Loading...
1/*
2 * Copyright (C) Ericsson AB 2007-2008
3 * Copyright (C) ST-Ericsson SA 2008-2010
4 * Author: Per Forlin <per.forlin@stericsson.com> for ST-Ericsson
5 * Author: Jonas Aaberg <jonas.aberg@stericsson.com> for ST-Ericsson
6 * License terms: GNU General Public License (GPL) version 2
7 */
8
9#include <linux/dma-mapping.h>
10#include <linux/kernel.h>
11#include <linux/slab.h>
12#include <linux/dmaengine.h>
13#include <linux/platform_device.h>
14#include <linux/clk.h>
15#include <linux/delay.h>
16#include <linux/err.h>
17#include <linux/amba/bus.h>
18
19#include <plat/ste_dma40.h>
20
21#include "ste_dma40_ll.h"
22
23#define D40_NAME "dma40"
24
25#define D40_PHY_CHAN -1
26
27/* For masking out/in 2 bit channel positions */
28#define D40_CHAN_POS(chan) (2 * (chan / 2))
29#define D40_CHAN_POS_MASK(chan) (0x3 << D40_CHAN_POS(chan))
30
31/* Maximum iterations taken before giving up suspending a channel */
32#define D40_SUSPEND_MAX_IT 500
33
34/* Hardware requirement on LCLA alignment */
35#define LCLA_ALIGNMENT 0x40000
36
37/* Max number of links per event group */
38#define D40_LCLA_LINK_PER_EVENT_GRP 128
39#define D40_LCLA_END D40_LCLA_LINK_PER_EVENT_GRP
40
41/* Attempts before giving up to trying to get pages that are aligned */
42#define MAX_LCLA_ALLOC_ATTEMPTS 256
43
44/* Bit markings for allocation map */
45#define D40_ALLOC_FREE (1 << 31)
46#define D40_ALLOC_PHY (1 << 30)
47#define D40_ALLOC_LOG_FREE 0
48
49/**
50 * enum 40_command - The different commands and/or statuses.
51 *
52 * @D40_DMA_STOP: DMA channel command STOP or status STOPPED,
53 * @D40_DMA_RUN: The DMA channel is RUNNING of the command RUN.
54 * @D40_DMA_SUSPEND_REQ: Request the DMA to SUSPEND as soon as possible.
55 * @D40_DMA_SUSPENDED: The DMA channel is SUSPENDED.
56 */
57enum d40_command {
58 D40_DMA_STOP = 0,
59 D40_DMA_RUN = 1,
60 D40_DMA_SUSPEND_REQ = 2,
61 D40_DMA_SUSPENDED = 3
62};
63
64/**
65 * struct d40_lli_pool - Structure for keeping LLIs in memory
66 *
67 * @base: Pointer to memory area when the pre_alloc_lli's are not large
68 * enough, IE bigger than the most common case, 1 dst and 1 src. NULL if
69 * pre_alloc_lli is used.
70 * @dma_addr: DMA address, if mapped
71 * @size: The size in bytes of the memory at base or the size of pre_alloc_lli.
72 * @pre_alloc_lli: Pre allocated area for the most common case of transfers,
73 * one buffer to one buffer.
74 */
75struct d40_lli_pool {
76 void *base;
77 int size;
78 dma_addr_t dma_addr;
79 /* Space for dst and src, plus an extra for padding */
80 u8 pre_alloc_lli[3 * sizeof(struct d40_phy_lli)];
81};
82
83/**
84 * struct d40_desc - A descriptor is one DMA job.
85 *
86 * @lli_phy: LLI settings for physical channel. Both src and dst=
87 * points into the lli_pool, to base if lli_len > 1 or to pre_alloc_lli if
88 * lli_len equals one.
89 * @lli_log: Same as above but for logical channels.
90 * @lli_pool: The pool with two entries pre-allocated.
91 * @lli_len: Number of llis of current descriptor.
92 * @lli_current: Number of transferred llis.
93 * @lcla_alloc: Number of LCLA entries allocated.
94 * @txd: DMA engine struct. Used for among other things for communication
95 * during a transfer.
96 * @node: List entry.
97 * @is_in_client_list: true if the client owns this descriptor.
98 * the previous one.
99 *
100 * This descriptor is used for both logical and physical transfers.
101 */
102struct d40_desc {
103 /* LLI physical */
104 struct d40_phy_lli_bidir lli_phy;
105 /* LLI logical */
106 struct d40_log_lli_bidir lli_log;
107
108 struct d40_lli_pool lli_pool;
109 int lli_len;
110 int lli_current;
111 int lcla_alloc;
112
113 struct dma_async_tx_descriptor txd;
114 struct list_head node;
115
116 bool is_in_client_list;
117 bool cyclic;
118};
119
120/**
121 * struct d40_lcla_pool - LCLA pool settings and data.
122 *
123 * @base: The virtual address of LCLA. 18 bit aligned.
124 * @base_unaligned: The orignal kmalloc pointer, if kmalloc is used.
125 * This pointer is only there for clean-up on error.
126 * @pages: The number of pages needed for all physical channels.
127 * Only used later for clean-up on error
128 * @lock: Lock to protect the content in this struct.
129 * @alloc_map: big map over which LCLA entry is own by which job.
130 */
131struct d40_lcla_pool {
132 void *base;
133 dma_addr_t dma_addr;
134 void *base_unaligned;
135 int pages;
136 spinlock_t lock;
137 struct d40_desc **alloc_map;
138};
139
140/**
141 * struct d40_phy_res - struct for handling eventlines mapped to physical
142 * channels.
143 *
144 * @lock: A lock protection this entity.
145 * @num: The physical channel number of this entity.
146 * @allocated_src: Bit mapped to show which src event line's are mapped to
147 * this physical channel. Can also be free or physically allocated.
148 * @allocated_dst: Same as for src but is dst.
149 * allocated_dst and allocated_src uses the D40_ALLOC* defines as well as
150 * event line number.
151 */
152struct d40_phy_res {
153 spinlock_t lock;
154 int num;
155 u32 allocated_src;
156 u32 allocated_dst;
157};
158
159struct d40_base;
160
161/**
162 * struct d40_chan - Struct that describes a channel.
163 *
164 * @lock: A spinlock to protect this struct.
165 * @log_num: The logical number, if any of this channel.
166 * @completed: Starts with 1, after first interrupt it is set to dma engine's
167 * current cookie.
168 * @pending_tx: The number of pending transfers. Used between interrupt handler
169 * and tasklet.
170 * @busy: Set to true when transfer is ongoing on this channel.
171 * @phy_chan: Pointer to physical channel which this instance runs on. If this
172 * point is NULL, then the channel is not allocated.
173 * @chan: DMA engine handle.
174 * @tasklet: Tasklet that gets scheduled from interrupt context to complete a
175 * transfer and call client callback.
176 * @client: Cliented owned descriptor list.
177 * @pending_queue: Submitted jobs, to be issued by issue_pending()
178 * @active: Active descriptor.
179 * @queue: Queued jobs.
180 * @prepare_queue: Prepared jobs.
181 * @dma_cfg: The client configuration of this dma channel.
182 * @configured: whether the dma_cfg configuration is valid
183 * @base: Pointer to the device instance struct.
184 * @src_def_cfg: Default cfg register setting for src.
185 * @dst_def_cfg: Default cfg register setting for dst.
186 * @log_def: Default logical channel settings.
187 * @lcla: Space for one dst src pair for logical channel transfers.
188 * @lcpa: Pointer to dst and src lcpa settings.
189 * @runtime_addr: runtime configured address.
190 * @runtime_direction: runtime configured direction.
191 *
192 * This struct can either "be" a logical or a physical channel.
193 */
194struct d40_chan {
195 spinlock_t lock;
196 int log_num;
197 /* ID of the most recent completed transfer */
198 int completed;
199 int pending_tx;
200 bool busy;
201 struct d40_phy_res *phy_chan;
202 struct dma_chan chan;
203 struct tasklet_struct tasklet;
204 struct list_head client;
205 struct list_head pending_queue;
206 struct list_head active;
207 struct list_head queue;
208 struct list_head prepare_queue;
209 struct stedma40_chan_cfg dma_cfg;
210 bool configured;
211 struct d40_base *base;
212 /* Default register configurations */
213 u32 src_def_cfg;
214 u32 dst_def_cfg;
215 struct d40_def_lcsp log_def;
216 struct d40_log_lli_full *lcpa;
217 /* Runtime reconfiguration */
218 dma_addr_t runtime_addr;
219 enum dma_data_direction runtime_direction;
220};
221
222/**
223 * struct d40_base - The big global struct, one for each probe'd instance.
224 *
225 * @interrupt_lock: Lock used to make sure one interrupt is handle a time.
226 * @execmd_lock: Lock for execute command usage since several channels share
227 * the same physical register.
228 * @dev: The device structure.
229 * @virtbase: The virtual base address of the DMA's register.
230 * @rev: silicon revision detected.
231 * @clk: Pointer to the DMA clock structure.
232 * @phy_start: Physical memory start of the DMA registers.
233 * @phy_size: Size of the DMA register map.
234 * @irq: The IRQ number.
235 * @num_phy_chans: The number of physical channels. Read from HW. This
236 * is the number of available channels for this driver, not counting "Secure
237 * mode" allocated physical channels.
238 * @num_log_chans: The number of logical channels. Calculated from
239 * num_phy_chans.
240 * @dma_both: dma_device channels that can do both memcpy and slave transfers.
241 * @dma_slave: dma_device channels that can do only do slave transfers.
242 * @dma_memcpy: dma_device channels that can do only do memcpy transfers.
243 * @log_chans: Room for all possible logical channels in system.
244 * @lookup_log_chans: Used to map interrupt number to logical channel. Points
245 * to log_chans entries.
246 * @lookup_phy_chans: Used to map interrupt number to physical channel. Points
247 * to phy_chans entries.
248 * @plat_data: Pointer to provided platform_data which is the driver
249 * configuration.
250 * @phy_res: Vector containing all physical channels.
251 * @lcla_pool: lcla pool settings and data.
252 * @lcpa_base: The virtual mapped address of LCPA.
253 * @phy_lcpa: The physical address of the LCPA.
254 * @lcpa_size: The size of the LCPA area.
255 * @desc_slab: cache for descriptors.
256 */
257struct d40_base {
258 spinlock_t interrupt_lock;
259 spinlock_t execmd_lock;
260 struct device *dev;
261 void __iomem *virtbase;
262 u8 rev:4;
263 struct clk *clk;
264 phys_addr_t phy_start;
265 resource_size_t phy_size;
266 int irq;
267 int num_phy_chans;
268 int num_log_chans;
269 struct dma_device dma_both;
270 struct dma_device dma_slave;
271 struct dma_device dma_memcpy;
272 struct d40_chan *phy_chans;
273 struct d40_chan *log_chans;
274 struct d40_chan **lookup_log_chans;
275 struct d40_chan **lookup_phy_chans;
276 struct stedma40_platform_data *plat_data;
277 /* Physical half channels */
278 struct d40_phy_res *phy_res;
279 struct d40_lcla_pool lcla_pool;
280 void *lcpa_base;
281 dma_addr_t phy_lcpa;
282 resource_size_t lcpa_size;
283 struct kmem_cache *desc_slab;
284};
285
286/**
287 * struct d40_interrupt_lookup - lookup table for interrupt handler
288 *
289 * @src: Interrupt mask register.
290 * @clr: Interrupt clear register.
291 * @is_error: true if this is an error interrupt.
292 * @offset: start delta in the lookup_log_chans in d40_base. If equals to
293 * D40_PHY_CHAN, the lookup_phy_chans shall be used instead.
294 */
295struct d40_interrupt_lookup {
296 u32 src;
297 u32 clr;
298 bool is_error;
299 int offset;
300};
301
302/**
303 * struct d40_reg_val - simple lookup struct
304 *
305 * @reg: The register.
306 * @val: The value that belongs to the register in reg.
307 */
308struct d40_reg_val {
309 unsigned int reg;
310 unsigned int val;
311};
312
313static struct device *chan2dev(struct d40_chan *d40c)
314{
315 return &d40c->chan.dev->device;
316}
317
318static bool chan_is_physical(struct d40_chan *chan)
319{
320 return chan->log_num == D40_PHY_CHAN;
321}
322
323static bool chan_is_logical(struct d40_chan *chan)
324{
325 return !chan_is_physical(chan);
326}
327
328static void __iomem *chan_base(struct d40_chan *chan)
329{
330 return chan->base->virtbase + D40_DREG_PCBASE +
331 chan->phy_chan->num * D40_DREG_PCDELTA;
332}
333
334#define d40_err(dev, format, arg...) \
335 dev_err(dev, "[%s] " format, __func__, ## arg)
336
337#define chan_err(d40c, format, arg...) \
338 d40_err(chan2dev(d40c), format, ## arg)
339
340static int d40_pool_lli_alloc(struct d40_chan *d40c, struct d40_desc *d40d,
341 int lli_len)
342{
343 bool is_log = chan_is_logical(d40c);
344 u32 align;
345 void *base;
346
347 if (is_log)
348 align = sizeof(struct d40_log_lli);
349 else
350 align = sizeof(struct d40_phy_lli);
351
352 if (lli_len == 1) {
353 base = d40d->lli_pool.pre_alloc_lli;
354 d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
355 d40d->lli_pool.base = NULL;
356 } else {
357 d40d->lli_pool.size = lli_len * 2 * align;
358
359 base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
360 d40d->lli_pool.base = base;
361
362 if (d40d->lli_pool.base == NULL)
363 return -ENOMEM;
364 }
365
366 if (is_log) {
367 d40d->lli_log.src = PTR_ALIGN(base, align);
368 d40d->lli_log.dst = d40d->lli_log.src + lli_len;
369
370 d40d->lli_pool.dma_addr = 0;
371 } else {
372 d40d->lli_phy.src = PTR_ALIGN(base, align);
373 d40d->lli_phy.dst = d40d->lli_phy.src + lli_len;
374
375 d40d->lli_pool.dma_addr = dma_map_single(d40c->base->dev,
376 d40d->lli_phy.src,
377 d40d->lli_pool.size,
378 DMA_TO_DEVICE);
379
380 if (dma_mapping_error(d40c->base->dev,
381 d40d->lli_pool.dma_addr)) {
382 kfree(d40d->lli_pool.base);
383 d40d->lli_pool.base = NULL;
384 d40d->lli_pool.dma_addr = 0;
385 return -ENOMEM;
386 }
387 }
388
389 return 0;
390}
391
392static void d40_pool_lli_free(struct d40_chan *d40c, struct d40_desc *d40d)
393{
394 if (d40d->lli_pool.dma_addr)
395 dma_unmap_single(d40c->base->dev, d40d->lli_pool.dma_addr,
396 d40d->lli_pool.size, DMA_TO_DEVICE);
397
398 kfree(d40d->lli_pool.base);
399 d40d->lli_pool.base = NULL;
400 d40d->lli_pool.size = 0;
401 d40d->lli_log.src = NULL;
402 d40d->lli_log.dst = NULL;
403 d40d->lli_phy.src = NULL;
404 d40d->lli_phy.dst = NULL;
405}
406
407static int d40_lcla_alloc_one(struct d40_chan *d40c,
408 struct d40_desc *d40d)
409{
410 unsigned long flags;
411 int i;
412 int ret = -EINVAL;
413 int p;
414
415 spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
416
417 p = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP;
418
419 /*
420 * Allocate both src and dst at the same time, therefore the half
421 * start on 1 since 0 can't be used since zero is used as end marker.
