617 } else if (ecc->tc_list) {
1618 /* memcpy channel */
1619 echan->tc = &ecc->tc_list[ecc->info->default_queue];
1620 eventq_no = echan->tc->id;
1621 }
1622
1623 ret = edma_alloc_channel(echan, eventq_no);
1624 if (ret)
1625 return ret;
1626
1627 echan->slot[0] = edma_alloc_slot(ecc, echan->ch_num);
1628 if (echan->slot[0] < 0) {
1629 dev_err(dev, "Entry slot allocation failed for channel %u\n",
1630 EDMA_CHAN_SLOT(echan->ch_num));
1631 ret = echan->slot[0];
1632 goto err_slot;
1633 }
1634
1635 /* Set up channel -> slot mapping for the entry slot */
1636 edma_set_chmap(echan, echan->slot[0]);
1637 echan->alloced = true;
1638
1639 dev_dbg(dev, "Got eDMA channel %d for virt channel %d (%s trigger)\n",
1640 EDMA_CHAN_SLOT(echan->ch_num), chan->chan_id,
1641 echan->hw_triggered ? "HW" : "SW");
1642
1643 return 0;
1644
1645 err_slot:
1646 edma_free_channel(echan);
1647 return ret;
1648 }
1649
1650 /* Free channel resources */
1651 static void edma_free_chan_resources(struct dma_chan *chan)
1652 {
1653 struct edma_chan *echan = to_edma_chan(chan);
1654 struct device *dev = echan->ecc->dev;
1655 int i;
1656
1657 /* Terminate transfers */
1658 edma_stop(echan);
1659
1660 vchan_free_chan_resources(&echan->vchan);
1661
1662 /* Free EDMA PaRAM slots */
1663 for (i = 0; i < EDMA_MAX_SLOTS; i++) {
1664 if (echan->slot[i] >= 0) {
1665 edma_free_slot(echan->ecc, echan->slot[i]);
1666 echan->slot[i] = -1;
1667 }
1668 }
1669
1670 /* Set entry slot to the dummy slot */
1671 edma_set_chmap(echan, echan->ecc->dummy_slot);
1672
1673 /* Free EDMA channel */
1674 if (echan->alloced) {
1675 edma_free_channel(echan);
1676 echan->alloced = false;
1677 }
1678
1679 echan->tc = NULL;
1680 echan->hw_triggered = false;
1681
1682 dev_dbg(dev, "Free eDMA channel %d for virt channel %d\n",
1683 EDMA_CHAN_SLOT(echan->ch_num), chan->chan_id);
1684 }
1685
1686 /* Send pending descriptor to hardware */
1687 static void edma_issue_pending(struct dma_chan *chan)
1688 {
1689 struct edma_chan *echan = to_edma_chan(chan);
1690 unsigned long flags;
1691
1692 spin_lock_irqsave(&echan->vchan.lock, flags);
1693 if (vchan_issue_pending(&echan->vchan) && !echan->edesc)
1694 edma_execute(echan);
1695 spin_unlock_irqrestore(&echan->vchan.lock, flags);
1696 }
1697
1698 /*
1699 * This limit exists to avoid a possible infinite loop when waiting for proof
1700 * that a particular transfer is completed. This limit can be hit if there
1701 * are large bursts to/from slow devices or the CPU is never able to catch
1702 * the DMA hardware idle. On an AM335x transfering 48 bytes from the UART
1703 * RX-FIFO, as many as 55 loops have been seen.
1704 */
1705 #define EDMA_MAX_TR_WAIT_LOOPS 1000
1706
1707 static u32 edma_residue(struct edma_desc *edesc)
1708 {
1709 bool dst = edesc->direction == DMA_DEV_TO_MEM;
1710 int loop_count = EDMA_MAX_TR_WAIT_LOOPS;
1711 struct edma_chan *echan = edesc->echan;
1712 struct edma_pset *pset = edesc->pset;
1713 dma_addr_t done, pos;
1714 int i;
1715
1716 /*
1717 * We always read the dst/src position from the first RamPar
1718 * pset. That's the one which is active now.
1719 */
1720 pos = edma_get_position(echan->ecc, echan->slot[0], dst);
1721
1722 /*
1723 * "pos" may represent a transfer request that is still being
1724 * processed by the EDMACC or EDMATC. We will busy wait until
1725 * any one of the situations occurs:
1726 * 1. the DMA hardware is idle
1727 * 2. a new transfer request is setup
1728 * 3. we hit the loop limit
1729 */
1730 while (edma_read(echan->ecc, EDMA_CCSTAT) & EDMA_CCSTAT_ACTV) {
1731 /* check if a new transfer request is setup */
1732 if (edma_get_position(echan->ecc,
1733 echan->slot[0], dst) != pos) {
1734 break;
1735 }
1736
1737 if (!--loop_count) {
1738 dev_dbg_ratelimited(echan->vchan.chan.device->dev,
1739 "%s: timeout waiting for PaRAM update\n",
1740 __func__);
1741 break;
1742 }
1743
1744 cpu_relax();
1745 }
1746
1747 /*
1748 * Cyclic is simple. Just subtract pset[0].addr from pos.