422 */
423 for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
424 if (!d40c->base->lcla_pool.alloc_map[p + i]) {
425 d40c->base->lcla_pool.alloc_map[p + i] = d40d;
426 d40d->lcla_alloc++;
427 ret = i;
428 break;
429 }
430 }
431
432 spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
433
434 return ret;
435}
436
437static int d40_lcla_free_all(struct d40_chan *d40c,
438 struct d40_desc *d40d)
439{
440 unsigned long flags;
441 int i;
442 int ret = -EINVAL;
443
444 if (chan_is_physical(d40c))
445 return 0;
446
447 spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
448
449 for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
450 if (d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num *
451 D40_LCLA_LINK_PER_EVENT_GRP + i] == d40d) {
452 d40c->base->lcla_pool.alloc_map[d40c->phy_chan->num *
453 D40_LCLA_LINK_PER_EVENT_GRP + i] = NULL;
454 d40d->lcla_alloc--;
455 if (d40d->lcla_alloc == 0) {
456 ret = 0;
457 break;
458 }
459 }
460 }
461
462 spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
463
464 return ret;
465
466}
467
468static void d40_desc_remove(struct d40_desc *d40d)
469{
470 list_del(&d40d->node);
471}
472
473static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
474{
475 struct d40_desc *desc = NULL;
476
477 if (!list_empty(&d40c->client)) {
478 struct d40_desc *d;
479 struct d40_desc *_d;
480
481 list_for_each_entry_safe(d, _d, &d40c->client, node)
482 if (async_tx_test_ack(&d->txd)) {
483 d40_desc_remove(d);
484 desc = d;
485 memset(desc, 0, sizeof(*desc));
486 break;
487 }
488 }
489
490 if (!desc)
491 desc = kmem_cache_zalloc(d40c->base->desc_slab, GFP_NOWAIT);
492
493 if (desc)
494 INIT_LIST_HEAD(&desc->node);
495
496 return desc;
497}
498
499static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
500{
501
502 d40_pool_lli_free(d40c, d40d);
503 d40_lcla_free_all(d40c, d40d);
504 kmem_cache_free(d40c->base->desc_slab, d40d);
505}
506
507static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
508{
509 list_add_tail(&desc->node, &d40c->active);
510}
511
512static void d40_phy_lli_load(struct d40_chan *chan, struct d40_desc *desc)
513{
514 struct d40_phy_lli *lli_dst = desc->lli_phy.dst;
515 struct d40_phy_lli *lli_src = desc->lli_phy.src;
516 void __iomem *base = chan_base(chan);
517
518 writel(lli_src->reg_cfg, base + D40_CHAN_REG_SSCFG);
519 writel(lli_src->reg_elt, base + D40_CHAN_REG_SSELT);
520 writel(lli_src->reg_ptr, base + D40_CHAN_REG_SSPTR);
521 writel(lli_src->reg_lnk, base + D40_CHAN_REG_SSLNK);
522
523 writel(lli_dst->reg_cfg, base + D40_CHAN_REG_SDCFG);
524 writel(lli_dst->reg_elt, base + D40_CHAN_REG_SDELT);
525 writel(lli_dst->reg_ptr, base + D40_CHAN_REG_SDPTR);
526 writel(lli_dst->reg_lnk, base + D40_CHAN_REG_SDLNK);
527}
528
529static void d40_log_lli_to_lcxa(struct d40_chan *chan, struct d40_desc *desc)
530{
531 struct d40_lcla_pool *pool = &chan->base->lcla_pool;
532 struct d40_log_lli_bidir *lli = &desc->lli_log;
533 int lli_current = desc->lli_current;
534 int lli_len = desc->lli_len;
535 bool cyclic = desc->cyclic;
536 int curr_lcla = -EINVAL;
537 int first_lcla = 0;
538 bool linkback;
539
540 /*
541 * We may have partially running cyclic transfers, in case we did't get
542 * enough LCLA entries.
543 */
544 linkback = cyclic && lli_current == 0;
545
546 /*
547 * For linkback, we need one LCLA even with only one link, because we
548 * can't link back to the one in LCPA space
549 */
550 if (linkback || (lli_len - lli_current > 1)) {
551 curr_lcla = d40_lcla_alloc_one(chan, desc);
552 first_lcla = curr_lcla;
553 }
554
555 /*
556 * For linkback, we normally load the LCPA in the loop since we need to
557 * link it to the second LCLA and not the first. However, if we
558 * couldn't even get a first LCLA, then we have to run in LCPA and
559 * reload manually.
560 */
561 if (!linkback || curr_lcla == -EINVAL) {
562 unsigned int flags = 0;
563
564 if (curr_lcla == -EINVAL)
565 flags |= LLI_TERM_INT;
566
567 d40_log_lli_lcpa_write(chan->lcpa,
568 &lli->dst[lli_current],
569 &lli->src[lli_current],
570 curr_lcla,
571 flags);
572 lli_current++;
573 }
574
575 if (curr_lcla < 0)
576 goto out;
577
578 for (; lli_current < lli_len; lli_current++) {
579 unsigned int lcla_offset = chan->phy_chan->num * 1024 +
580 8 * curr_lcla * 2;
581 struct d40_log_lli *lcla = pool->base + lcla_offset;
582 unsigned int flags = 0;
583 int next_lcla;
584
585 if (lli_current + 1 < lli_len)
586 next_lcla = d40_lcla_alloc_one(chan, desc);
587 else
588 next_lcla = linkback ? first_lcla : -EINVAL;
589
590 if (cyclic || next_lcla == -EINVAL)
591 flags |= LLI_TERM_INT;
592
593 if (linkback && curr_lcla == first_lcla) {
594 /* First link goes in both LCPA and LCLA */
595 d40_log_lli_lcpa_write(chan->lcpa,
596 &lli->dst[lli_current],
597 &lli->src[lli_current],
598 next_lcla, flags);
599 }
600
601 /*
602 * One unused LCLA in the cyclic case if the very first
603 * next_lcla fails...
604 */
605 d40_log_lli_lcla_write(lcla,
606 &lli->dst[lli_current],
607 &lli->src[lli_current],
608 next_lcla, flags);
609
610 dma_sync_single_range_for_device(chan->base->dev,
611 pool->dma_addr, lcla_offset,
612 2 * sizeof(struct d40_log_lli),
613 DMA_TO_DEVICE);
614
615 curr_lcla = next_lcla;
616
617 if (curr_lcla == -EINVAL || curr_lcla == first_lcla) {
618 lli_current++;
619 break;
620 }
621 }
622
623out:
624 desc->lli_current = lli_current;
625}
626
627static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
628{
629 if (chan_is_physical(d40c)) {
630 d40_phy_lli_load(d40c, d40d);
631 d40d->lli_current = d40d->lli_len;
632 } else
633 d40_log_lli_to_lcxa(d40c, d40d);
634}
635
636static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
637{
638 struct d40_desc *d;
639
640 if (list_empty(&d40c->active))
641 return NULL;
642
643 d = list_first_entry(&d40c->active,
644 struct d40_desc,
645 node);
646 return d;
647}
648
649/* remove desc from current queue and add it to the pending_queue */
650static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
651{
652 d40_desc_remove(desc);
653 desc->is_in_client_list = false;
654 list_add_tail(&desc->node, &d40c->pending_queue);
655}
656
657static struct d40_desc *d40_first_pending(struct d40_chan *d40c)
658{
659 struct d40_desc *d;
660
661 if (list_empty(&d40c->pending_queue))
662 return NULL;
663
664 d = list_first_entry(&d40c->pending_queue,
665 struct d40_desc,
666 node);
667 return d;
668}
669
670static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
671{
672 struct d40_desc *d;
673
674 if (list_empty(&d40c->queue))
675 return NULL;
676
677 d = list_first_entry(&d40c->queue,
678 struct d40_desc,
679 node);
680 return d;
681}
682
683static int d40_psize_2_burst_size(bool is_log, int psize)
684{
685 if (is_log) {
686 if (psize == STEDMA40_PSIZE_LOG_1)
687 return 1;
688 } else {
689 if (psize == STEDMA40_PSIZE_PHY_1)
690 return 1;
691 }
692
693 return 2 << psize;
694}
695
696/*
697 * The dma only supports transmitting packages up to
698 * STEDMA40_MAX_SEG_SIZE << data_width. Calculate the total number of
699 * dma elements required to send the entire sg list
700 */
701static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2)
702{
703 int dmalen;
704 u32 max_w = max(data_width1, data_width2);
705 u32 min_w = min(data_width1, data_width2);
706 u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE << min_w, 1 << max_w);
707
708 if (seg_max > STEDMA40_MAX_SEG_SIZE)
709 seg_max -= (1 << max_w);
710
711 if (!IS_ALIGNED(size, 1 << max_w))
712 return -EINVAL;
713
714 if (size <= seg_max)
715 dmalen = 1;
716 else {
717 dmalen = size / seg_max;
718 if (dmalen * seg_max < size)
719 dmalen++;
720 }
721 return dmalen;
722}
723
724static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len,
725 u32 data_width1, u32 data_width2)
726{
727 struct scatterlist *sg;
728 int i;
729 int len = 0;
730 int ret;
731
732 for_each_sg(sgl, sg, sg_len, i) {
733 ret = d40_size_2_dmalen(sg_dma_len(sg),
734 data_width1, data_width2);
735 if (ret < 0)
736 return ret;
737 len += ret;
738 }
739 return len;
740}
741
742/* Support functions for logical channels */
743
744static int d40_channel_execute_command(struct d40_chan *d40c,
745 enum d40_command command)
746{
747 u32 status;
748 int i;
749 void __iomem *active_reg;
750 int ret = 0;
751 unsigned long flags;
752 u32 wmask;
753
754 spin_lock_irqsave(&d40c->base->execmd_lock, flags);
755
756 if (d40c->phy_chan->num % 2 == 0)
757 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
758 else
759 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
760
761 if (command == D40_DMA_SUSPEND_REQ) {
762 status = (readl(active_reg) &
763 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
764 D40_CHAN_POS(d40c->phy_chan->num);
765
766 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
767 goto done;
768 }
769
770 wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
771 writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
772 active_reg);
773
774 if (command == D40_DMA_SUSPEND_REQ) {
775
776 for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
777 status = (readl(active_reg) &
778 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
779 D40_CHAN_POS(d40c->phy_chan->num);
780
781 cpu_relax();
782 /*
783 * Reduce the number of bus accesses while
784 * waiting for the DMA to suspend.
785 */
786 udelay(3);
787
788 if (status == D40_DMA_STOP ||
789 status == D40_DMA_SUSPENDED)
790 break;
791 }
792
793 if (i == D40_SUSPEND_MAX_IT) {
794 chan_err(d40c,
795 "unable to suspend the chl %d (log: %d) status %x\n",
796 d40c->phy_chan->num, d40c->log_num,
797 status);
798 dump_stack();
799 ret = -EBUSY;
800 }
801
802 }
803done:
804 spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
805 return ret;
806}
807
808static void d40_term_all(struct d40_chan *d40c)
809{
810 struct d40_desc *d40d;
811 struct d40_desc *_d;
812
813 /* Release active descriptors */
814 while ((d40d = d40_first_active_get(d40c))) {
815 d40_desc_remove(d40d);
816 d40_desc_free(d40c, d40d);
817 }
818
819 /* Release queued descriptors waiting for transfer */
820 while ((d40d = d40_first_queued(d40c))) {
821 d40_desc_remove(d40d);
822 d40_desc_free(d40c, d40d);
823 }
824
825 /* Release pending descriptors */
826 while ((d40d = d40_first_pending(d40c))) {
827 d40_desc_remove(d40d);
828 d40_desc_free(d40c, d40d);
829 }
830
831 /* Release client owned descriptors */
832 if (!list_empty(&d40c->client))
833 list_for_each_entry_safe(d40d, _d, &d40c->client, node) {
834 d40_desc_remove(d40d);
835 d40_desc_free(d40c, d40d);
836 }
837
838 /* Release descriptors in prepare queue */
839 if (!list_empty(&d40c->prepare_queue))
840 list_for_each_entry_safe(d40d, _d,
841 &d40c->prepare_queue, node) {
842 d40_desc_remove(d40d);
843 d40_desc_free(d40c, d40d);
844 }
845
846 d40c->pending_tx = 0;
847 d40c->busy = false;
848}
849
850static void __d40_config_set_event(struct d40_chan *d40c, bool enable,
851 u32 event, int reg)
852{
853 void __iomem *addr = chan_base(d40c) + reg;
854 int tries;
855
856 if (!enable) {
857 writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
858 | ~D40_EVENTLINE_MASK(event), addr);
859 return;
860 }
861
862 /*
863 * The hardware sometimes doesn't register the enable when src and dst
864 * event lines are active on the same logical channel. Retry to ensure
865 * it does. Usually only one retry is sufficient.
866 */
867 tries = 100;
868 while (--tries) {
869 writel((D40_ACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
870 | ~D40_EVENTLINE_MASK(event), addr);
871
872 if (readl(addr) & D40_EVENTLINE_MASK(event))
873 break;
874 }
875
876 if (tries != 99)
877 dev_dbg(chan2dev(d40c),
878 "[%s] workaround enable S%cLNK (%d tries)\n",
879 __func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D',
880 100 - tries);
881
882 WARN_ON(!tries);
883}
884
885static void d40_config_set_event(struct d40_chan *d40c, bool do_enable)
886{
887 unsigned long flags;
888
889 spin_lock_irqsave(&d40c->phy_chan->lock, flags);
890
891 /* Enable event line connected to device (or memcpy) */
892 if ((d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) ||
893 (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH)) {
894 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
895
896 __d40_config_set_event(d40c, do_enable, event,
897 D40_CHAN_REG_SSLNK);
898 }
899
900 if (d40c->dma_cfg.dir != STEDMA40_PERIPH_TO_MEM) {
901 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
902
903 __d40_config_set_event(d40c, do_enable, event,
904 D40_CHAN_REG_SDLNK);
905 }
906
907 spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
908}
909
910static u32 d40_chan_has_events(struct d40_chan *d40c)
911{
912 void __iomem *chanbase = chan_base(d40c);
913 u32 val;
914
915 val = readl(chanbase + D40_CHAN_REG_SSLNK);
916 val |= readl(chanbase + D40_CHAN_REG_SDLNK);
917
918 return val;
919}
920
921static u32 d40_get_prmo(struct d40_chan *d40c)
922{
923 static const unsigned int phy_map[] = {
924 [STEDMA40_PCHAN_BASIC_MODE]
925 = D40_DREG_PRMO_PCHAN_BASIC,
926 [STEDMA40_PCHAN_MODULO_MODE]
927 = D40_DREG_PRMO_PCHAN_MODULO,
928 [STEDMA40_PCHAN_DOUBLE_DST_MODE]
929 = D40_DREG_PRMO_PCHAN_DOUBLE_DST,
930 };
931 static const unsigned int log_map[] = {
932 [STEDMA40_LCHAN_SRC_PHY_DST_LOG]
933 = D40_DREG_PRMO_LCHAN_SRC_PHY_DST_LOG,
934 [STEDMA40_LCHAN_SRC_LOG_DST_PHY]
935 = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_PHY,
936 [STEDMA40_LCHAN_SRC_LOG_DST_LOG]
937 = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_LOG,
938 };
939
940 if (chan_is_physical(d40c))
941 return phy_map[d40c->dma_cfg.mode_opt];
942 else
943 return log_map[d40c->dma_cfg.mode_opt];
944}
945
946static void d40_config_write(struct d40_chan *d40c)
947{
948 u32 addr_base;
949 u32 var;
950
951 /* Odd addresses are even addresses + 4 */
952 addr_base = (d40c->phy_chan->num % 2) * 4;
953 /* Setup channel mode to logical or physical */
954 var = ((u32)(chan_is_logical(d40c)) + 1) <<
955 D40_CHAN_POS(d40c->phy_chan->num);
956 writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
957
958 /* Setup operational mode option register */
959 var = d40_get_prmo(d40c) << D40_CHAN_POS(d40c->phy_chan->num);
960
961 writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
962
963 if (chan_is_logical(d40c)) {
964 int lidx = (d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS)
965 & D40_SREG_ELEM_LOG_LIDX_MASK;
966 void __iomem *chanbase = chan_base(d40c);
967
968 /* Set default config for CFG reg */
969 writel(d40c->src_def_cfg, chanbase + D40_CHAN_REG_SSCFG);
970 writel(d40c->dst_def_cfg, chanbase + D40_CHAN_REG_SDCFG);
971
972 /* Set LIDX for lcla */
973 writel(lidx, chanbase + D40_CHAN_REG_SSELT);
974 writel(lidx, chanbase + D40_CHAN_REG_SDELT);
975 }
976}
977
978static u32 d40_residue(struct d40_chan *d40c)
979{
980 u32 num_elt;
981
982 if (chan_is_logical(d40c))
983 num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
984 >> D40_MEM_LCSP2_ECNT_POS;
985 else {
986 u32 val = readl(chan_base(d40c) + D40_CHAN_REG_SDELT);
987 num_elt = (val & D40_SREG_ELEM_PHY_ECNT_MASK)
988 >> D40_SREG_ELEM_PHY_ECNT_POS;
989 }
990
991 return num_elt * (1 << d40c->dma_cfg.dst_info.data_width);
992}
993
994static bool d40_tx_is_linked(struct d40_chan *d40c)
995{
996 bool is_link;
997
998 if (chan_is_logical(d40c))
999 is_link = readl(&d40c->lcpa->lcsp3) & D40_MEM_LCSP3_DLOS_MASK;
1000 else
1001 is_link = readl(chan_base(d40c) + D40_CHAN_REG_SDLNK)
1002 & D40_SREG_LNK_PHYS_LNK_MASK;
1003
1004 return is_link;
1005}
1006
1007static int d40_pause(struct d40_chan *d40c)
1008{
1009 int res = 0;
1010 unsigned long flags;
1011
1012 if (!d40c->busy)
1013 return 0;
1014
1015 spin_lock_irqsave(&d40c->lock, flags);
1016
1017 res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1018 if (res == 0) {
1019 if (chan_is_logical(d40c)) {
1020 d40_config_set_event(d40c, false);
1021 /* Resume the other logical channels if any */
1022 if (d40_chan_has_events(d40c))
1023 res = d40_channel_execute_command(d40c,
1024 D40_DMA_RUN);
1025 }
1026 }
1027
1028 spin_unlock_irqrestore(&d40c->lock, flags);
1029 return res;
1030}
1031
1032static int d40_resume(struct d40_chan *d40c)
1033{
1034 int res = 0;
1035 unsigned long flags;
1036
1037 if (!d40c->busy)
1038 return 0;
1039
1040 spin_lock_irqsave(&d40c->lock, flags);
1041
1042 if (d40c->base->rev == 0)
1043 if (chan_is_logical(d40c)) {
1044 res = d40_channel_execute_command(d40c,
1045 D40_DMA_SUSPEND_REQ);
1046 goto no_suspend;
1047 }
1048
1049 /* If bytes left to transfer or linked tx resume job */
1050 if (d40_residue(d40c) || d40_tx_is_linked(d40c)) {
1051
1052 if (chan_is_logical(d40c))
1053 d40_config_set_event(d40c, true);
1054
1055 res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1056 }
1057
1058no_suspend:
1059 spin_unlock_irqrestore(&d40c->lock, flags);
1060 return res;
1061}
1062
1063static int d40_terminate_all(struct d40_chan *chan)
1064{
1065 unsigned long flags;
1066 int ret = 0;
1067
1068 ret = d40_pause(chan);
1069 if (!ret && chan_is_physical(chan))
1070 ret = d40_channel_execute_command(chan, D40_DMA_STOP);
1071
1072 spin_lock_irqsave(&chan->lock, flags);
1073 d40_term_all(chan);
1074 spin_unlock_irqrestore(&chan->lock, flags);
1075
1076 return ret;
1077}
1078
1079static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
1080{
1081 struct d40_chan *d40c = container_of(tx->chan,
1082 struct d40_chan,
1083 chan);
1084 struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
1085 unsigned long flags;
1086
1087 spin_lock_irqsave(&d40c->lock, flags);
1088
1089 d40c->chan.cookie++;
1090
1091 if (d40c->chan.cookie < 0)
1092 d40c->chan.cookie = 1;
1093
1094 d40d->txd.cookie = d40c->chan.cookie;
1095
1096 d40_desc_queue(d40c, d40d);
1097
1098 spin_unlock_irqrestore(&d40c->lock, flags);
1099
1100 return tx->cookie;
1101}
1102
1103static int d40_start(struct d40_chan *d40c)
1104{
1105 if (d40c->base->rev == 0) {
1106 int err;
1107
1108 if (chan_is_logical(d40c)) {
1109 err = d40_channel_execute_command(d40c,
1110 D40_DMA_SUSPEND_REQ);
1111 if (err)
1112 return err;
1113 }
1114 }
1115
1116 if (chan_is_logical(d40c))
1117 d40_config_set_event(d40c, true);
1118
1119 return d40_channel_execute_command(d40c, D40_DMA_RUN);
1120}
1121
1122static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
1123{
1124 struct d40_desc *d40d;
1125 int err;
1126
1127 /* Start queued jobs, if any */
1128 d40d = d40_first_queued(d40c);
1129
1130 if (d40d != NULL) {
1131 d40c->busy = true;
1132
1133 /* Remove from queue */
1134 d40_desc_remove(d40d);
1135
1136 /* Add to active queue */
1137 d40_desc_submit(d40c, d40d);
1138
1139 /* Initiate DMA job */
1140 d40_desc_load(d40c, d40d);
1141
1142 /* Start dma job */
1143 err = d40_start(d40c);
1144
1145 if (err)
1146 return NULL;
1147 }
1148
1149 return d40d;
1150}
1151
1152/* called from interrupt context */
1153static void dma_tc_handle(struct d40_chan *d40c)
1154{
1155 struct d40_desc *d40d;
1156
1157 /* Get first active entry from list */
1158 d40d = d40_first_active_get(d40c);
1159
1160 if (d40d == NULL)
1161 return;
1162
1163 if (d40d->cyclic) {
1164 /*
1165 * If this was a paritially loaded list, we need to reloaded
1166 * it, and only when the list is completed. We need to check
1167 * for done because the interrupt will hit for every link, and
1168 * not just the last one.
1169 */
1170 if (d40d->lli_current < d40d->lli_len
1171 && !d40_tx_is_linked(d40c)
1172 && !d40_residue(d40c)) {
1173 d40_lcla_free_all(d40c, d40d);
1174 d40_desc_load(d40c, d40d);
1175 (void) d40_start(d40c);
1176
1177 if (d40d->lli_current == d40d->lli_len)
1178 d40d->lli_current = 0;
1179 }
1180 } else {
1181 d40_lcla_free_all(d40c, d40d);
1182
1183 if (d40d->lli_current < d40d->lli_len) {
1184 d40_desc_load(d40c, d40d);
1185 /* Start dma job */
1186 (void) d40_start(d40c);
1187 return;
1188 }
1189
1190 if (d40_queue_start(d40c) == NULL)
1191 d40c->busy = false;
1192 }
1193
1194 d40c->pending_tx++;
1195 tasklet_schedule(&d40c->tasklet);
1196
1197}
1198
1199static void dma_tasklet(unsigned long data)
1200{
1201 struct d40_chan *d40c = (struct d40_chan *) data;
1202 struct d40_desc *d40d;
1203 unsigned long flags;
1204 dma_async_tx_callback callback;
1205 void *callback_param;
1206
1207 spin_lock_irqsave(&d40c->lock, flags);
1208
1209 /* Get first active entry from list */
1210 d40d = d40_first_active_get(d40c);
1211 if (d40d == NULL)
1212 goto err;
1213
1214 if (!d40d->cyclic)
1215 d40c->completed = d40d->txd.cookie;
1216
1217 /*
1218 * If terminating a channel pending_tx is set to zero.
1219 * This prevents any finished active jobs to return to the client.
1220 */
1221 if (d40c->pending_tx == 0) {
1222 spin_unlock_irqrestore(&d40c->lock, flags);
1223 return;
1224 }
1225
1226 /* Callback to client */
1227 callback = d40d->txd.callback;
1228 callback_param = d40d->txd.callback_param;
1229
1230 if (!d40d->cyclic) {
1231 if (async_tx_test_ack(&d40d->txd)) {
1232 d40_desc_remove(d40d);
1233 d40_desc_free(d40c, d40d);
1234 } else {
1235 if (!d40d->is_in_client_list) {
1236 d40_desc_remove(d40d);
1237 d40_lcla_free_all(d40c, d40d);
1238 list_add_tail(&d40d->node, &d40c->client);
1239 d40d->is_in_client_list = true;
1240 }
1241 }
1242 }
1243
1244 d40c->pending_tx--;
1245
1246 if (d40c->pending_tx)
1247 tasklet_schedule(&d40c->tasklet);
1248
1249 spin_unlock_irqrestore(&d40c->lock, flags);
1250
1251 if (callback && (d40d->txd.flags & DMA_PREP_INTERRUPT))
1252 callback(callback_param);
1253
1254 return;
1255
1256 err:
1257 /* Rescue manoeuvre if receiving double interrupts */
1258 if (d40c->pending_tx > 0)
1259 d40c->pending_tx--;
1260 spin_unlock_irqrestore(&d40c->lock, flags);
1261}
1262
1263static irqreturn_t d40_handle_interrupt(int irq, void *data)
1264{
1265 static const struct d40_interrupt_lookup il[] = {
1266 {D40_DREG_LCTIS0, D40_DREG_LCICR0, false, 0},
1267 {D40_DREG_LCTIS1, D40_DREG_LCICR1, false, 32},
1268 {D40_DREG_LCTIS2, D40_DREG_LCICR2, false, 64},
1269 {D40_DREG_LCTIS3, D40_DREG_LCICR3, false, 96},
1270 {D40_DREG_LCEIS0, D40_DREG_LCICR0, true, 0},
1271 {D40_DREG_LCEIS1, D40_DREG_LCICR1, true, 32},
1272 {D40_DREG_LCEIS2, D40_DREG_LCICR2, true, 64},
1273 {D40_DREG_LCEIS3, D40_DREG_LCICR3, true, 96},
1274 {D40_DREG_PCTIS, D40_DREG_PCICR, false, D40_PHY_CHAN},
1275 {D40_DREG_PCEIS, D40_DREG_PCICR, true, D40_PHY_CHAN},
1276 };
1277
1278 int i;
1279 u32 regs[ARRAY_SIZE(il)];
1280 u32 idx;
1281 u32 row;
1282 long chan = -1;
1283 struct d40_chan *d40c;
1284 unsigned long flags;
1285 struct d40_base *base = data;
1286
1287 spin_lock_irqsave(&base->interrupt_lock, flags);
1288
1289 /* Read interrupt status of both logical and physical channels */
1290 for (i = 0; i < ARRAY_SIZE(il); i++)
1291 regs[i] = readl(base->virtbase + il[i].src);
1292
1293 for (;;) {
1294
1295 chan = find_next_bit((unsigned long *)regs,
1296 BITS_PER_LONG * ARRAY_SIZE(il), chan + 1);
1297
1298 /* No more set bits found? */
1299 if (chan == BITS_PER_LONG * ARRAY_SIZE(il))
1300 break;
1301
1302 row = chan / BITS_PER_LONG;
1303 idx = chan & (BITS_PER_LONG - 1);
1304
1305 /* ACK interrupt */
1306 writel(1 << idx, base->virtbase + il[row].clr);
1307
1308 if (il[row].offset == D40_PHY_CHAN)
1309 d40c = base->lookup_phy_chans[idx];
1310 else
1311 d40c = base->lookup_log_chans[il[row].offset + idx];
1312 spin_lock(&d40c->lock);
1313
1314 if (!il[row].is_error)
1315 dma_tc_handle(d40c);
1316 else
1317 d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1318 chan, il[row].offset, idx);
1319
1320 spin_unlock(&d40c->lock);
1321 }
1322
1323 spin_unlock_irqrestore(&base->interrupt_lock, flags);
1324
1325 return IRQ_HANDLED;
1326}
1327
1328static int d40_validate_conf(struct d40_chan *d40c,
1329 struct stedma40_chan_cfg *conf)
1330{
1331 int res = 0;
1332 u32 dst_event_group = D40_TYPE_TO_GROUP(conf->dst_dev_type);
1333 u32 src_event_group = D40_TYPE_TO_GROUP(conf->src_dev_type);
1334 bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1335
1336 if (!conf->dir) {
1337 chan_err(d40c, "Invalid direction.\n");
1338 res = -EINVAL;
1339 }
1340
1341 if (conf->dst_dev_type != STEDMA40_DEV_DST_MEMORY &&
1342 d40c->base->plat_data->dev_tx[conf->dst_dev_type] == 0 &&
1343 d40c->runtime_addr == 0) {
1344
1345 chan_err(d40c, "Invalid TX channel address (%d)\n",
1346 conf->dst_dev_type);
1347 res = -EINVAL;
1348 }
1349
1350 if (conf->src_dev_type != STEDMA40_DEV_SRC_MEMORY &&
1351 d40c->base->plat_data->dev_rx[conf->src_dev_type] == 0 &&
1352 d40c->runtime_addr == 0) {
1353 chan_err(d40c, "Invalid RX channel address (%d)\n",
1354 conf->src_dev_type);
1355 res = -EINVAL;
1356 }
1357
1358 if (conf->dir == STEDMA40_MEM_TO_PERIPH &&
1359 dst_event_group == STEDMA40_DEV_DST_MEMORY) {
1360 chan_err(d40c, "Invalid dst\n");
1361 res = -EINVAL;
1362 }
1363
1364 if (conf->dir == STEDMA40_PERIPH_TO_MEM &&
1365 src_event_group == STEDMA40_DEV_SRC_MEMORY) {
1366 chan_err(d40c, "Invalid src\n");
1367 res = -EINVAL;
1368 }
1369
1370 if (src_event_group == STEDMA40_DEV_SRC_MEMORY &&
1371 dst_event_group == STEDMA40_DEV_DST_MEMORY && is_log) {
1372 chan_err(d40c, "No event line\n");
1373 res = -EINVAL;
1374 }
1375
1376 if (conf->dir == STEDMA40_PERIPH_TO_PERIPH &&
1377 (src_event_group != dst_event_group)) {
1378 chan_err(d40c, "Invalid event group\n");
1379 res = -EINVAL;
1380 }
1381
1382 if (conf->dir == STEDMA40_PERIPH_TO_PERIPH) {
1383 /*
1384 * DMAC HW supports it. Will be added to this driver,
1385 * in case any dma client requires it.
1386 */
1387 chan_err(d40c, "periph to periph not supported\n");
1388 res = -EINVAL;
1389 }
1390
1391 if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1392 (1 << conf->src_info.data_width) !=
1393 d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1394 (1 << conf->dst_info.data_width)) {
1395 /*
1396 * The DMAC hardware only supports
1397 * src (burst x width) == dst (burst x width)
1398 */
1399
1400 chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1401 res = -EINVAL;
1402 }
1403
1404 return res;
1405}
1406
1407static bool d40_alloc_mask_set(struct d40_phy_res *phy, bool is_src,
1408 int log_event_line, bool is_log)
1409{
1410 unsigned long flags;
1411 spin_lock_irqsave(&phy->lock, flags);
1412 if (!is_log) {
1413 /* Physical interrupts are masked per physical full channel */
1414 if (phy->allocated_src == D40_ALLOC_FREE &&
1415 phy->allocated_dst == D40_ALLOC_FREE) {
1416 phy->allocated_dst = D40_ALLOC_PHY;
1417 phy->allocated_src = D40_ALLOC_PHY;
1418 goto found;
1419 } else
1420 goto not_found;
1421 }
1422
1423 /* Logical channel */
1424 if (is_src) {
1425 if (phy->allocated_src == D40_ALLOC_PHY)
1426 goto not_found;
1427
1428 if (phy->allocated_src == D40_ALLOC_FREE)
1429 phy->allocated_src = D40_ALLOC_LOG_FREE;
1430
1431 if (!(phy->allocated_src & (1 << log_event_line))) {
1432 phy->allocated_src |= 1 << log_event_line;
1433 goto found;
1434 } else
1435 goto not_found;
1436 } else {
1437 if (phy->allocated_dst == D40_ALLOC_PHY)
1438 goto not_found;
1439
1440 if (phy->allocated_dst == D40_ALLOC_FREE)
1441 phy->allocated_dst = D40_ALLOC_LOG_FREE;
1442
1443 if (!(phy->allocated_dst & (1 << log_event_line))) {
1444 phy->allocated_dst |= 1 << log_event_line;
1445 goto found;
1446 } else
1447 goto not_found;
1448 }
1449
1450not_found:
1451 spin_unlock_irqrestore(&phy->lock, flags);
1452 return false;
1453found:
1454 spin_unlock_irqrestore(&phy->lock, flags);
1455 return true;
1456}
1457
1458static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1459 int log_event_line)
1460{
1461 unsigned long flags;
1462 bool is_free = false;
1463
1464 spin_lock_irqsave(&phy->lock, flags);
1465 if (!log_event_line) {
1466 phy->allocated_dst = D40_ALLOC_FREE;
1467 phy->allocated_src = D40_ALLOC_FREE;
1468 is_free = true;
1469 goto out;
1470 }
1471
1472 /* Logical channel */
1473 if (is_src) {
1474 phy->allocated_src &= ~(1 << log_event_line);
1475 if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1476 phy->allocated_src = D40_ALLOC_FREE;
1477 } else {
1478 phy->allocated_dst &= ~(1 << log_event_line);
1479 if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1480 phy->allocated_dst = D40_ALLOC_FREE;
1481 }
1482
1483 is_free = ((phy->allocated_src | phy->allocated_dst) ==
1484 D40_ALLOC_FREE);
1485
1486out:
1487 spin_unlock_irqrestore(&phy->lock, flags);
1488
1489 return is_free;
1490}
1491
1492static int d40_allocate_channel(struct d40_chan *d40c)
1493{
1494 int dev_type;
1495 int event_group;
1496 int event_line;
1497 struct d40_phy_res *phys;
1498 int i;
1499 int j;
1500 int log_num;
1501 bool is_src;
1502 bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1503
1504 phys = d40c->base->phy_res;
1505
1506 if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1507 dev_type = d40c->dma_cfg.src_dev_type;
1508 log_num = 2 * dev_type;
1509 is_src = true;
1510 } else if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1511 d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1512 /* dst event lines are used for logical memcpy */
1513 dev_type = d40c->dma_cfg.dst_dev_type;
1514 log_num = 2 * dev_type + 1;
1515 is_src = false;
1516 } else
1517 return -EINVAL;
1518
1519 event_group = D40_TYPE_TO_GROUP(dev_type);
1520 event_line = D40_TYPE_TO_EVENT(dev_type);
1521
1522 if (!is_log) {
1523 if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1524 /* Find physical half channel */
1525 for (i = 0; i < d40c->base->num_phy_chans; i++) {
1526
1527 if (d40_alloc_mask_set(&phys[i], is_src,
1528 0, is_log))
1529 goto found_phy;
1530 }
1531 } else
1532 for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1533 int phy_num = j + event_group * 2;
1534 for (i = phy_num; i < phy_num + 2; i++) {
1535 if (d40_alloc_mask_set(&phys[i],
1536 is_src,
1537 0,
1538 is_log))
1539 goto found_phy;
1540 }
1541 }
1542 return -EINVAL;
1543found_phy:
1544 d40c->phy_chan = &phys[i];
1545 d40c->log_num = D40_PHY_CHAN;
1546 goto out;
1547 }
1548 if (dev_type == -1)
1549 return -EINVAL;
1550
1551 /* Find logical channel */
1552 for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1553 int phy_num = j + event_group * 2;
1554 /*
1555 * Spread logical channels across all available physical rather
1556 * than pack every logical channel at the first available phy
1557 * channels.