1749 *
1750 * We never update edesc->residue in the cyclic case, so we
1751 * can tell the remaining room to the end of the circular
1752 * buffer.
1753 */
1754 if (edesc->cyclic) {
1755 done = pos - pset->addr;
1756 edesc->residue_stat = edesc->residue - done;
1757 return edesc->residue_stat;
1758 }
1759
1760 /*
1761 * For SG operation we catch up with the last processed
1762 * status.
1763 */
1764 pset += edesc->processed_stat;
1765
1766 for (i = edesc->processed_stat; i < edesc->processed; i++, pset++) {
1767 /*
1768 * If we are inside this pset address range, we know
1769 * this is the active one. Get the current delta and
1770 * stop walking the psets.
1771 */
1772 if (pos >= pset->addr && pos < pset->addr + pset->len)
1773 return edesc->residue_stat - (pos - pset->addr);
1774
1775 /* Otherwise mark it done and update residue_stat. */
1776 edesc->processed_stat++;
1777 edesc->residue_stat -= pset->len;
1778 }
1779 return edesc->residue_stat;
1780 }
1781
1782 /* Check request completion status */
1783 static enum dma_status edma_tx_status(struct dma_chan *chan,
1784 dma_cookie_t cookie,
1785 struct dma_tx_state *txstate)
1786 {
1787 struct edma_chan *echan = to_edma_chan(chan);
1788 struct virt_dma_desc *vdesc;
1789 enum dma_status ret;
1790 unsigned long flags;
1791
1792 ret = dma_cookie_status(chan, cookie, txstate);
1793 if (ret == DMA_COMPLETE || !txstate)
1794 return ret;
1795
1796 spin_lock_irqsave(&echan->vchan.lock, flags);
1797 if (echan->edesc && echan->edesc->vdesc.tx.cookie == cookie)
1798 txstate->residue = edma_residue(echan->edesc);
1799 else if ((vdesc = vchan_find_desc(&echan->vchan, cookie)))
1800 txstate->residue = to_edma_desc(&vdesc->tx)->residue;
1801 spin_unlock_irqrestore(&echan->vchan.lock, flags);
1802
1803 return ret;
1804 }
1805
1806 static bool edma_is_memcpy_channel(int ch_num, s32 *memcpy_channels)
1807 {
1808 if (!memcpy_channels)
1809 return false;
1810 while (*memcpy_channels != -1) {
1811 if (*memcpy_channels == ch_num)
1812 return true;
1813 memcpy_channels++;
1814 }
1815 return false;
1816 }
1817
1818 #define EDMA_DMA_BUSWIDTHS (BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
1819 BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
1820 BIT(DMA_SLAVE_BUSWIDTH_3_BYTES) | \
1821 BIT(DMA_SLAVE_BUSWIDTH_4_BYTES))
1822
1823 static void edma_dma_init(struct edma_cc *ecc, bool legacy_mode)
1824 {
1825 struct dma_device *s_ddev = &ecc->dma_slave;
1826 struct dma_device *m_ddev = NULL;
1827 s32 *memcpy_channels = ecc->info->memcpy_channels;
1828 int i, j;
1829
1830 dma_cap_zero(s_ddev->cap_mask);
1831 dma_cap_set(DMA_SLAVE, s_ddev->cap_mask);
1832 dma_cap_set(DMA_CYCLIC, s_ddev->cap_mask);
1833 if (ecc->legacy_mode && !memcpy_channels) {
1834 dev_warn(ecc->dev,
1835 "Legacy memcpy is enabled, things might not work\n");
1836
1837 dma_cap_set(DMA_MEMCPY, s_ddev->cap_mask);
1838 s_ddev->device_prep_dma_memcpy = edma_prep_dma_memcpy;
1839 s_ddev->directions = BIT(DMA_MEM_TO_MEM);
1840 }
1841
1842 s_ddev->device_prep_slave_sg = edma_prep_slave_sg;
1843 s_ddev->device_prep_dma_cyclic = edma_prep_dma_cyclic;
1844 s_ddev->device_alloc_chan_resources = edma_alloc_chan_resources;
1845 s_ddev->device_free_chan_resources = edma_free_chan_resources;
1846 s_ddev->device_issue_pending = edma_issue_pending;
1847 s_ddev->device_tx_status = edma_tx_status;
1848 s_ddev->device_config = edma_slave_config;
1849 s_ddev->device_pause = edma_dma_pause;
1850 s_ddev->device_resume = edma_dma_resume;
1851 s_ddev->device_terminate_all = edma_terminate_all;