1558 */
1559 if (is_src) {
1560 for (i = phy_num; i < phy_num + 2; i++) {
1561 if (d40_alloc_mask_set(&phys[i], is_src,
1562 event_line, is_log))
1563 goto found_log;
1564 }
1565 } else {
1566 for (i = phy_num + 1; i >= phy_num; i--) {
1567 if (d40_alloc_mask_set(&phys[i], is_src,
1568 event_line, is_log))
1569 goto found_log;
1570 }
1571 }
1572 }
1573 return -EINVAL;
1574
1575found_log:
1576 d40c->phy_chan = &phys[i];
1577 d40c->log_num = log_num;
1578out:
1579
1580 if (is_log)
1581 d40c->base->lookup_log_chans[d40c->log_num] = d40c;
1582 else
1583 d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
1584
1585 return 0;
1586
1587}
1588
1589static int d40_config_memcpy(struct d40_chan *d40c)
1590{
1591 dma_cap_mask_t cap = d40c->chan.device->cap_mask;
1592
1593 if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
1594 d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_log;
1595 d40c->dma_cfg.src_dev_type = STEDMA40_DEV_SRC_MEMORY;
1596 d40c->dma_cfg.dst_dev_type = d40c->base->plat_data->
1597 memcpy[d40c->chan.chan_id];
1598
1599 } else if (dma_has_cap(DMA_MEMCPY, cap) &&
1600 dma_has_cap(DMA_SLAVE, cap)) {
1601 d40c->dma_cfg = *d40c->base->plat_data->memcpy_conf_phy;
1602 } else {
1603 chan_err(d40c, "No memcpy\n");
1604 return -EINVAL;
1605 }
1606
1607 return 0;
1608}
1609
1610
1611static int d40_free_dma(struct d40_chan *d40c)
1612{
1613
1614 int res = 0;
1615 u32 event;
1616 struct d40_phy_res *phy = d40c->phy_chan;
1617 bool is_src;
1618
1619 /* Terminate all queued and active transfers */
1620 d40_term_all(d40c);
1621
1622 if (phy == NULL) {
1623 chan_err(d40c, "phy == null\n");
1624 return -EINVAL;
1625 }
1626
1627 if (phy->allocated_src == D40_ALLOC_FREE &&
1628 phy->allocated_dst == D40_ALLOC_FREE) {
1629 chan_err(d40c, "channel already free\n");
1630 return -EINVAL;
1631 }
1632
1633 if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1634 d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1635 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
1636 is_src = false;
1637 } else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1638 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
1639 is_src = true;
1640 } else {
1641 chan_err(d40c, "Unknown direction\n");
1642 return -EINVAL;
1643 }
1644
1645 res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1646 if (res) {
1647 chan_err(d40c, "suspend failed\n");
1648 return res;
1649 }
1650
1651 if (chan_is_logical(d40c)) {
1652 /* Release logical channel, deactivate the event line */
1653
1654 d40_config_set_event(d40c, false);
1655 d40c->base->lookup_log_chans[d40c->log_num] = NULL;
1656
1657 /*
1658 * Check if there are more logical allocation
1659 * on this phy channel.
1660 */
1661 if (!d40_alloc_mask_free(phy, is_src, event)) {
1662 /* Resume the other logical channels if any */
1663 if (d40_chan_has_events(d40c)) {
1664 res = d40_channel_execute_command(d40c,
1665 D40_DMA_RUN);
1666 if (res) {
1667 chan_err(d40c,
1668 "Executing RUN command\n");
1669 return res;
1670 }
1671 }
1672 return 0;
1673 }
1674 } else {
1675 (void) d40_alloc_mask_free(phy, is_src, 0);
1676 }
1677
1678 /* Release physical channel */
1679 res = d40_channel_execute_command(d40c, D40_DMA_STOP);
1680 if (res) {
1681 chan_err(d40c, "Failed to stop channel\n");
1682 return res;
1683 }
1684 d40c->phy_chan = NULL;
1685 d40c->configured = false;
1686 d40c->base->lookup_phy_chans[phy->num] = NULL;
1687
1688 return 0;
1689}
1690
1691static bool d40_is_paused(struct d40_chan *d40c)
1692{
1693 void __iomem *chanbase = chan_base(d40c);
1694 bool is_paused = false;
1695 unsigned long flags;
1696 void __iomem *active_reg;
1697 u32 status;
1698 u32 event;
1699
1700 spin_lock_irqsave(&d40c->lock, flags);
1701
1702 if (chan_is_physical(d40c)) {
1703 if (d40c->phy_chan->num % 2 == 0)
1704 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1705 else
1706 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1707
1708 status = (readl(active_reg) &
1709 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1710 D40_CHAN_POS(d40c->phy_chan->num);
1711 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1712 is_paused = true;
1713
1714 goto _exit;
1715 }
1716
1717 if (d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH ||
1718 d40c->dma_cfg.dir == STEDMA40_MEM_TO_MEM) {
1719 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dst_dev_type);
1720 status = readl(chanbase + D40_CHAN_REG_SDLNK);
1721 } else if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) {
1722 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.src_dev_type);
1723 status = readl(chanbase + D40_CHAN_REG_SSLNK);
1724 } else {
1725 chan_err(d40c, "Unknown direction\n");
1726 goto _exit;
1727 }
1728
1729 status = (status & D40_EVENTLINE_MASK(event)) >>
1730 D40_EVENTLINE_POS(event);
1731
1732 if (status != D40_DMA_RUN)
1733 is_paused = true;
1734_exit:
1735 spin_unlock_irqrestore(&d40c->lock, flags);
1736 return is_paused;
1737
1738}
1739
1740
1741static u32 stedma40_residue(struct dma_chan *chan)
1742{
1743 struct d40_chan *d40c =
1744 container_of(chan, struct d40_chan, chan);
1745 u32 bytes_left;
1746 unsigned long flags;
1747
1748 spin_lock_irqsave(&d40c->lock, flags);
1749 bytes_left = d40_residue(d40c);
1750 spin_unlock_irqrestore(&d40c->lock, flags);
1751
1752 return bytes_left;
1753}
1754
1755static int
1756d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
1757 struct scatterlist *sg_src, struct scatterlist *sg_dst,
1758 unsigned int sg_len, dma_addr_t src_dev_addr,
1759 dma_addr_t dst_dev_addr)
1760{
1761 struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
1762 struct stedma40_half_channel_info *src_info = &cfg->src_info;
1763 struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
1764 int ret;
1765
1766 ret = d40_log_sg_to_lli(sg_src, sg_len,
1767 src_dev_addr,
1768 desc->lli_log.src,
1769 chan->log_def.lcsp1,
1770 src_info->data_width,
1771 dst_info->data_width);
1772
1773 ret = d40_log_sg_to_lli(sg_dst, sg_len,
1774 dst_dev_addr,
1775 desc->lli_log.dst,
1776 chan->log_def.lcsp3,
1777 dst_info->data_width,
1778 src_info->data_width);
1779
1780 return ret < 0 ? ret : 0;
1781}
1782
1783static int
1784d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
1785 struct scatterlist *sg_src, struct scatterlist *sg_dst,
1786 unsigned int sg_len, dma_addr_t src_dev_addr,
1787 dma_addr_t dst_dev_addr)
1788{
1789 struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
1790 struct stedma40_half_channel_info *src_info = &cfg->src_info;
1791 struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
1792 unsigned long flags = 0;
1793 int ret;
1794
1795 if (desc->cyclic)
1796 flags |= LLI_CYCLIC | LLI_TERM_INT;
1797
1798 ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
1799 desc->lli_phy.src,
1800 virt_to_phys(desc->lli_phy.src),
1801 chan->src_def_cfg,
1802 src_info, dst_info, flags);
1803
1804 ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
1805 desc->lli_phy.dst,
1806 virt_to_phys(desc->lli_phy.dst),
1807 chan->dst_def_cfg,
1808 dst_info, src_info, flags);
1809
1810 dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
1811 desc->lli_pool.size, DMA_TO_DEVICE);
1812
1813 return ret < 0 ? ret : 0;
1814}
1815
1816
1817static struct d40_desc *
1818d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
1819 unsigned int sg_len, unsigned long dma_flags)
1820{
1821 struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
1822 struct d40_desc *desc;
1823 int ret;
1824
1825 desc = d40_desc_get(chan);
1826 if (!desc)
1827 return NULL;
1828
1829 desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
1830 cfg->dst_info.data_width);
1831 if (desc->lli_len < 0) {
1832 chan_err(chan, "Unaligned size\n");
1833 goto err;
1834 }
1835
1836 ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
1837 if (ret < 0) {
1838 chan_err(chan, "Could not allocate lli\n");
1839 goto err;
1840 }
1841
1842
1843 desc->lli_current = 0;
1844 desc->txd.flags = dma_flags;
1845 desc->txd.tx_submit = d40_tx_submit;
1846
1847 dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
1848
1849 return desc;
1850
1851err:
1852 d40_desc_free(chan, desc);
1853 return NULL;
1854}
1855
1856static dma_addr_t
1857d40_get_dev_addr(struct d40_chan *chan, enum dma_data_direction direction)
1858{
1859 struct stedma40_platform_data *plat = chan->base->plat_data;
1860 struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
1861 dma_addr_t addr = 0;
1862
1863 if (chan->runtime_addr)
1864 return chan->runtime_addr;
1865
1866 if (direction == DMA_FROM_DEVICE)
1867 addr = plat->dev_rx[cfg->src_dev_type];
1868 else if (direction == DMA_TO_DEVICE)
1869 addr = plat->dev_tx[cfg->dst_dev_type];
1870
1871 return addr;
1872}
1873
1874static struct dma_async_tx_descriptor *
1875d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
1876 struct scatterlist *sg_dst, unsigned int sg_len,
1877 enum dma_data_direction direction, unsigned long dma_flags)
1878{
1879 struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
1880 dma_addr_t src_dev_addr = 0;
1881 dma_addr_t dst_dev_addr = 0;
1882 struct d40_desc *desc;
1883 unsigned long flags;
1884 int ret;
1885
1886 if (!chan->phy_chan) {
1887 chan_err(chan, "Cannot prepare unallocated channel\n");
1888 return NULL;
1889 }
1890
1891
1892 spin_lock_irqsave(&chan->lock, flags);
1893
1894 desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
1895 if (desc == NULL)
1896 goto err;
1897
1898 if (sg_next(&sg_src[sg_len - 1]) == sg_src)
1899 desc->cyclic = true;
1900
1901 if (direction != DMA_NONE) {
1902 dma_addr_t dev_addr = d40_get_dev_addr(chan, direction);
1903
1904 if (direction == DMA_FROM_DEVICE)
1905 src_dev_addr = dev_addr;
1906 else if (direction == DMA_TO_DEVICE)
1907 dst_dev_addr = dev_addr;
1908 }
1909
1910 if (chan_is_logical(chan))
1911 ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
1912 sg_len, src_dev_addr, dst_dev_addr);
1913 else
1914 ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
1915 sg_len, src_dev_addr, dst_dev_addr);
1916
1917 if (ret) {
1918 chan_err(chan, "Failed to prepare %s sg job: %d\n",
1919 chan_is_logical(chan) ? "log" : "phy", ret);
1920 goto err;
1921 }
1922
1923 /*
1924 * add descriptor to the prepare queue in order to be able
1925 * to free them later in terminate_all
1926 */
1927 list_add_tail(&desc->node, &chan->prepare_queue);
1928
1929 spin_unlock_irqrestore(&chan->lock, flags);
1930
1931 return &desc->txd;
1932
1933err:
1934 if (desc)
1935 d40_desc_free(chan, desc);
1936 spin_unlock_irqrestore(&chan->lock, flags);
1937 return NULL;
1938}
1939
1940bool stedma40_filter(struct dma_chan *chan, void *data)
1941{
1942 struct stedma40_chan_cfg *info = data;
1943 struct d40_chan *d40c =
1944 container_of(chan, struct d40_chan, chan);
1945 int err;
1946
1947 if (data) {
1948 err = d40_validate_conf(d40c, info);
1949 if (!err)
1950 d40c->dma_cfg = *info;
1951 } else
1952 err = d40_config_memcpy(d40c);
1953
1954 if (!err)
1955 d40c->configured = true;
1956
1957 return err == 0;
1958}
1959EXPORT_SYMBOL(stedma40_filter);
1960
1961static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
1962{
1963 bool realtime = d40c->dma_cfg.realtime;
1964 bool highprio = d40c->dma_cfg.high_priority;
1965 u32 prioreg = highprio ? D40_DREG_PSEG1 : D40_DREG_PCEG1;
1966 u32 rtreg = realtime ? D40_DREG_RSEG1 : D40_DREG_RCEG1;
1967 u32 event = D40_TYPE_TO_EVENT(dev_type);
1968 u32 group = D40_TYPE_TO_GROUP(dev_type);
1969 u32 bit = 1 << event;
1970
1971 /* Destination event lines are stored in the upper halfword */
1972 if (!src)
1973 bit <<= 16;
1974
1975 writel(bit, d40c->base->virtbase + prioreg + group * 4);
1976 writel(bit, d40c->base->virtbase + rtreg + group * 4);
1977}
1978
1979static void d40_set_prio_realtime(struct d40_chan *d40c)
1980{
1981 if (d40c->base->rev < 3)
1982 return;
1983
1984 if ((d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM) ||
1985 (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH))
1986 __d40_set_prio_rt(d40c, d40c->dma_cfg.src_dev_type, true);
1987
1988 if ((d40c->dma_cfg.dir == STEDMA40_MEM_TO_PERIPH) ||
1989 (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_PERIPH))
1990 __d40_set_prio_rt(d40c, d40c->dma_cfg.dst_dev_type, false);
1991}
1992
1993/* DMA ENGINE functions */
1994static int d40_alloc_chan_resources(struct dma_chan *chan)
1995{
1996 int err;
1997 unsigned long flags;
1998 struct d40_chan *d40c =
1999 container_of(chan, struct d40_chan, chan);
2000 bool is_free_phy;
2001 spin_lock_irqsave(&d40c->lock, flags);
2002
2003 d40c->completed = chan->cookie = 1;
2004
2005 /* If no dma configuration is set use default configuration (memcpy) */
2006 if (!d40c->configured) {
2007 err = d40_config_memcpy(d40c);
2008 if (err) {
2009 chan_err(d40c, "Failed to configure memcpy channel\n");
2010 goto fail;
2011 }
2012 }
2013 is_free_phy = (d40c->phy_chan == NULL);
2014
2015 err = d40_allocate_channel(d40c);
2016 if (err) {
2017 chan_err(d40c, "Failed to allocate channel\n");
2018 goto fail;
2019 }
2020
2021 /* Fill in basic CFG register values */
2022 d40_phy_cfg(&d40c->dma_cfg, &d40c->src_def_cfg,
2023 &d40c->dst_def_cfg, chan_is_logical(d40c));
2024
2025 d40_set_prio_realtime(d40c);
2026
2027 if (chan_is_logical(d40c)) {
2028 d40_log_cfg(&d40c->dma_cfg,
2029 &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2030
2031 if (d40c->dma_cfg.dir == STEDMA40_PERIPH_TO_MEM)
2032 d40c->lcpa = d40c->base->lcpa_base +
2033 d40c->dma_cfg.src_dev_type * D40_LCPA_CHAN_SIZE;
2034 else
2035 d40c->lcpa = d40c->base->lcpa_base +
2036 d40c->dma_cfg.dst_dev_type *
2037 D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2038 }
2039
2040 /*
2041 * Only write channel configuration to the DMA if the physical
2042 * resource is free. In case of multiple logical channels
2043 * on the same physical resource, only the first write is necessary.