1852 s_ddev->device_synchronize = edma_synchronize;
1853
1854 s_ddev->src_addr_widths = EDMA_DMA_BUSWIDTHS;
1855 s_ddev->dst_addr_widths = EDMA_DMA_BUSWIDTHS;
1856 s_ddev->directions |= (BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV));
1857 s_ddev->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
1858
1859 s_ddev->dev = ecc->dev;
1860 INIT_LIST_HEAD(&s_ddev->channels);
1861
1862 if (memcpy_channels) {
1863 m_ddev = devm_kzalloc(ecc->dev, sizeof(*m_ddev), GFP_KERNEL);
1864 ecc->dma_memcpy = m_ddev;
1865
1866 dma_cap_zero(m_ddev->cap_mask);
1867 dma_cap_set(DMA_MEMCPY, m_ddev->cap_mask);
1868
1869 m_ddev->device_prep_dma_memcpy = edma_prep_dma_memcpy;
1870 m_ddev->device_alloc_chan_resources = edma_alloc_chan_resources;
1871 m_ddev->device_free_chan_resources = edma_free_chan_resources;
1872 m_ddev->device_issue_pending = edma_issue_pending;
1873 m_ddev->device_tx_status = edma_tx_status;
1874 m_ddev->device_config = edma_slave_config;
1875 m_ddev->device_pause = edma_dma_pause;
1876 m_ddev->device_resume = edma_dma_resume;
1877 m_ddev->device_terminate_all = edma_terminate_all;
1878 m_ddev->device_synchronize = edma_synchronize;
1879
1880 m_ddev->src_addr_widths = EDMA_DMA_BUSWIDTHS;
1881 m_ddev->dst_addr_widths = EDMA_DMA_BUSWIDTHS;
1882 m_ddev->directions = BIT(DMA_MEM_TO_MEM);
1883 m_ddev->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
1884
1885 m_ddev->dev = ecc->dev;
1886 INIT_LIST_HEAD(&m_ddev->channels);
1887 } else if (!ecc->legacy_mode) {
1888 dev_info(ecc->dev, "memcpy is disabled\n");
1889 }
1890
1891 for (i = 0; i < ecc->num_channels; i++) {
1892 struct edma_chan *echan = &ecc->slave_chans[i];
1893 echan->ch_num = EDMA_CTLR_CHAN(ecc->id, i);
1894 echan->ecc = ecc;
1895 echan->vchan.desc_free = edma_desc_free;
1896
1897 if (m_ddev && edma_is_memcpy_channel(i, memcpy_channels))
1898 vchan_init(&echan->vchan, m_ddev);
1899 else
1900 vchan_init(&echan->vchan, s_ddev);
1901
1902 INIT_LIST_HEAD(&echan->node);
1903 for (j = 0; j < EDMA_MAX_SLOTS; j++)
1904 echan->slot[j] = -1;
1905 }
1906 }
1907
1908 static int edma_setup_from_hw(struct device *dev, struct edma_soc_info *pdata,
1909 struct edma_cc *ecc)
1910 {
1911 int i;
1912 u32 value, cccfg;
1913 s8 (*queue_priority_map)[2];
1914
1915 /* Decode the eDMA3 configuration from CCCFG register */
1916 cccfg = edma_read(ecc, EDMA_CCCFG);
1917
1918 value = GET_NUM_REGN(cccfg);
1919 ecc->num_region = BIT(value);
1920
1921 value = GET_NUM_DMACH(cccfg);
1922 ecc->num_channels = BIT(value + 1);
1923
1924 value = GET_NUM_QDMACH(cccfg);
1925 ecc->num_qchannels = value * 2;
1926
1927 value = GET_NUM_PAENTRY(cccfg);
1928 ecc->num_slots = BIT(value + 4);
1929
1930 value = GET_NUM_EVQUE(cccfg);
1931 ecc->num_tc = value + 1;
1932
1933 ecc->chmap_exist = (cccfg & CHMAP_EXIST) ? true : false;
1934
1935 dev_dbg(dev, "eDMA3 CC HW configuration (cccfg: 0x%08x):\n", cccfg);
1936 dev_dbg(dev, "num_region: %u\n", ecc->num_region);
1937 dev_dbg(dev, "num_channels: %u\n", ecc->num_channels);
1938 dev_dbg(dev, "num_qchannels: %u\n", ecc->num_qchannels);
1939 dev_dbg(dev, "num_slots: %u\n", ecc->num_slots);
1940 dev_dbg(dev, "num_tc: %u\n", ecc->num_tc);
1941 dev_dbg(dev, "chmap_exist: %s\n", ecc->chmap_exist ? "yes" : "no");
1942
1943 /* Nothing need to be done if queue priority is provided */
1944 if (pdata->queue_priority_mapping)
1945 return 0;
1946
1947 /*
1948 * Configure TC/queue priority as follows:
1949 * Q0 - priority 0
1950 * Q1 - priority 1
1951 * Q2 - priority 2
1952 * ...
1953 * The meaning of priority numbers: 0 highest priority, 7 lowest
1954 * priority. So Q0 is the highest priority queue and the last queue has
1955 * the lowest priority.
1956 */
1957 queue_priority_map = devm_kcalloc(dev, ecc->num_tc + 1, sizeof(s8),
1958 GFP_KERNEL);
1959 if (!queue_priority_map)
1960 return -ENOMEM;
1961
1962 for (i = 0; i < ecc->num_tc; i++) {
1963 queue_priority_map[i][0] = i;
1964 queue_priority_map[i][1] = i;
1965 }
1966 queue_priority_map[i][0] = -1;
1967 queue_priority_map[i][1] = -1;
1968
1969 pdata->queue_priority_mapping = queue_priority_map;
1970 /* Default queue has the lowest priority */
1971 pdata->default_queue = i - 1;
1972
1973 return 0;
1974 }
1975
1976 #if IS_ENABLED(CONFIG_OF)
1977 static int edma_xbar_event_map(struct device *dev, struct edma_soc_info *pdata,
1978 size_t sz)
1979 {
1980 const char pname[] = "ti,edma-xbar-event-map";
1981 struct resource res;
1982 void __iomem *xbar;
1983 s16 (*xbar_chans)[2];
1984 size_t nelm = sz / sizeof(s16);
1985 u32 shift, offset, mux;
1986 int ret, i;
1987
1988 xbar_chans = devm_kcalloc(dev, nelm + 2, sizeof(s16), GFP_KERNEL);
1989 if (!xbar_chans)
1990 return -ENOMEM;
1991
1992 ret = of_address_to_resource(dev->of_node, 1, &res);
1993 if (ret)
1994 return -ENOMEM;
1995
1996 xbar = devm_ioremap(dev, res.start, resource_size(&res));
1997 if (!xbar)
1998 return -ENOMEM;
1999
2000 ret = of_property_read_u16_array(dev->of_node, pname, (u16 *)xbar_chans,
2001 nelm);
2002 if (ret)
2003 return -EIO;
2004
2005 /* Invalidate last entry for the other user of this mess */
2006 nelm >>= 1;
2007 xbar_chans[nelm][0] = -1;
2008 xbar_chans[nelm][1] = -1;
2009
2010 for (i = 0; i < nelm; i++) {
2011 shift = (xbar_chans[i][1] & 0x03) << 3;
2012 offset = xbar_chans[i][1] & 0xfffffffc;
2013 mux = readl(xbar + offset);
2014 mux &= ~(0xff << shift);
2015 mux |= xbar_chans[i][0] << shift;
2016 writel(mux, (xbar + offset));
2017 }
2018
2019 pdata->xbar_chans = (const s16 (*)[2]) xbar_chans;
2020 return 0;
2021 }
2022
2023 static struct edma_soc_info *edma_setup_info_from_dt(struct device *dev,
2024 bool legacy_mode)
2025 {
2026 struct edma_soc_info *info;
2027 struct property *prop;
2028 int sz, ret;
2029
2030 info = devm_kzalloc(dev, sizeof(struct edma_soc_info), GFP_KERNEL);
2031 if (!info)
2032 return ERR_PTR(-ENOMEM);
2033
2034 if (legacy_mode) {
2035 prop = of_find_property(dev->of_node, "ti,edma-xbar-event-map",
2036 &sz);
2037 if (prop) {
2038 ret = edma_xbar_event_map(dev, info, sz);
2039 if (ret)
2040 return ERR_PTR(ret);
2041 }
2042 return info;
2043 }
2044
2045 /* Get the list of channels allocated to be used for memcpy */
2046 prop = of_find_property(dev->of_node, "ti,edma-memcpy-channels", &sz);
2047 if (prop) {
2048 const char pname[] = "ti,edma-memcpy-channels";
2049 size_t nelm = sz / sizeof(s32);
2050 s32 *memcpy_ch;
2051
2052 memcpy_ch = devm_kcalloc(dev, nelm + 1, sizeof(s32),
2053 GFP_KERNEL);
2054 if (!memcpy_ch)
2055 return ERR_PTR(-ENOMEM);
2056
2057 ret = of_property_read_u32_array(dev->of_node, pname,
2058 (u32 *)memcpy_ch, nelm);
2059 if (ret)
2060 return ERR_PTR(ret);
2061
2062 memcpy_ch[nelm] = -1;
2063 info->memcpy_channels = memcpy_ch;
2064 }
2065
2066 prop = of_find_property(dev->of_node, "ti,edma-reserved-slot-ranges",
2067 &sz);