2044 */
2045 if (is_free_phy)
2046 d40_config_write(d40c);
2047fail:
2048 spin_unlock_irqrestore(&d40c->lock, flags);
2049 return err;
2050}
2051
2052static void d40_free_chan_resources(struct dma_chan *chan)
2053{
2054 struct d40_chan *d40c =
2055 container_of(chan, struct d40_chan, chan);
2056 int err;
2057 unsigned long flags;
2058
2059 if (d40c->phy_chan == NULL) {
2060 chan_err(d40c, "Cannot free unallocated channel\n");
2061 return;
2062 }
2063
2064
2065 spin_lock_irqsave(&d40c->lock, flags);
2066
2067 err = d40_free_dma(d40c);
2068
2069 if (err)
2070 chan_err(d40c, "Failed to free channel\n");
2071 spin_unlock_irqrestore(&d40c->lock, flags);
2072}
2073
2074static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2075 dma_addr_t dst,
2076 dma_addr_t src,
2077 size_t size,
2078 unsigned long dma_flags)
2079{
2080 struct scatterlist dst_sg;
2081 struct scatterlist src_sg;
2082
2083 sg_init_table(&dst_sg, 1);
2084 sg_init_table(&src_sg, 1);
2085
2086 sg_dma_address(&dst_sg) = dst;
2087 sg_dma_address(&src_sg) = src;
2088
2089 sg_dma_len(&dst_sg) = size;
2090 sg_dma_len(&src_sg) = size;
2091
2092 return d40_prep_sg(chan, &src_sg, &dst_sg, 1, DMA_NONE, dma_flags);
2093}
2094
2095static struct dma_async_tx_descriptor *
2096d40_prep_memcpy_sg(struct dma_chan *chan,
2097 struct scatterlist *dst_sg, unsigned int dst_nents,
2098 struct scatterlist *src_sg, unsigned int src_nents,
2099 unsigned long dma_flags)
2100{
2101 if (dst_nents != src_nents)
2102 return NULL;
2103
2104 return d40_prep_sg(chan, src_sg, dst_sg, src_nents, DMA_NONE, dma_flags);
2105}
2106
2107static struct dma_async_tx_descriptor *d40_prep_slave_sg(struct dma_chan *chan,
2108 struct scatterlist *sgl,
2109 unsigned int sg_len,
2110 enum dma_data_direction direction,
2111 unsigned long dma_flags)
2112{
2113 if (direction != DMA_FROM_DEVICE && direction != DMA_TO_DEVICE)
2114 return NULL;
2115
2116 return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2117}
2118
2119static struct dma_async_tx_descriptor *
2120dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2121 size_t buf_len, size_t period_len,
2122 enum dma_data_direction direction)
2123{
2124 unsigned int periods = buf_len / period_len;
2125 struct dma_async_tx_descriptor *txd;
2126 struct scatterlist *sg;
2127 int i;
2128
2129 sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2130 for (i = 0; i < periods; i++) {
2131 sg_dma_address(&sg[i]) = dma_addr;
2132 sg_dma_len(&sg[i]) = period_len;
2133 dma_addr += period_len;
2134 }
2135
2136 sg[periods].offset = 0;
2137 sg[periods].length = 0;
2138 sg[periods].page_link =
2139 ((unsigned long)sg | 0x01) & ~0x02;
2140
2141 txd = d40_prep_sg(chan, sg, sg, periods, direction,
2142 DMA_PREP_INTERRUPT);
2143
2144 kfree(sg);
2145
2146 return txd;
2147}
2148
2149static enum dma_status d40_tx_status(struct dma_chan *chan,
2150 dma_cookie_t cookie,
2151 struct dma_tx_state *txstate)
2152{
2153 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2154 dma_cookie_t last_used;
2155 dma_cookie_t last_complete;
2156 int ret;
2157
2158 if (d40c->phy_chan == NULL) {
2159 chan_err(d40c, "Cannot read status of unallocated channel\n");
2160 return -EINVAL;
2161 }
2162
2163 last_complete = d40c->completed;
2164 last_used = chan->cookie;
2165
2166 if (d40_is_paused(d40c))
2167 ret = DMA_PAUSED;
2168 else
2169 ret = dma_async_is_complete(cookie, last_complete, last_used);
2170
2171 dma_set_tx_state(txstate, last_complete, last_used,
2172 stedma40_residue(chan));
2173
2174 return ret;
2175}
2176
2177static void d40_issue_pending(struct dma_chan *chan)
2178{
2179 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2180 unsigned long flags;
2181
2182 if (d40c->phy_chan == NULL) {
2183 chan_err(d40c, "Channel is not allocated!\n");
2184 return;
2185 }
2186
2187 spin_lock_irqsave(&d40c->lock, flags);
2188
2189 list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2190
2191 /* Busy means that queued jobs are already being processed */
2192 if (!d40c->busy)
2193 (void) d40_queue_start(d40c);
2194
2195 spin_unlock_irqrestore(&d40c->lock, flags);
2196}
2197
2198static int
2199dma40_config_to_halfchannel(struct d40_chan *d40c,
2200 struct stedma40_half_channel_info *info,
2201 enum dma_slave_buswidth width,
2202 u32 maxburst)
2203{
2204 enum stedma40_periph_data_width addr_width;
2205 int psize;
2206
2207 switch (width) {
2208 case DMA_SLAVE_BUSWIDTH_1_BYTE:
2209 addr_width = STEDMA40_BYTE_WIDTH;
2210 break;
2211 case DMA_SLAVE_BUSWIDTH_2_BYTES:
2212 addr_width = STEDMA40_HALFWORD_WIDTH;
2213 break;
2214 case DMA_SLAVE_BUSWIDTH_4_BYTES:
2215 addr_width = STEDMA40_WORD_WIDTH;
2216 break;
2217 case DMA_SLAVE_BUSWIDTH_8_BYTES:
2218 addr_width = STEDMA40_DOUBLEWORD_WIDTH;
2219 break;
2220 default:
2221 dev_err(d40c->base->dev,
2222 "illegal peripheral address width "
2223 "requested (%d)\n",
2224 width);
2225 return -EINVAL;
2226 }
2227
2228 if (chan_is_logical(d40c)) {
2229 if (maxburst >= 16)
2230 psize = STEDMA40_PSIZE_LOG_16;
2231 else if (maxburst >= 8)
2232 psize = STEDMA40_PSIZE_LOG_8;
2233 else if (maxburst >= 4)
2234 psize = STEDMA40_PSIZE_LOG_4;
2235 else
2236 psize = STEDMA40_PSIZE_LOG_1;
2237 } else {
2238 if (maxburst >= 16)
2239 psize = STEDMA40_PSIZE_PHY_16;
2240 else if (maxburst >= 8)
2241 psize = STEDMA40_PSIZE_PHY_8;
2242 else if (maxburst >= 4)
2243 psize = STEDMA40_PSIZE_PHY_4;
2244 else
2245 psize = STEDMA40_PSIZE_PHY_1;
2246 }
2247
2248 info->data_width = addr_width;
2249 info->psize = psize;
2250 info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2251
2252 return 0;
2253}
2254
2255/* Runtime reconfiguration extension */
2256static int d40_set_runtime_config(struct dma_chan *chan,
2257 struct dma_slave_config *config)
2258{
2259 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2260 struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2261 enum dma_slave_buswidth src_addr_width, dst_addr_width;
2262 dma_addr_t config_addr;
2263 u32 src_maxburst, dst_maxburst;
2264 int ret;
2265
2266 src_addr_width = config->src_addr_width;
2267 src_maxburst = config->src_maxburst;
2268 dst_addr_width = config->dst_addr_width;
2269 dst_maxburst = config->dst_maxburst;
2270
2271 if (config->direction == DMA_FROM_DEVICE) {
2272 dma_addr_t dev_addr_rx =
2273 d40c->base->plat_data->dev_rx[cfg->src_dev_type];
2274
2275 config_addr = config->src_addr;
2276 if (dev_addr_rx)
2277 dev_dbg(d40c->base->dev,
2278 "channel has a pre-wired RX address %08x "
2279 "overriding with %08x\n",
2280 dev_addr_rx, config_addr);
2281 if (cfg->dir != STEDMA40_PERIPH_TO_MEM)
2282 dev_dbg(d40c->base->dev,
2283 "channel was not configured for peripheral "
2284 "to memory transfer (%d) overriding\n",
2285 cfg->dir);
2286 cfg->dir = STEDMA40_PERIPH_TO_MEM;
2287
2288 /* Configure the memory side */
2289 if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2290 dst_addr_width = src_addr_width;
2291 if (dst_maxburst == 0)
2292 dst_maxburst = src_maxburst;
2293
2294 } else if (config->direction == DMA_TO_DEVICE) {
2295 dma_addr_t dev_addr_tx =
2296 d40c->base->plat_data->dev_tx[cfg->dst_dev_type];
2297
2298 config_addr = config->dst_addr;
2299 if (dev_addr_tx)
2300 dev_dbg(d40c->base->dev,
2301 "channel has a pre-wired TX address %08x "
2302 "overriding with %08x\n",
2303 dev_addr_tx, config_addr);
2304 if (cfg->dir != STEDMA40_MEM_TO_PERIPH)
2305 dev_dbg(d40c->base->dev,
2306 "channel was not configured for memory "
2307 "to peripheral transfer (%d) overriding\n",
2308 cfg->dir);
2309 cfg->dir = STEDMA40_MEM_TO_PERIPH;
2310
2311 /* Configure the memory side */
2312 if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2313 src_addr_width = dst_addr_width;
2314 if (src_maxburst == 0)
2315 src_maxburst = dst_maxburst;
2316 } else {
2317 dev_err(d40c->base->dev,
2318 "unrecognized channel direction %d\n",
2319 config->direction);
2320 return -EINVAL;
2321 }
2322
2323 if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2324 dev_err(d40c->base->dev,
2325 "src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2326 src_maxburst,
2327 src_addr_width,
2328 dst_maxburst,
2329 dst_addr_width);
2330 return -EINVAL;
2331 }
2332
2333 ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2334 src_addr_width,
2335 src_maxburst);
2336 if (ret)
2337 return ret;
2338
2339 ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2340 dst_addr_width,
2341 dst_maxburst);
2342 if (ret)
2343 return ret;
2344
2345 /* Fill in register values */
2346 if (chan_is_logical(d40c))
2347 d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2348 else
2349 d40_phy_cfg(cfg, &d40c->src_def_cfg,
2350 &d40c->dst_def_cfg, false);
2351
2352 /* These settings will take precedence later */
2353 d40c->runtime_addr = config_addr;
2354 d40c->runtime_direction = config->direction;
2355 dev_dbg(d40c->base->dev,
2356 "configured channel %s for %s, data width %d/%d, "
2357 "maxburst %d/%d elements, LE, no flow control\n",
2358 dma_chan_name(chan),
2359 (config->direction == DMA_FROM_DEVICE) ? "RX" : "TX",
2360 src_addr_width, dst_addr_width,
2361 src_maxburst, dst_maxburst);
2362
2363 return 0;
2364}
2365
2366static int d40_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
2367 unsigned long arg)
2368{
2369 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2370
2371 if (d40c->phy_chan == NULL) {
2372 chan_err(d40c, "Channel is not allocated!\n");
2373 return -EINVAL;
2374 }
2375
2376 switch (cmd) {
2377 case DMA_TERMINATE_ALL:
2378 return d40_terminate_all(d40c);
2379 case DMA_PAUSE:
2380 return d40_pause(d40c);
2381 case DMA_RESUME:
2382 return d40_resume(d40c);
2383 case DMA_SLAVE_CONFIG:
2384 return d40_set_runtime_config(chan,
2385 (struct dma_slave_config *) arg);
2386 default:
2387 break;
2388 }
2389
2390 /* Other commands are unimplemented */
2391 return -ENXIO;
2392}
2393
2394/* Initialization functions */
2395
2396static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2397 struct d40_chan *chans, int offset,
2398 int num_chans)
2399{
2400 int i = 0;
2401 struct d40_chan *d40c;
2402
2403 INIT_LIST_HEAD(&dma->channels);
2404
2405 for (i = offset; i < offset + num_chans; i++) {
2406 d40c = &chans[i];
2407 d40c->base = base;
2408 d40c->chan.device = dma;
2409
2410 spin_lock_init(&d40c->lock);
2411
2412 d40c->log_num = D40_PHY_CHAN;
2413
2414 INIT_LIST_HEAD(&d40c->active);
2415 INIT_LIST_HEAD(&d40c->queue);
2416 INIT_LIST_HEAD(&d40c->pending_queue);
2417 INIT_LIST_HEAD(&d40c->client);
2418 INIT_LIST_HEAD(&d40c->prepare_queue);
2419
2420 tasklet_init(&d40c->tasklet, dma_tasklet,
2421 (unsigned long) d40c);
2422
2423 list_add_tail(&d40c->chan.device_node,
2424 &dma->channels);
2425 }
2426}
2427
2428static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2429{
2430 if (dma_has_cap(DMA_SLAVE, dev->cap_mask))
2431 dev->device_prep_slave_sg = d40_prep_slave_sg;
2432
2433 if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2434 dev->device_prep_dma_memcpy = d40_prep_memcpy;
2435
2436 /*
2437 * This controller can only access address at even
2438 * 32bit boundaries, i.e. 2^2
2439 */
2440 dev->copy_align = 2;
2441 }
2442
2443 if (dma_has_cap(DMA_SG, dev->cap_mask))
2444 dev->device_prep_dma_sg = d40_prep_memcpy_sg;
2445
2446 if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2447 dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2448
2449 dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2450 dev->device_free_chan_resources = d40_free_chan_resources;
2451 dev->device_issue_pending = d40_issue_pending;
2452 dev->device_tx_status = d40_tx_status;
2453 dev->device_control = d40_control;
2454 dev->dev = base->dev;
2455}
2456
2457static int __init d40_dmaengine_init(struct d40_base *base,
2458 int num_reserved_chans)
2459{
2460 int err ;
2461
2462 d40_chan_init(base, &base->dma_slave, base->log_chans,
2463 0, base->num_log_chans);
2464
2465 dma_cap_zero(base->dma_slave.cap_mask);
2466 dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2467 dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2468
2469 d40_ops_init(base, &base->dma_slave);
2470
2471 err = dma_async_device_register(&base->dma_slave);
2472
2473 if (err) {
2474 d40_err(base->dev, "Failed to register slave channels\n");
2475 goto failure1;
2476 }
2477
2478 d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2479 base->num_log_chans, base->plat_data->memcpy_len);
2480
2481 dma_cap_zero(base->dma_memcpy.cap_mask);
2482 dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2483 dma_cap_set(DMA_SG, base->dma_memcpy.cap_mask);
2484
2485 d40_ops_init(base, &base->dma_memcpy);
2486
2487 err = dma_async_device_register(&base->dma_memcpy);
2488
2489 if (err) {
2490 d40_err(base->dev,
2491 "Failed to regsiter memcpy only channels\n");
2492 goto failure2;
2493 }
2494
2495 d40_chan_init(base, &base->dma_both, base->phy_chans,
2496 0, num_reserved_chans);
2497
2498 dma_cap_zero(base->dma_both.cap_mask);
2499 dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2500 dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2501 dma_cap_set(DMA_SG, base->dma_both.cap_mask);
2502 dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2503
2504 d40_ops_init(base, &base->dma_both);
2505 err = dma_async_device_register(&base->dma_both);
2506
2507 if (err) {
2508 d40_err(base->dev,
2509 "Failed to register logical and physical capable channels\n");
2510 goto failure3;
2511 }
2512 return 0;
2513failure3:
2514 dma_async_device_unregister(&base->dma_memcpy);
2515failure2:
2516 dma_async_device_unregister(&base->dma_slave);
2517failure1:
2518 return err;
2519}
2520
2521/* Initialization functions. */
2522
2523static int __init d40_phy_res_init(struct d40_base *base)
2524{
2525 int i;
2526 int num_phy_chans_avail = 0;
2527 u32 val[2];
2528 int odd_even_bit = -2;
2529
2530 val[0] = readl(base->virtbase + D40_DREG_PRSME);
2531 val[1] = readl(base->virtbase + D40_DREG_PRSMO);
2532
2533 for (i = 0; i < base->num_phy_chans; i++) {
2534 base->phy_res[i].num = i;
2535 odd_even_bit += 2 * ((i % 2) == 0);
2536 if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
2537 /* Mark security only channels as occupied */
2538 base->phy_res[i].allocated_src = D40_ALLOC_PHY;
2539 base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
2540 } else {
2541 base->phy_res[i].allocated_src = D40_ALLOC_FREE;
2542 base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
2543 num_phy_chans_avail++;
2544 }
2545 spin_lock_init(&base->phy_res[i].lock);
2546 }
2547
2548 /* Mark disabled channels as occupied */
2549 for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
2550 int chan = base->plat_data->disabled_channels[i];
2551
2552 base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
2553 base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
2554 num_phy_chans_avail--;
2555 }
2556
2557 dev_info(base->dev, "%d of %d physical DMA channels available\n",