2068 if (prop) {
2069 const char pname[] = "ti,edma-reserved-slot-ranges";
2070 u32 (*tmp)[2];
2071 s16 (*rsv_slots)[2];
2072 size_t nelm = sz / sizeof(*tmp);
2073 struct edma_rsv_info *rsv_info;
2074 int i;
2075
2076 if (!nelm)
2077 return info;
2078
2079 tmp = kcalloc(nelm, sizeof(*tmp), GFP_KERNEL);
2080 if (!tmp)
2081 return ERR_PTR(-ENOMEM);
2082
2083 rsv_info = devm_kzalloc(dev, sizeof(*rsv_info), GFP_KERNEL);
2084 if (!rsv_info) {
2085 kfree(tmp);
2086 return ERR_PTR(-ENOMEM);
2087 }
2088
2089 rsv_slots = devm_kcalloc(dev, nelm + 1, sizeof(*rsv_slots),
2090 GFP_KERNEL);
2091 if (!rsv_slots) {
2092 kfree(tmp);
2093 return ERR_PTR(-ENOMEM);
2094 }
2095
2096 ret = of_property_read_u32_array(dev->of_node, pname,
2097 (u32 *)tmp, nelm * 2);
2098 if (ret) {
2099 kfree(tmp);
2100 return ERR_PTR(ret);
2101 }
2102
2103 for (i = 0; i < nelm; i++) {
2104 rsv_slots[i][0] = tmp[i][0];
2105 rsv_slots[i][1] = tmp[i][1];
2106 }
2107 rsv_slots[nelm][0] = -1;
2108 rsv_slots[nelm][1] = -1;
2109
2110 info->rsv = rsv_info;
2111 info->rsv->rsv_slots = (const s16 (*)[2])rsv_slots;
2112
2113 kfree(tmp);
2114 }
2115
2116 return info;
2117 }
2118
2119 static struct dma_chan *of_edma_xlate(struct of_phandle_args *dma_spec,
2120 struct of_dma *ofdma)
2121 {
2122 struct edma_cc *ecc = ofdma->of_dma_data;
2123 struct dma_chan *chan = NULL;
2124 struct edma_chan *echan;
2125 int i;
2126
2127 if (!ecc || dma_spec->args_count < 1)
2128 return NULL;
2129
2130 for (i = 0; i < ecc->num_channels; i++) {
2131 echan = &ecc->slave_chans[i];
2132 if (echan->ch_num == dma_spec->args[0]) {
2133 chan = &echan->vchan.chan;
2134 break;
2135 }
2136 }
2137
2138 if (!chan)
2139 return NULL;
2140
2141 if (echan->ecc->legacy_mode && dma_spec->args_count == 1)
2142 goto out;
2143
2144 if (!echan->ecc->legacy_mode && dma_spec->args_count == 2 &&
2145 dma_spec->args[1] < echan->ecc->num_tc) {
2146 echan->tc = &echan->ecc->tc_list[dma_spec->args[1]];
2147 goto out;
2148 }
2149
2150 return NULL;
2151 out:
2152 /* The channel is going to be used as HW synchronized */
2153 echan->hw_triggered = true;
2154 return dma_get_slave_channel(chan);
2155 }
2156 #else
2157 static struct edma_soc_info *edma_setup_info_from_dt(struct device *dev,
2158 bool legacy_mode)
2159 {
2160 return ERR_PTR(-EINVAL);
2161 }
2162
2163 static struct dma_chan *of_edma_xlate(struct of_phandle_args *dma_spec,
2164 struct of_dma *ofdma)
2165 {
2166 return NULL;
2167 }
2168 #endif
2169
2170 static int edma_probe(struct platform_device *pdev)
2171 {
2172 struct edma_soc_info *info = pdev->dev.platform_data;
2173 s8 (*queue_priority_mapping)[2];
2174 int i, off, ln;
2175 const s16 (*rsv_slots)[2];
2176 const s16 (*xbar_chans)[2];
2177 int irq;
2178 char *irq_name;
2179 struct resource *mem;
2180 struct device_node *node = pdev->dev.of_node;
2181 struct device *dev = &pdev->dev;
2182 struct edma_cc *ecc;
2183 bool legacy_mode = true;
2184 int ret;
2185
2186 if (node) {
2187 const struct of_device_id *match;
2188
2189 match = of_match_node(edma_of_ids, node);
2190 if (match && (*(u32 *)match->data) == EDMA_BINDING_TPCC)
2191 legacy_mode = false;
2192
2193 info = edma_setup_info_from_dt(dev, legacy_mode);
2194 if (IS_ERR(info)) {
2195 dev_err(dev, "failed to get DT data\n");
2196 return PTR_ERR(info);
2197 }
2198 }
2199
2200 if (!info)
2201 return -ENODEV;
2202
2203 pm_runtime_enable(dev);
2204 ret = pm_runtime_get_sync(dev);
2205 if (ret < 0) {
2206 dev_err(dev, "pm_runtime_get_sync() failed\n");
2207 return ret;
2208 }
2209
2210 ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
2211 if (ret)
2212 return ret;
2213
2214 ecc = devm_kzalloc(dev, sizeof(*ecc), GFP_KERNEL);
2215 if (!ecc)
2216 return -ENOMEM;
2217
2218 ecc->dev = dev;
2219 ecc->id = pdev->id;
2220 ecc->legacy_mode = legacy_mode;
2221 /* When booting with DT the pdev->id is -1 */
2222 if (ecc->id < 0)
2223 ecc->id = 0;
2224
2225 mem = platform_get_resource_byname(pdev, IORESOURCE_MEM, "edma3_cc");
2226 if (!mem) {
2227 dev_dbg(dev, "mem resource not found, using index 0\n");
2228 mem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
2229 if (!mem) {
2230 dev_err(dev, "no mem resource?\n");
2231 return -ENODEV;
2232 }
2233 }
2234 ecc->base = devm_ioremap_resource(dev, mem);
2235 if (IS_ERR(ecc->base))
2236 return PTR_ERR(ecc->base);
2237
2238 platform_set_drvdata(pdev, ecc);
2239
2240 /* Get eDMA3 configuration from IP */
2241 ret = edma_setup_from_hw(dev, info, ecc);
2242 if (ret)
2243 return ret;
2244
2245 /* Allocate memory based on the information we got from the IP */
2246 ecc->slave_chans = devm_kcalloc(dev, ecc->num_channels,
2247 sizeof(*ecc->slave_chans), GFP_KERNEL);
2248 if (!ecc->slave_chans)
2249 return -ENOMEM;
2250
2251 ecc->slot_inuse = devm_kcalloc(dev, BITS_TO_LONGS(ecc->num_slots),
2252 sizeof(unsigned long), GFP_KERNEL);
2253 if (!ecc->slot_inuse)
2254 return -ENOMEM;
2255
2256 ecc->default_queue = info->default_queue;
2257
2258 for (i = 0; i < ecc->num_slots; i++)
2259 edma_write_slot(ecc, i, &dummy_paramset);
2260
2261 if (info->rsv) {
2262 /* Set the reserved slots in inuse list */
2263 rsv_slots = info->rsv->rsv_slots;
2264 if (rsv_slots) {
2265 for (i = 0; rsv_slots[i][0] != -1; i++) {
2266 off = rsv_slots[i][0];
2267 ln = rsv_slots[i][1];
2268 edma_set_bits(off, ln, ecc->slot_inuse);
2269 }
2270 }
2271 }
2272
2273 /* Clear the xbar mapped channels in unused list */
2274 xbar_chans = info->xbar_chans;
2275 if (xbar_chans) {
2276 for (i = 0; xbar_chans[i][1] != -1; i++) {
2277 off = xbar_chans[i][1];
2278 }
2279 }
2280
2281 irq = platform_get_irq_byname(pdev, "edma3_ccint");
2282 if (irq < 0 && node)
2283 irq = irq_of_parse_and_map(node, 0);
2284
2285 if (irq >= 0) {
2286 irq_name = devm_kasprintf(dev, GFP_KERNEL, "%s_ccint",
2287 dev_name(dev));
2288 ret = devm_request_irq(dev, irq, dma_irq_handler, 0, irq_name,
2289 ecc);
2290 if (ret) {
2291 dev_err(dev, "CCINT (%d) failed --> %d\n", irq, ret);
2292 return ret;
2293 }
2294 ecc->ccint = irq;
2295 }
2296
2297 irq = platform_get_irq_byname(pdev, "edma3_ccerrint");
2298 if (irq < 0 && node)
2299 irq = irq_of_parse_and_map(node, 2);
2300
2301 if (irq >= 0) {
2302 irq_name = devm_kasprintf(dev, GFP_KERNEL, "%s_ccerrint",
2303 dev_name(dev));
2304 ret = devm_request_irq(dev, irq, dma_ccerr_handler, 0, irq_name,
2305 ecc);
2306 if (ret) {
2307 dev_err(dev, "CCERRINT (%d) failed --> %d\n", irq, ret);
2308 return ret;
2309 }
2310 ecc->ccerrint = irq;
2311 }
2312
2313 ecc->dummy_slot = edma_alloc_slot(ecc, EDMA_SLOT_ANY);
2314 if (ecc->dummy_slot < 0) {