2558 num_phy_chans_avail, base->num_phy_chans);
2559
2560 /* Verify settings extended vs standard */
2561 val[0] = readl(base->virtbase + D40_DREG_PRTYP);
2562
2563 for (i = 0; i < base->num_phy_chans; i++) {
2564
2565 if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
2566 (val[0] & 0x3) != 1)
2567 dev_info(base->dev,
2568 "[%s] INFO: channel %d is misconfigured (%d)\n",
2569 __func__, i, val[0] & 0x3);
2570
2571 val[0] = val[0] >> 2;
2572 }
2573
2574 return num_phy_chans_avail;
2575}
2576
2577static struct d40_base * __init d40_hw_detect_init(struct platform_device *pdev)
2578{
2579 struct stedma40_platform_data *plat_data;
2580 struct clk *clk = NULL;
2581 void __iomem *virtbase = NULL;
2582 struct resource *res = NULL;
2583 struct d40_base *base = NULL;
2584 int num_log_chans = 0;
2585 int num_phy_chans;
2586 int i;
2587 u32 pid;
2588 u32 cid;
2589 u8 rev;
2590
2591 clk = clk_get(&pdev->dev, NULL);
2592
2593 if (IS_ERR(clk)) {
2594 d40_err(&pdev->dev, "No matching clock found\n");
2595 goto failure;
2596 }
2597
2598 clk_enable(clk);
2599
2600 /* Get IO for DMAC base address */
2601 res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "base");
2602 if (!res)
2603 goto failure;
2604
2605 if (request_mem_region(res->start, resource_size(res),
2606 D40_NAME " I/O base") == NULL)
2607 goto failure;
2608
2609 virtbase = ioremap(res->start, resource_size(res));
2610 if (!virtbase)
2611 goto failure;
2612
2613 /* This is just a regular AMBA PrimeCell ID actually */
2614 for (pid = 0, i = 0; i < 4; i++)
2615 pid |= (readl(virtbase + resource_size(res) - 0x20 + 4 * i)
2616 & 255) << (i * 8);
2617 for (cid = 0, i = 0; i < 4; i++)
2618 cid |= (readl(virtbase + resource_size(res) - 0x10 + 4 * i)
2619 & 255) << (i * 8);
2620
2621 if (cid != AMBA_CID) {
2622 d40_err(&pdev->dev, "Unknown hardware! No PrimeCell ID\n");
2623 goto failure;
2624 }
2625 if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
2626 d40_err(&pdev->dev, "Unknown designer! Got %x wanted %x\n",
2627 AMBA_MANF_BITS(pid),
2628 AMBA_VENDOR_ST);
2629 goto failure;
2630 }
2631 /*
2632 * HW revision:
2633 * DB8500ed has revision 0
2634 * ? has revision 1
2635 * DB8500v1 has revision 2
2636 * DB8500v2 has revision 3
2637 */
2638 rev = AMBA_REV_BITS(pid);
2639
2640 /* The number of physical channels on this HW */
2641 num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
2642
2643 dev_info(&pdev->dev, "hardware revision: %d @ 0x%x\n",
2644 rev, res->start);
2645
2646 plat_data = pdev->dev.platform_data;
2647
2648 /* Count the number of logical channels in use */
2649 for (i = 0; i < plat_data->dev_len; i++)
2650 if (plat_data->dev_rx[i] != 0)
2651 num_log_chans++;
2652
2653 for (i = 0; i < plat_data->dev_len; i++)
2654 if (plat_data->dev_tx[i] != 0)
2655 num_log_chans++;
2656
2657 base = kzalloc(ALIGN(sizeof(struct d40_base), 4) +
2658 (num_phy_chans + num_log_chans + plat_data->memcpy_len) *
2659 sizeof(struct d40_chan), GFP_KERNEL);
2660
2661 if (base == NULL) {
2662 d40_err(&pdev->dev, "Out of memory\n");
2663 goto failure;
2664 }
2665
2666 base->rev = rev;
2667 base->clk = clk;
2668 base->num_phy_chans = num_phy_chans;
2669 base->num_log_chans = num_log_chans;
2670 base->phy_start = res->start;
2671 base->phy_size = resource_size(res);
2672 base->virtbase = virtbase;
2673 base->plat_data = plat_data;
2674 base->dev = &pdev->dev;
2675 base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
2676 base->log_chans = &base->phy_chans[num_phy_chans];
2677
2678 base->phy_res = kzalloc(num_phy_chans * sizeof(struct d40_phy_res),
2679 GFP_KERNEL);
2680 if (!base->phy_res)
2681 goto failure;
2682
2683 base->lookup_phy_chans = kzalloc(num_phy_chans *
2684 sizeof(struct d40_chan *),
2685 GFP_KERNEL);
2686 if (!base->lookup_phy_chans)
2687 goto failure;
2688
2689 if (num_log_chans + plat_data->memcpy_len) {
2690 /*
2691 * The max number of logical channels are event lines for all
2692 * src devices and dst devices
2693 */
2694 base->lookup_log_chans = kzalloc(plat_data->dev_len * 2 *
2695 sizeof(struct d40_chan *),
2696 GFP_KERNEL);
2697 if (!base->lookup_log_chans)
2698 goto failure;
2699 }
2700
2701 base->lcla_pool.alloc_map = kzalloc(num_phy_chans *
2702 sizeof(struct d40_desc *) *
2703 D40_LCLA_LINK_PER_EVENT_GRP,
2704 GFP_KERNEL);
2705 if (!base->lcla_pool.alloc_map)
2706 goto failure;
2707
2708 base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
2709 0, SLAB_HWCACHE_ALIGN,
2710 NULL);
2711 if (base->desc_slab == NULL)
2712 goto failure;
2713
2714 return base;
2715
2716failure:
2717 if (!IS_ERR(clk)) {
2718 clk_disable(clk);
2719 clk_put(clk);
2720 }
2721 if (virtbase)
2722 iounmap(virtbase);
2723 if (res)
2724 release_mem_region(res->start,
2725 resource_size(res));
2726 if (virtbase)
2727 iounmap(virtbase);
2728
2729 if (base) {
2730 kfree(base->lcla_pool.alloc_map);
2731 kfree(base->lookup_log_chans);
2732 kfree(base->lookup_phy_chans);
2733 kfree(base->phy_res);
2734 kfree(base);
2735 }
2736
2737 return NULL;
2738}
2739
2740static void __init d40_hw_init(struct d40_base *base)
2741{
2742
2743 static const struct d40_reg_val dma_init_reg[] = {
2744 /* Clock every part of the DMA block from start */
2745 { .reg = D40_DREG_GCC, .val = 0x0000ff01},
2746
2747 /* Interrupts on all logical channels */
2748 { .reg = D40_DREG_LCMIS0, .val = 0xFFFFFFFF},
2749 { .reg = D40_DREG_LCMIS1, .val = 0xFFFFFFFF},
2750 { .reg = D40_DREG_LCMIS2, .val = 0xFFFFFFFF},
2751 { .reg = D40_DREG_LCMIS3, .val = 0xFFFFFFFF},
2752 { .reg = D40_DREG_LCICR0, .val = 0xFFFFFFFF},
2753 { .reg = D40_DREG_LCICR1, .val = 0xFFFFFFFF},
2754 { .reg = D40_DREG_LCICR2, .val = 0xFFFFFFFF},
2755 { .reg = D40_DREG_LCICR3, .val = 0xFFFFFFFF},
2756 { .reg = D40_DREG_LCTIS0, .val = 0xFFFFFFFF},
2757 { .reg = D40_DREG_LCTIS1, .val = 0xFFFFFFFF},
2758 { .reg = D40_DREG_LCTIS2, .val = 0xFFFFFFFF},
2759 { .reg = D40_DREG_LCTIS3, .val = 0xFFFFFFFF}
2760 };
2761 int i;
2762 u32 prmseo[2] = {0, 0};
2763 u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
2764 u32 pcmis = 0;
2765 u32 pcicr = 0;
2766
2767 for (i = 0; i < ARRAY_SIZE(dma_init_reg); i++)
2768 writel(dma_init_reg[i].val,
2769 base->virtbase + dma_init_reg[i].reg);
2770
2771 /* Configure all our dma channels to default settings */
2772 for (i = 0; i < base->num_phy_chans; i++) {
2773
2774 activeo[i % 2] = activeo[i % 2] << 2;
2775
2776 if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
2777 == D40_ALLOC_PHY) {
2778 activeo[i % 2] |= 3;
2779 continue;
2780 }
2781
2782 /* Enable interrupt # */
2783 pcmis = (pcmis << 1) | 1;
2784
2785 /* Clear interrupt # */
2786 pcicr = (pcicr << 1) | 1;
2787
2788 /* Set channel to physical mode */
2789 prmseo[i % 2] = prmseo[i % 2] << 2;
2790 prmseo[i % 2] |= 1;
2791
2792 }
2793
2794 writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
2795 writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
2796 writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
2797 writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
2798
2799 /* Write which interrupt to enable */
2800 writel(pcmis, base->virtbase + D40_DREG_PCMIS);
2801
2802 /* Write which interrupt to clear */
2803 writel(pcicr, base->virtbase + D40_DREG_PCICR);
2804
2805}
2806
2807static int __init d40_lcla_allocate(struct d40_base *base)
2808{
2809 struct d40_lcla_pool *pool = &base->lcla_pool;
2810 unsigned long *page_list;
2811 int i, j;
2812 int ret = 0;
2813
2814 /*
2815 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
2816 * To full fill this hardware requirement without wasting 256 kb
2817 * we allocate pages until we get an aligned one.
2818 */
2819 page_list = kmalloc(sizeof(unsigned long) * MAX_LCLA_ALLOC_ATTEMPTS,
2820 GFP_KERNEL);
2821
2822 if (!page_list) {
2823 ret = -ENOMEM;
2824 goto failure;
2825 }
2826
2827 /* Calculating how many pages that are required */
2828 base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
2829
2830 for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
2831 page_list[i] = __get_free_pages(GFP_KERNEL,
2832 base->lcla_pool.pages);
2833 if (!page_list[i]) {
2834
2835 d40_err(base->dev, "Failed to allocate %d pages.\n",
2836 base->lcla_pool.pages);
2837
2838 for (j = 0; j < i; j++)
2839 free_pages(page_list[j], base->lcla_pool.pages);
2840 goto failure;
2841 }
2842
2843 if ((virt_to_phys((void *)page_list[i]) &
2844 (LCLA_ALIGNMENT - 1)) == 0)
2845 break;
2846 }
2847
2848 for (j = 0; j < i; j++)
2849 free_pages(page_list[j], base->lcla_pool.pages);
2850
2851 if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
2852 base->lcla_pool.base = (void *)page_list[i];
2853 } else {
2854 /*
2855 * After many attempts and no succees with finding the correct
2856 * alignment, try with allocating a big buffer.
2857 */
2858 dev_warn(base->dev,
2859 "[%s] Failed to get %d pages @ 18 bit align.\n",
2860 __func__, base->lcla_pool.pages);
2861 base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
2862 base->num_phy_chans +
2863 LCLA_ALIGNMENT,
2864 GFP_KERNEL);
2865 if (!base->lcla_pool.base_unaligned) {
2866 ret = -ENOMEM;
2867 goto failure;
2868 }
2869
2870 base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
2871 LCLA_ALIGNMENT);
2872 }
2873
2874 pool->dma_addr = dma_map_single(base->dev, pool->base,
2875 SZ_1K * base->num_phy_chans,
2876 DMA_TO_DEVICE);
2877 if (dma_mapping_error(base->dev, pool->dma_addr)) {
2878 pool->dma_addr = 0;
2879 ret = -ENOMEM;
2880 goto failure;
2881 }
2882
2883 writel(virt_to_phys(base->lcla_pool.base),
2884 base->virtbase + D40_DREG_LCLA);
2885failure:
2886 kfree(page_list);
2887 return ret;
2888}
2889
2890static int __init d40_probe(struct platform_device *pdev)
2891{
2892 int err;
2893 int ret = -ENOENT;
2894 struct d40_base *base;
2895 struct resource *res = NULL;
2896 int num_reserved_chans;
2897 u32 val;
2898
2899 base = d40_hw_detect_init(pdev);
2900
2901 if (!base)
2902 goto failure;
2903
2904 num_reserved_chans = d40_phy_res_init(base);
2905
2906 platform_set_drvdata(pdev, base);
2907
2908 spin_lock_init(&base->interrupt_lock);
2909 spin_lock_init(&base->execmd_lock);
2910
2911 /* Get IO for logical channel parameter address */
2912 res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "lcpa");
2913 if (!res) {
2914 ret = -ENOENT;
2915 d40_err(&pdev->dev, "No \"lcpa\" memory resource\n");
2916 goto failure;
2917 }
2918 base->lcpa_size = resource_size(res);
2919 base->phy_lcpa = res->start;
2920
2921 if (request_mem_region(res->start, resource_size(res),
2922 D40_NAME " I/O lcpa") == NULL) {
2923 ret = -EBUSY;
2924 d40_err(&pdev->dev,
2925 "Failed to request LCPA region 0x%x-0x%x\n",
2926 res->start, res->end);
2927 goto failure;
2928 }
2929
2930 /* We make use of ESRAM memory for this. */
2931 val = readl(base->virtbase + D40_DREG_LCPA);
2932 if (res->start != val && val != 0) {
2933 dev_warn(&pdev->dev,
2934 "[%s] Mismatch LCPA dma 0x%x, def 0x%x\n",
2935 __func__, val, res->start);
2936 } else
2937 writel(res->start, base->virtbase + D40_DREG_LCPA);
2938
2939 base->lcpa_base = ioremap(res->start, resource_size(res));
2940 if (!base->lcpa_base) {
2941 ret = -ENOMEM;
2942 d40_err(&pdev->dev, "Failed to ioremap LCPA region\n");
2943 goto failure;
2944 }
2945
2946 ret = d40_lcla_allocate(base);
2947 if (ret) {
2948 d40_err(&pdev->dev, "Failed to allocate LCLA area\n");
2949 goto failure;
2950 }
2951
2952 spin_lock_init(&base->lcla_pool.lock);
2953
2954 base->irq = platform_get_irq(pdev, 0);
2955
2956 ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
2957 if (ret) {
2958 d40_err(&pdev->dev, "No IRQ defined\n");
2959 goto failure;
2960 }
2961
2962 err = d40_dmaengine_init(base, num_reserved_chans);
2963 if (err)
2964 goto failure;
2965
2966 d40_hw_init(base);
2967
2968 dev_info(base->dev, "initialized\n");
2969 return 0;
2970
2971failure:
2972 if (base) {
2973 if (base->desc_slab)
2974 kmem_cache_destroy(base->desc_slab);
2975 if (base->virtbase)
2976 iounmap(base->virtbase);
2977
2978 if (base->lcla_pool.dma_addr)
2979 dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
2980 SZ_1K * base->num_phy_chans,
2981 DMA_TO_DEVICE);
2982
2983 if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
2984 free_pages((unsigned long)base->lcla_pool.base,
2985 base->lcla_pool.pages);
2986
2987 kfree(base->lcla_pool.base_unaligned);
2988
2989 if (base->phy_lcpa)
2990 release_mem_region(base->phy_lcpa,
2991 base->lcpa_size);
2992 if (base->phy_start)
2993 release_mem_region(base->phy_start,
2994 base->phy_size);
2995 if (base->clk) {
2996 clk_disable(base->clk);
2997 clk_put(base->clk);
2998 }
2999
3000 kfree(base->lcla_pool.alloc_map);
3001 kfree(base->lookup_log_chans);
3002 kfree(base->lookup_phy_chans);
3003 kfree(base->phy_res);
3004 kfree(base);
3005 }
3006
3007 d40_err(&pdev->dev, "probe failed\n");
3008 return ret;
3009}
3010
3011static struct platform_driver d40_driver = {
3012 .driver = {
3013 .owner = THIS_MODULE,
3014 .name = D40_NAME,
3015 },
3016};
3017
3018static int __init stedma40_init(void)
3019{
3020 return platform_driver_probe(&d40_driver, d40_probe);
3021}
3022subsys_initcall(stedma40_init);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) Ericsson AB 2007-2008
4 * Copyright (C) ST-Ericsson SA 2008-2010
5 * Author: Per Forlin <per.forlin@stericsson.com> for ST-Ericsson
6 * Author: Jonas Aaberg <jonas.aberg@stericsson.com> for ST-Ericsson
7 */
8
9#include <linux/dma-mapping.h>
10#include <linux/kernel.h>
11#include <linux/slab.h>
12#include <linux/export.h>
13#include <linux/dmaengine.h>
14#include <linux/platform_device.h>
15#include <linux/clk.h>
16#include <linux/delay.h>
17#include <linux/log2.h>
18#include <linux/pm.h>
19#include <linux/pm_runtime.h>
20#include <linux/err.h>
21#include <linux/of.h>
22#include <linux/of_address.h>
23#include <linux/of_dma.h>
24#include <linux/amba/bus.h>
25#include <linux/regulator/consumer.h>
26
27#include "dmaengine.h"
28#include "ste_dma40.h"
29#include "ste_dma40_ll.h"
30
31/**
32 * struct stedma40_platform_data - Configuration struct for the dma device.
33 *
34 * @disabled_channels: A vector, ending with -1, that marks physical channels
35 * that are for different reasons not available for the driver.
36 * @soft_lli_chans: A vector, that marks physical channels will use LLI by SW
37 * which avoids HW bug that exists in some versions of the controller.
38 * SoftLLI introduces relink overhead that could impact performance for
39 * certain use cases.
40 * @num_of_soft_lli_chans: The number of channels that needs to be configured
41 * to use SoftLLI.
42 * @use_esram_lcla: flag for mapping the lcla into esram region
43 * @num_of_memcpy_chans: The number of channels reserved for memcpy.
44 * @num_of_phy_chans: The number of physical channels implemented in HW.
45 * 0 means reading the number of channels from DMA HW but this is only valid
46 * for 'multiple of 4' channels, like 8.