2315 dev_err(dev, "Can't allocate PaRAM dummy slot\n");
2316 return ecc->dummy_slot;
2317 }
2318
2319 queue_priority_mapping = info->queue_priority_mapping;
2320
2321 if (!ecc->legacy_mode) {
2322 int lowest_priority = 0;
2323 struct of_phandle_args tc_args;
2324
2325 ecc->tc_list = devm_kcalloc(dev, ecc->num_tc,
2326 sizeof(*ecc->tc_list), GFP_KERNEL);
2327 if (!ecc->tc_list)
2328 return -ENOMEM;
2329
2330 for (i = 0;; i++) {
2331 ret = of_parse_phandle_with_fixed_args(node, "ti,tptcs",
2332 1, i, &tc_args);
2333 if (ret || i == ecc->num_tc)
2334 break;
2335
2336 ecc->tc_list[i].node = tc_args.np;
2337 ecc->tc_list[i].id = i;
2338 queue_priority_mapping[i][1] = tc_args.args[0];
2339 if (queue_priority_mapping[i][1] > lowest_priority) {
2340 lowest_priority = queue_priority_mapping[i][1];
2341 info->default_queue = i;
2342 }
2343 }
2344 }
2345
2346 /* Event queue priority mapping */
2347 for (i = 0; queue_priority_mapping[i][0] != -1; i++)
2348 edma_assign_priority_to_queue(ecc, queue_priority_mapping[i][0],
2349 queue_priority_mapping[i][1]);
2350
2351 for (i = 0; i < ecc->num_region; i++) {
2352 edma_write_array2(ecc, EDMA_DRAE, i, 0, 0x0);
2353 edma_write_array2(ecc, EDMA_DRAE, i, 1, 0x0);
2354 edma_write_array(ecc, EDMA_QRAE, i, 0x0);
2355 }
2356 ecc->info = info;
2357
2358 /* Init the dma device and channels */
2359 edma_dma_init(ecc, legacy_mode);
2360
2361 for (i = 0; i < ecc->num_channels; i++) {
2362 /* Assign all channels to the default queue */
2363 edma_assign_channel_eventq(&ecc->slave_chans[i],
2364 info->default_queue);
2365 /* Set entry slot to the dummy slot */
2366 edma_set_chmap(&ecc->slave_chans[i], ecc->dummy_slot);
2367 }
2368
2369 ecc->dma_slave.filter.map = info->slave_map;
2370 ecc->dma_slave.filter.mapcnt = info->slavecnt;
2371 ecc->dma_slave.filter.fn = edma_filter_fn;
2372
2373 ret = dma_async_device_register(&ecc->dma_slave);
2374 if (ret) {
2375 dev_err(dev, "slave ddev registration failed (%d)\n", ret);
2376 goto err_reg1;
2377 }
2378
2379 if (ecc->dma_memcpy) {
2380 ret = dma_async_device_register(ecc->dma_memcpy);
2381 if (ret) {
2382 dev_err(dev, "memcpy ddev registration failed (%d)\n",
2383 ret);
2384 dma_async_device_unregister(&ecc->dma_slave);
2385 goto err_reg1;
2386 }
2387 }
2388
2389 if (node)
2390 of_dma_controller_register(node, of_edma_xlate, ecc);
2391
2392 dev_info(dev, "TI EDMA DMA engine driver\n");
2393
2394 return 0;
2395
2396 err_reg1:
2397 edma_free_slot(ecc, ecc->dummy_slot);
2398 return ret;
2399 }
2400
2401 static void edma_cleanupp_vchan(struct dma_device *dmadev)
2402 {
2403 struct edma_chan *echan, *_echan;
2404
2405 list_for_each_entry_safe(echan, _echan,
2406 &dmadev->channels, vchan.chan.device_node) {
2407 list_del(&echan->vchan.chan.device_node);
2408 tasklet_kill(&echan->vchan.task);
2409 }
2410 }
2411
2412 static int edma_remove(struct platform_device *pdev)
2413 {
2414 struct device *dev = &pdev->dev;
2415 struct edma_cc *ecc = dev_get_drvdata(dev);
2416
2417 devm_free_irq(dev, ecc->ccint, ecc);
2418 devm_free_irq(dev, ecc->ccerrint, ecc);
2419
2420 edma_cleanupp_vchan(&ecc->dma_slave);
2421
2422 if (dev->of_node)
2423 of_dma_controller_free(dev->of_node);
2424 dma_async_device_unregister(&ecc->dma_slave);
2425 if (ecc->dma_memcpy)
2426 dma_async_device_unregister(ecc->dma_memcpy);
2427 edma_free_slot(ecc, ecc->dummy_slot);
2428
2429 return 0;
2430 }
2431
2432 #ifdef CONFIG_PM_SLEEP
2433 static int edma_pm_suspend(struct device *dev)
2434 {
2435 struct edma_cc *ecc = dev_get_drvdata(dev);
2436 struct edma_chan *echan = ecc->slave_chans;
2437 int i;
2438
2439 for (i = 0; i < ecc->num_channels; i++) {
2440 if (echan[i].alloced)
2441 edma_setup_interrupt(&echan[i], false);
2442 }
2443
2444 return 0;
2445 }
2446
2447 static int edma_pm_resume(struct device *dev)
2448 {
2449 struct edma_cc *ecc = dev_get_drvdata(dev);
2450 struct edma_chan *echan = ecc->slave_chans;
2451 int i;
2452 s8 (*queue_priority_mapping)[2];
2453
2454 /* re initialize dummy slot to dummy param set */
2455 edma_write_slot(ecc, ecc->dummy_slot, &dummy_paramset);
2456
2457 queue_priority_mapping = ecc->info->queue_priority_mapping;
2458
2459 /* Event queue priority mapping */
2460 for (i = 0; queue_priority_mapping[i][0] != -1; i++)
2461 edma_assign_priority_to_queue(ecc, queue_priority_mapping[i][0],
2462 queue_priority_mapping[i][1]);
2463
2464 for (i = 0; i < ecc->num_channels; i++) {
2465 if (echan[i].alloced) {
2466 /* ensure access through shadow region 0 */
2467 edma_or_array2(ecc, EDMA_DRAE, 0, i >> 5,
2468 BIT(i & 0x1f));
2469
2470 edma_setup_interrupt(&echan[i], true);
2471
2472 /* Set up channel -> slot mapping for the entry slot */
2473 edma_set_chmap(&echan[i], echan[i].slot[0]);
2474 }
2475 }
2476
2477 return 0;
2478 }
2479 #endif
2480
2481 static const struct dev_pm_ops edma_pm_ops = {
2482 SET_LATE_SYSTEM_SLEEP_PM_OPS(edma_pm_suspend, edma_pm_resume)
2483 };
2484
2485 static struct platform_driver edma_driver = {
2486 .probe = edma_probe,
2487 .remove = edma_remove,
2488 .driver = {
2489 .name = "edma",
2490 .pm = &edma_pm_ops,
2491 .of_match_table = edma_of_ids,
2492 },
2493 };
2494
2495 static int edma_tptc_probe(struct platform_device *pdev)
2496 {
2497 pm_runtime_enable(&pdev->dev);
2498 return pm_runtime_get_sync(&pdev->dev);
2499 }
2500
2501 static struct platform_driver edma_tptc_driver = {
2502 .probe = edma_tptc_probe,
2503 .driver = {
2504 .name = "edma3-tptc",
2505 .of_match_table = edma_tptc_of_ids,
2506 },
2507 };
2508
2509 bool edma_filter_fn(struct dma_chan *chan, void *param)
2510 {
2511 bool match = false;
2512
2513 if (chan->device->dev->driver == &edma_driver.driver) {
2514 struct edma_chan *echan = to_edma_chan(chan);
2515 unsigned ch_req = *(unsigned *)param;
2516 if (ch_req == echan->ch_num) {
2517 /* The channel is going to be used as HW synchronized */
2518 echan->hw_triggered = true;
2519 match = true;
2520 }
2521 }
2522 return match;
2523 }
2524 EXPORT_SYMBOL(edma_filter_fn);
2525
2526 static int edma_init(void)
2527 {
2528 int ret;
2529
2530 ret = platform_driver_register(&edma_tptc_driver);
2531 if (ret)
2532 return ret;
2533
2534 return platform_driver_register(&edma_driver);
2535 }
2536 subsys_initcall(edma_init);
2537
2538 static void __exit edma_exit(void)
2539 {
2540 platform_driver_unregister(&edma_driver);
2541 platform_driver_unregister(&edma_tptc_driver);
2542 }
2543 module_exit(edma_exit);
2544
2545 MODULE_AUTHOR("Matt Porter <matt.porter@linaro.org>");
2546 MODULE_DESCRIPTION("TI EDMA DMA engine driver");
2547 MODULE_LICENSE("GPL v2");
Linux with added FPC-III support