47 */
48struct stedma40_platform_data {
49 int disabled_channels[STEDMA40_MAX_PHYS];
50 int *soft_lli_chans;
51 int num_of_soft_lli_chans;
52 bool use_esram_lcla;
53 int num_of_memcpy_chans;
54 int num_of_phy_chans;
55};
56
57#define D40_NAME "dma40"
58
59#define D40_PHY_CHAN -1
60
61/* For masking out/in 2 bit channel positions */
62#define D40_CHAN_POS(chan) (2 * (chan / 2))
63#define D40_CHAN_POS_MASK(chan) (0x3 << D40_CHAN_POS(chan))
64
65/* Maximum iterations taken before giving up suspending a channel */
66#define D40_SUSPEND_MAX_IT 500
67
68/* Milliseconds */
69#define DMA40_AUTOSUSPEND_DELAY 100
70
71/* Hardware requirement on LCLA alignment */
72#define LCLA_ALIGNMENT 0x40000
73
74/* Max number of links per event group */
75#define D40_LCLA_LINK_PER_EVENT_GRP 128
76#define D40_LCLA_END D40_LCLA_LINK_PER_EVENT_GRP
77
78/* Max number of logical channels per physical channel */
79#define D40_MAX_LOG_CHAN_PER_PHY 32
80
81/* Attempts before giving up to trying to get pages that are aligned */
82#define MAX_LCLA_ALLOC_ATTEMPTS 256
83
84/* Bit markings for allocation map */
85#define D40_ALLOC_FREE BIT(31)
86#define D40_ALLOC_PHY BIT(30)
87#define D40_ALLOC_LOG_FREE 0
88
89#define D40_MEMCPY_MAX_CHANS 8
90
91/* Reserved event lines for memcpy only. */
92#define DB8500_DMA_MEMCPY_EV_0 51
93#define DB8500_DMA_MEMCPY_EV_1 56
94#define DB8500_DMA_MEMCPY_EV_2 57
95#define DB8500_DMA_MEMCPY_EV_3 58
96#define DB8500_DMA_MEMCPY_EV_4 59
97#define DB8500_DMA_MEMCPY_EV_5 60
98
99static int dma40_memcpy_channels[] = {
100 DB8500_DMA_MEMCPY_EV_0,
101 DB8500_DMA_MEMCPY_EV_1,
102 DB8500_DMA_MEMCPY_EV_2,
103 DB8500_DMA_MEMCPY_EV_3,
104 DB8500_DMA_MEMCPY_EV_4,
105 DB8500_DMA_MEMCPY_EV_5,
106};
107
108/* Default configuration for physical memcpy */
109static const struct stedma40_chan_cfg dma40_memcpy_conf_phy = {
110 .mode = STEDMA40_MODE_PHYSICAL,
111 .dir = DMA_MEM_TO_MEM,
112
113 .src_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
114 .src_info.psize = STEDMA40_PSIZE_PHY_1,
115 .src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
116
117 .dst_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
118 .dst_info.psize = STEDMA40_PSIZE_PHY_1,
119 .dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
120};
121
122/* Default configuration for logical memcpy */
123static const struct stedma40_chan_cfg dma40_memcpy_conf_log = {
124 .mode = STEDMA40_MODE_LOGICAL,
125 .dir = DMA_MEM_TO_MEM,
126
127 .src_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
128 .src_info.psize = STEDMA40_PSIZE_LOG_1,
129 .src_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
130
131 .dst_info.data_width = DMA_SLAVE_BUSWIDTH_1_BYTE,
132 .dst_info.psize = STEDMA40_PSIZE_LOG_1,
133 .dst_info.flow_ctrl = STEDMA40_NO_FLOW_CTRL,
134};
135
136/**
137 * enum d40_command - The different commands and/or statuses.
138 *
139 * @D40_DMA_STOP: DMA channel command STOP or status STOPPED,
140 * @D40_DMA_RUN: The DMA channel is RUNNING of the command RUN.
141 * @D40_DMA_SUSPEND_REQ: Request the DMA to SUSPEND as soon as possible.
142 * @D40_DMA_SUSPENDED: The DMA channel is SUSPENDED.
143 */
144enum d40_command {
145 D40_DMA_STOP = 0,
146 D40_DMA_RUN = 1,
147 D40_DMA_SUSPEND_REQ = 2,
148 D40_DMA_SUSPENDED = 3
149};
150
151/*
152 * enum d40_events - The different Event Enables for the event lines.
153 *
154 * @D40_DEACTIVATE_EVENTLINE: De-activate Event line, stopping the logical chan.
155 * @D40_ACTIVATE_EVENTLINE: Activate the Event line, to start a logical chan.
156 * @D40_SUSPEND_REQ_EVENTLINE: Requesting for suspending a event line.
157 * @D40_ROUND_EVENTLINE: Status check for event line.
158 */
159
160enum d40_events {
161 D40_DEACTIVATE_EVENTLINE = 0,
162 D40_ACTIVATE_EVENTLINE = 1,
163 D40_SUSPEND_REQ_EVENTLINE = 2,
164 D40_ROUND_EVENTLINE = 3
165};
166
167/*
168 * These are the registers that has to be saved and later restored
169 * when the DMA hw is powered off.
170 * TODO: Add save/restore of D40_DREG_GCC on dma40 v3 or later, if that works.
171 */
172static __maybe_unused u32 d40_backup_regs[] = {
173 D40_DREG_LCPA,
174 D40_DREG_LCLA,
175 D40_DREG_PRMSE,
176 D40_DREG_PRMSO,
177 D40_DREG_PRMOE,
178 D40_DREG_PRMOO,
179};
180
181#define BACKUP_REGS_SZ ARRAY_SIZE(d40_backup_regs)
182
183/*
184 * since 9540 and 8540 has the same HW revision
185 * use v4a for 9540 or earlier
186 * use v4b for 8540 or later
187 * HW revision:
188 * DB8500ed has revision 0
189 * DB8500v1 has revision 2
190 * DB8500v2 has revision 3
191 * AP9540v1 has revision 4
192 * DB8540v1 has revision 4
193 * TODO: Check if all these registers have to be saved/restored on dma40 v4a
194 */
195static u32 d40_backup_regs_v4a[] = {
196 D40_DREG_PSEG1,
197 D40_DREG_PSEG2,
198 D40_DREG_PSEG3,
199 D40_DREG_PSEG4,
200 D40_DREG_PCEG1,
201 D40_DREG_PCEG2,
202 D40_DREG_PCEG3,
203 D40_DREG_PCEG4,
204 D40_DREG_RSEG1,
205 D40_DREG_RSEG2,
206 D40_DREG_RSEG3,
207 D40_DREG_RSEG4,
208 D40_DREG_RCEG1,
209 D40_DREG_RCEG2,
210 D40_DREG_RCEG3,
211 D40_DREG_RCEG4,
212};
213
214#define BACKUP_REGS_SZ_V4A ARRAY_SIZE(d40_backup_regs_v4a)
215
216static u32 d40_backup_regs_v4b[] = {
217 D40_DREG_CPSEG1,
218 D40_DREG_CPSEG2,
219 D40_DREG_CPSEG3,
220 D40_DREG_CPSEG4,
221 D40_DREG_CPSEG5,
222 D40_DREG_CPCEG1,
223 D40_DREG_CPCEG2,
224 D40_DREG_CPCEG3,
225 D40_DREG_CPCEG4,
226 D40_DREG_CPCEG5,
227 D40_DREG_CRSEG1,
228 D40_DREG_CRSEG2,
229 D40_DREG_CRSEG3,
230 D40_DREG_CRSEG4,
231 D40_DREG_CRSEG5,
232 D40_DREG_CRCEG1,
233 D40_DREG_CRCEG2,
234 D40_DREG_CRCEG3,
235 D40_DREG_CRCEG4,
236 D40_DREG_CRCEG5,
237};
238
239#define BACKUP_REGS_SZ_V4B ARRAY_SIZE(d40_backup_regs_v4b)
240
241static __maybe_unused u32 d40_backup_regs_chan[] = {
242 D40_CHAN_REG_SSCFG,
243 D40_CHAN_REG_SSELT,
244 D40_CHAN_REG_SSPTR,
245 D40_CHAN_REG_SSLNK,
246 D40_CHAN_REG_SDCFG,
247 D40_CHAN_REG_SDELT,
248 D40_CHAN_REG_SDPTR,
249 D40_CHAN_REG_SDLNK,
250};
251
252#define BACKUP_REGS_SZ_MAX ((BACKUP_REGS_SZ_V4A > BACKUP_REGS_SZ_V4B) ? \
253 BACKUP_REGS_SZ_V4A : BACKUP_REGS_SZ_V4B)
254
255/**
256 * struct d40_interrupt_lookup - lookup table for interrupt handler
257 *
258 * @src: Interrupt mask register.
259 * @clr: Interrupt clear register.
260 * @is_error: true if this is an error interrupt.
261 * @offset: start delta in the lookup_log_chans in d40_base. If equals to
262 * D40_PHY_CHAN, the lookup_phy_chans shall be used instead.
263 */
264struct d40_interrupt_lookup {
265 u32 src;
266 u32 clr;
267 bool is_error;
268 int offset;
269};
270
271
272static struct d40_interrupt_lookup il_v4a[] = {
273 {D40_DREG_LCTIS0, D40_DREG_LCICR0, false, 0},
274 {D40_DREG_LCTIS1, D40_DREG_LCICR1, false, 32},
275 {D40_DREG_LCTIS2, D40_DREG_LCICR2, false, 64},
276 {D40_DREG_LCTIS3, D40_DREG_LCICR3, false, 96},
277 {D40_DREG_LCEIS0, D40_DREG_LCICR0, true, 0},
278 {D40_DREG_LCEIS1, D40_DREG_LCICR1, true, 32},
279 {D40_DREG_LCEIS2, D40_DREG_LCICR2, true, 64},
280 {D40_DREG_LCEIS3, D40_DREG_LCICR3, true, 96},
281 {D40_DREG_PCTIS, D40_DREG_PCICR, false, D40_PHY_CHAN},
282 {D40_DREG_PCEIS, D40_DREG_PCICR, true, D40_PHY_CHAN},
283};
284
285static struct d40_interrupt_lookup il_v4b[] = {
286 {D40_DREG_CLCTIS1, D40_DREG_CLCICR1, false, 0},
287 {D40_DREG_CLCTIS2, D40_DREG_CLCICR2, false, 32},
288 {D40_DREG_CLCTIS3, D40_DREG_CLCICR3, false, 64},
289 {D40_DREG_CLCTIS4, D40_DREG_CLCICR4, false, 96},
290 {D40_DREG_CLCTIS5, D40_DREG_CLCICR5, false, 128},
291 {D40_DREG_CLCEIS1, D40_DREG_CLCICR1, true, 0},
292 {D40_DREG_CLCEIS2, D40_DREG_CLCICR2, true, 32},
293 {D40_DREG_CLCEIS3, D40_DREG_CLCICR3, true, 64},
294 {D40_DREG_CLCEIS4, D40_DREG_CLCICR4, true, 96},
295 {D40_DREG_CLCEIS5, D40_DREG_CLCICR5, true, 128},
296 {D40_DREG_CPCTIS, D40_DREG_CPCICR, false, D40_PHY_CHAN},
297 {D40_DREG_CPCEIS, D40_DREG_CPCICR, true, D40_PHY_CHAN},
298};
299
300/**
301 * struct d40_reg_val - simple lookup struct
302 *
303 * @reg: The register.
304 * @val: The value that belongs to the register in reg.
305 */
306struct d40_reg_val {
307 unsigned int reg;
308 unsigned int val;
309};
310
311static __initdata struct d40_reg_val dma_init_reg_v4a[] = {
312 /* Clock every part of the DMA block from start */
313 { .reg = D40_DREG_GCC, .val = D40_DREG_GCC_ENABLE_ALL},
314
315 /* Interrupts on all logical channels */
316 { .reg = D40_DREG_LCMIS0, .val = 0xFFFFFFFF},
317 { .reg = D40_DREG_LCMIS1, .val = 0xFFFFFFFF},
318 { .reg = D40_DREG_LCMIS2, .val = 0xFFFFFFFF},
319 { .reg = D40_DREG_LCMIS3, .val = 0xFFFFFFFF},
320 { .reg = D40_DREG_LCICR0, .val = 0xFFFFFFFF},
321 { .reg = D40_DREG_LCICR1, .val = 0xFFFFFFFF},
322 { .reg = D40_DREG_LCICR2, .val = 0xFFFFFFFF},
323 { .reg = D40_DREG_LCICR3, .val = 0xFFFFFFFF},
324 { .reg = D40_DREG_LCTIS0, .val = 0xFFFFFFFF},
325 { .reg = D40_DREG_LCTIS1, .val = 0xFFFFFFFF},
326 { .reg = D40_DREG_LCTIS2, .val = 0xFFFFFFFF},
327 { .reg = D40_DREG_LCTIS3, .val = 0xFFFFFFFF}
328};
329static __initdata struct d40_reg_val dma_init_reg_v4b[] = {
330 /* Clock every part of the DMA block from start */
331 { .reg = D40_DREG_GCC, .val = D40_DREG_GCC_ENABLE_ALL},
332
333 /* Interrupts on all logical channels */
334 { .reg = D40_DREG_CLCMIS1, .val = 0xFFFFFFFF},
335 { .reg = D40_DREG_CLCMIS2, .val = 0xFFFFFFFF},
336 { .reg = D40_DREG_CLCMIS3, .val = 0xFFFFFFFF},
337 { .reg = D40_DREG_CLCMIS4, .val = 0xFFFFFFFF},
338 { .reg = D40_DREG_CLCMIS5, .val = 0xFFFFFFFF},
339 { .reg = D40_DREG_CLCICR1, .val = 0xFFFFFFFF},
340 { .reg = D40_DREG_CLCICR2, .val = 0xFFFFFFFF},
341 { .reg = D40_DREG_CLCICR3, .val = 0xFFFFFFFF},
342 { .reg = D40_DREG_CLCICR4, .val = 0xFFFFFFFF},
343 { .reg = D40_DREG_CLCICR5, .val = 0xFFFFFFFF},
344 { .reg = D40_DREG_CLCTIS1, .val = 0xFFFFFFFF},
345 { .reg = D40_DREG_CLCTIS2, .val = 0xFFFFFFFF},
346 { .reg = D40_DREG_CLCTIS3, .val = 0xFFFFFFFF},
347 { .reg = D40_DREG_CLCTIS4, .val = 0xFFFFFFFF},
348 { .reg = D40_DREG_CLCTIS5, .val = 0xFFFFFFFF}
349};
350
351/**
352 * struct d40_lli_pool - Structure for keeping LLIs in memory
353 *
354 * @base: Pointer to memory area when the pre_alloc_lli's are not large
355 * enough, IE bigger than the most common case, 1 dst and 1 src. NULL if
356 * pre_alloc_lli is used.
357 * @dma_addr: DMA address, if mapped
358 * @size: The size in bytes of the memory at base or the size of pre_alloc_lli.
359 * @pre_alloc_lli: Pre allocated area for the most common case of transfers,
360 * one buffer to one buffer.
361 */
362struct d40_lli_pool {
363 void *base;
364 int size;
365 dma_addr_t dma_addr;
366 /* Space for dst and src, plus an extra for padding */
367 u8 pre_alloc_lli[3 * sizeof(struct d40_phy_lli)];
368};
369
370/**
371 * struct d40_desc - A descriptor is one DMA job.
372 *
373 * @lli_phy: LLI settings for physical channel. Both src and dst=
374 * points into the lli_pool, to base if lli_len > 1 or to pre_alloc_lli if
375 * lli_len equals one.
376 * @lli_log: Same as above but for logical channels.
377 * @lli_pool: The pool with two entries pre-allocated.
378 * @lli_len: Number of llis of current descriptor.
379 * @lli_current: Number of transferred llis.
380 * @lcla_alloc: Number of LCLA entries allocated.
381 * @txd: DMA engine struct. Used for among other things for communication
382 * during a transfer.
383 * @node: List entry.
384 * @is_in_client_list: true if the client owns this descriptor.
385 * @cyclic: true if this is a cyclic job
386 *
387 * This descriptor is used for both logical and physical transfers.
388 */
389struct d40_desc {
390 /* LLI physical */
391 struct d40_phy_lli_bidir lli_phy;
392 /* LLI logical */
393 struct d40_log_lli_bidir lli_log;
394
395 struct d40_lli_pool lli_pool;
396 int lli_len;
397 int lli_current;
398 int lcla_alloc;
399
400 struct dma_async_tx_descriptor txd;
401 struct list_head node;
402
403 bool is_in_client_list;
404 bool cyclic;
405};
406
407/**
408 * struct d40_lcla_pool - LCLA pool settings and data.
409 *
410 * @base: The virtual address of LCLA. 18 bit aligned.
411 * @dma_addr: DMA address, if mapped
412 * @base_unaligned: The original kmalloc pointer, if kmalloc is used.
413 * This pointer is only there for clean-up on error.
414 * @pages: The number of pages needed for all physical channels.
415 * Only used later for clean-up on error
416 * @lock: Lock to protect the content in this struct.
417 * @alloc_map: big map over which LCLA entry is own by which job.
418 */
419struct d40_lcla_pool {
420 void *base;
421 dma_addr_t dma_addr;
422 void *base_unaligned;
423 int pages;
424 spinlock_t lock;
425 struct d40_desc **alloc_map;
426};
427
428/**
429 * struct d40_phy_res - struct for handling eventlines mapped to physical
430 * channels.
431 *
432 * @lock: A lock protection this entity.
433 * @reserved: True if used by secure world or otherwise.
434 * @num: The physical channel number of this entity.
435 * @allocated_src: Bit mapped to show which src event line's are mapped to
436 * this physical channel. Can also be free or physically allocated.
437 * @allocated_dst: Same as for src but is dst.
438 * allocated_dst and allocated_src uses the D40_ALLOC* defines as well as
439 * event line number.
440 * @use_soft_lli: To mark if the linked lists of channel are managed by SW.
441 */
442struct d40_phy_res {
443 spinlock_t lock;
444 bool reserved;
445 int num;
446 u32 allocated_src;
447 u32 allocated_dst;
448 bool use_soft_lli;
449};
450
451struct d40_base;
452
453/**
454 * struct d40_chan - Struct that describes a channel.
455 *
456 * @lock: A spinlock to protect this struct.
457 * @log_num: The logical number, if any of this channel.
458 * @pending_tx: The number of pending transfers. Used between interrupt handler
459 * and tasklet.
460 * @busy: Set to true when transfer is ongoing on this channel.
461 * @phy_chan: Pointer to physical channel which this instance runs on. If this
462 * point is NULL, then the channel is not allocated.
463 * @chan: DMA engine handle.
464 * @tasklet: Tasklet that gets scheduled from interrupt context to complete a
465 * transfer and call client callback.
466 * @client: Cliented owned descriptor list.
467 * @pending_queue: Submitted jobs, to be issued by issue_pending()
468 * @active: Active descriptor.
469 * @done: Completed jobs
470 * @queue: Queued jobs.
471 * @prepare_queue: Prepared jobs.
472 * @dma_cfg: The client configuration of this dma channel.
473 * @slave_config: DMA slave configuration.
474 * @configured: whether the dma_cfg configuration is valid
475 * @base: Pointer to the device instance struct.
476 * @src_def_cfg: Default cfg register setting for src.
477 * @dst_def_cfg: Default cfg register setting for dst.
478 * @log_def: Default logical channel settings.
479 * @lcpa: Pointer to dst and src lcpa settings.
480 * @runtime_addr: runtime configured address.
481 * @runtime_direction: runtime configured direction.
482 *
483 * This struct can either "be" a logical or a physical channel.
484 */
485struct d40_chan {
486 spinlock_t lock;
487 int log_num;
488 int pending_tx;
489 bool busy;
490 struct d40_phy_res *phy_chan;
491 struct dma_chan chan;
492 struct tasklet_struct tasklet;
493 struct list_head client;
494 struct list_head pending_queue;
495 struct list_head active;
496 struct list_head done;
497 struct list_head queue;
498 struct list_head prepare_queue;
499 struct stedma40_chan_cfg dma_cfg;
500 struct dma_slave_config slave_config;
501 bool configured;
502 struct d40_base *base;
503 /* Default register configurations */
504 u32 src_def_cfg;
505 u32 dst_def_cfg;
506 struct d40_def_lcsp log_def;
507 struct d40_log_lli_full *lcpa;
508 /* Runtime reconfiguration */
509 dma_addr_t runtime_addr;
510 enum dma_transfer_direction runtime_direction;
511};
512
513/**
514 * struct d40_gen_dmac - generic values to represent u8500/u8540 DMA
515 * controller
516 *
517 * @backup: the pointer to the registers address array for backup
518 * @backup_size: the size of the registers address array for backup
519 * @realtime_en: the realtime enable register
520 * @realtime_clear: the realtime clear register
521 * @high_prio_en: the high priority enable register
522 * @high_prio_clear: the high priority clear register
523 * @interrupt_en: the interrupt enable register
524 * @interrupt_clear: the interrupt clear register
525 * @il: the pointer to struct d40_interrupt_lookup
526 * @il_size: the size of d40_interrupt_lookup array
527 * @init_reg: the pointer to the struct d40_reg_val
528 * @init_reg_size: the size of d40_reg_val array
529 */
530struct d40_gen_dmac {
531 u32 *backup;
532 u32 backup_size;
533 u32 realtime_en;
534 u32 realtime_clear;
535 u32 high_prio_en;
536 u32 high_prio_clear;
537 u32 interrupt_en;
538 u32 interrupt_clear;
539 struct d40_interrupt_lookup *il;
540 u32 il_size;
541 struct d40_reg_val *init_reg;
542 u32 init_reg_size;
543};
544
545/**
546 * struct d40_base - The big global struct, one for each probe'd instance.
547 *
548 * @interrupt_lock: Lock used to make sure one interrupt is handle a time.
549 * @execmd_lock: Lock for execute command usage since several channels share
550 * the same physical register.
551 * @dev: The device structure.
552 * @virtbase: The virtual base address of the DMA's register.
553 * @rev: silicon revision detected.
554 * @clk: Pointer to the DMA clock structure.
555 * @irq: The IRQ number.
556 * @num_memcpy_chans: The number of channels used for memcpy (mem-to-mem
557 * transfers).
558 * @num_phy_chans: The number of physical channels. Read from HW. This
559 * is the number of available channels for this driver, not counting "Secure
560 * mode" allocated physical channels.
561 * @num_log_chans: The number of logical channels. Calculated from
562 * num_phy_chans.
563 * @dma_both: dma_device channels that can do both memcpy and slave transfers.
564 * @dma_slave: dma_device channels that can do only do slave transfers.
565 * @dma_memcpy: dma_device channels that can do only do memcpy transfers.
566 * @phy_chans: Room for all possible physical channels in system.
567 * @log_chans: Room for all possible logical channels in system.
568 * @lookup_log_chans: Used to map interrupt number to logical channel. Points
569 * to log_chans entries.
570 * @lookup_phy_chans: Used to map interrupt number to physical channel. Points
571 * to phy_chans entries.
572 * @plat_data: Pointer to provided platform_data which is the driver
573 * configuration.
574 * @lcpa_regulator: Pointer to hold the regulator for the esram bank for lcla.
575 * @phy_res: Vector containing all physical channels.
576 * @lcla_pool: lcla pool settings and data.
577 * @lcpa_base: The virtual mapped address of LCPA.
578 * @phy_lcpa: The physical address of the LCPA.
579 * @lcpa_size: The size of the LCPA area.
580 * @desc_slab: cache for descriptors.
581 * @reg_val_backup: Here the values of some hardware registers are stored
582 * before the DMA is powered off. They are restored when the power is back on.
583 * @reg_val_backup_v4: Backup of registers that only exits on dma40 v3 and
584 * later
585 * @reg_val_backup_chan: Backup data for standard channel parameter registers.
586 * @regs_interrupt: Scratch space for registers during interrupt.
587 * @gcc_pwr_off_mask: Mask to maintain the channels that can be turned off.
588 * @gen_dmac: the struct for generic registers values to represent u8500/8540
589 * DMA controller
590 */
591struct d40_base {
592 spinlock_t interrupt_lock;
593 spinlock_t execmd_lock;
594 struct device *dev;
595 void __iomem *virtbase;
596 u8 rev:4;
597 struct clk *clk;
598 int irq;
599 int num_memcpy_chans;
600 int num_phy_chans;
601 int num_log_chans;
602 struct dma_device dma_both;
603 struct dma_device dma_slave;
604 struct dma_device dma_memcpy;
605 struct d40_chan *phy_chans;
606 struct d40_chan *log_chans;
607 struct d40_chan **lookup_log_chans;
608 struct d40_chan **lookup_phy_chans;
609 struct stedma40_platform_data *plat_data;
610 struct regulator *lcpa_regulator;
611 /* Physical half channels */
612 struct d40_phy_res *phy_res;
613 struct d40_lcla_pool lcla_pool;
614 void *lcpa_base;
615 dma_addr_t phy_lcpa;
616 resource_size_t lcpa_size;
617 struct kmem_cache *desc_slab;
618 u32 reg_val_backup[BACKUP_REGS_SZ];
619 u32 reg_val_backup_v4[BACKUP_REGS_SZ_MAX];
620 u32 *reg_val_backup_chan;
621 u32 *regs_interrupt;
622 u16 gcc_pwr_off_mask;
623 struct d40_gen_dmac gen_dmac;
624};
625
626static struct device *chan2dev(struct d40_chan *d40c)
627{
628 return &d40c->chan.dev->device;
629}
630
631static bool chan_is_physical(struct d40_chan *chan)
632{
633 return chan->log_num == D40_PHY_CHAN;
634}
635
636static bool chan_is_logical(struct d40_chan *chan)
637{
638 return !chan_is_physical(chan);
639}
640
641static void __iomem *chan_base(struct d40_chan *chan)
642{
643 return chan->base->virtbase + D40_DREG_PCBASE +
644 chan->phy_chan->num * D40_DREG_PCDELTA;
645}
646
647#define d40_err(dev, format, arg...) \
648 dev_err(dev, "[%s] " format, __func__, ## arg)
649
650#define chan_err(d40c, format, arg...) \
651 d40_err(chan2dev(d40c), format, ## arg)
652
653static int d40_set_runtime_config_write(struct dma_chan *chan,
654 struct dma_slave_config *config,
655 enum dma_transfer_direction direction);
656
657static int d40_pool_lli_alloc(struct d40_chan *d40c, struct d40_desc *d40d,
658 int lli_len)
659{
660 bool is_log = chan_is_logical(d40c);
661 u32 align;
662 void *base;
663
664 if (is_log)
665 align = sizeof(struct d40_log_lli);
666 else
667 align = sizeof(struct d40_phy_lli);
668
669 if (lli_len == 1) {
670 base = d40d->lli_pool.pre_alloc_lli;
671 d40d->lli_pool.size = sizeof(d40d->lli_pool.pre_alloc_lli);
672 d40d->lli_pool.base = NULL;
673 } else {
674 d40d->lli_pool.size = lli_len * 2 * align;
675
676 base = kmalloc(d40d->lli_pool.size + align, GFP_NOWAIT);
677 d40d->lli_pool.base = base;
678
679 if (d40d->lli_pool.base == NULL)
680 return -ENOMEM;
681 }
682
683 if (is_log) {
684 d40d->lli_log.src = PTR_ALIGN(base, align);
685 d40d->lli_log.dst = d40d->lli_log.src + lli_len;
686
687 d40d->lli_pool.dma_addr = 0;
688 } else {
689 d40d->lli_phy.src = PTR_ALIGN(base, align);
690 d40d->lli_phy.dst = d40d->lli_phy.src + lli_len;
691
692 d40d->lli_pool.dma_addr = dma_map_single(d40c->base->dev,
693 d40d->lli_phy.src,
694 d40d->lli_pool.size,
695 DMA_TO_DEVICE);
696
697 if (dma_mapping_error(d40c->base->dev,
698 d40d->lli_pool.dma_addr)) {
699 kfree(d40d->lli_pool.base);
700 d40d->lli_pool.base = NULL;
701 d40d->lli_pool.dma_addr = 0;
702 return -ENOMEM;
703 }
704 }
705
706 return 0;
707}
708
709static void d40_pool_lli_free(struct d40_chan *d40c, struct d40_desc *d40d)
710{
711 if (d40d->lli_pool.dma_addr)
712 dma_unmap_single(d40c->base->dev, d40d->lli_pool.dma_addr,
713 d40d->lli_pool.size, DMA_TO_DEVICE);
714
715 kfree(d40d->lli_pool.base);
716 d40d->lli_pool.base = NULL;
717 d40d->lli_pool.size = 0;
718 d40d->lli_log.src = NULL;
719 d40d->lli_log.dst = NULL;
720 d40d->lli_phy.src = NULL;
721 d40d->lli_phy.dst = NULL;
722}
723
724static int d40_lcla_alloc_one(struct d40_chan *d40c,
725 struct d40_desc *d40d)
726{
727 unsigned long flags;
728 int i;
729 int ret = -EINVAL;
730
731 spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
732
733 /*
734 * Allocate both src and dst at the same time, therefore the half
735 * start on 1 since 0 can't be used since zero is used as end marker.
736 */
737 for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
738 int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
739
740 if (!d40c->base->lcla_pool.alloc_map[idx]) {
741 d40c->base->lcla_pool.alloc_map[idx] = d40d;
742 d40d->lcla_alloc++;
743 ret = i;
744 break;
745 }
746 }
747
748 spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
749
750 return ret;
751}
752
753static int d40_lcla_free_all(struct d40_chan *d40c,
754 struct d40_desc *d40d)
755{
756 unsigned long flags;
757 int i;
758 int ret = -EINVAL;
759
760 if (chan_is_physical(d40c))
761 return 0;
762
763 spin_lock_irqsave(&d40c->base->lcla_pool.lock, flags);
764
765 for (i = 1 ; i < D40_LCLA_LINK_PER_EVENT_GRP / 2; i++) {
766 int idx = d40c->phy_chan->num * D40_LCLA_LINK_PER_EVENT_GRP + i;
767
768 if (d40c->base->lcla_pool.alloc_map[idx] == d40d) {
769 d40c->base->lcla_pool.alloc_map[idx] = NULL;
770 d40d->lcla_alloc--;
771 if (d40d->lcla_alloc == 0) {
772 ret = 0;
773 break;
774 }
775 }
776 }
777
778 spin_unlock_irqrestore(&d40c->base->lcla_pool.lock, flags);
779
780 return ret;
781
782}
783
784static void d40_desc_remove(struct d40_desc *d40d)
785{
786 list_del(&d40d->node);
787}
788
789static struct d40_desc *d40_desc_get(struct d40_chan *d40c)
790{
791 struct d40_desc *desc = NULL;
792
793 if (!list_empty(&d40c->client)) {
794 struct d40_desc *d;
795 struct d40_desc *_d;
796
797 list_for_each_entry_safe(d, _d, &d40c->client, node) {
798 if (async_tx_test_ack(&d->txd)) {
799 d40_desc_remove(d);
800 desc = d;
801 memset(desc, 0, sizeof(*desc));
802 break;
803 }
804 }
805 }
806
807 if (!desc)
808 desc = kmem_cache_zalloc(d40c->base->desc_slab, GFP_NOWAIT);
809
810 if (desc)
811 INIT_LIST_HEAD(&desc->node);
812
813 return desc;
814}
815
816static void d40_desc_free(struct d40_chan *d40c, struct d40_desc *d40d)
817{
818
819 d40_pool_lli_free(d40c, d40d);
820 d40_lcla_free_all(d40c, d40d);
821 kmem_cache_free(d40c->base->desc_slab, d40d);
822}
823
824static void d40_desc_submit(struct d40_chan *d40c, struct d40_desc *desc)
825{
826 list_add_tail(&desc->node, &d40c->active);
827}
828
829static void d40_phy_lli_load(struct d40_chan *chan, struct d40_desc *desc)
830{
831 struct d40_phy_lli *lli_dst = desc->lli_phy.dst;
832 struct d40_phy_lli *lli_src = desc->lli_phy.src;
833 void __iomem *base = chan_base(chan);
834
835 writel(lli_src->reg_cfg, base + D40_CHAN_REG_SSCFG);
836 writel(lli_src->reg_elt, base + D40_CHAN_REG_SSELT);
837 writel(lli_src->reg_ptr, base + D40_CHAN_REG_SSPTR);
838 writel(lli_src->reg_lnk, base + D40_CHAN_REG_SSLNK);
839
840 writel(lli_dst->reg_cfg, base + D40_CHAN_REG_SDCFG);
841 writel(lli_dst->reg_elt, base + D40_CHAN_REG_SDELT);
842 writel(lli_dst->reg_ptr, base + D40_CHAN_REG_SDPTR);
843 writel(lli_dst->reg_lnk, base + D40_CHAN_REG_SDLNK);
844}
845
846static void d40_desc_done(struct d40_chan *d40c, struct d40_desc *desc)
847{
848 list_add_tail(&desc->node, &d40c->done);
849}
850
851static void d40_log_lli_to_lcxa(struct d40_chan *chan, struct d40_desc *desc)
852{
853 struct d40_lcla_pool *pool = &chan->base->lcla_pool;
854 struct d40_log_lli_bidir *lli = &desc->lli_log;
855 int lli_current = desc->lli_current;
856 int lli_len = desc->lli_len;
857 bool cyclic = desc->cyclic;
858 int curr_lcla = -EINVAL;
859 int first_lcla = 0;
860 bool use_esram_lcla = chan->base->plat_data->use_esram_lcla;
861 bool linkback;
862
863 /*
864 * We may have partially running cyclic transfers, in case we did't get
865 * enough LCLA entries.
866 */
867 linkback = cyclic && lli_current == 0;
868
869 /*
870 * For linkback, we need one LCLA even with only one link, because we
871 * can't link back to the one in LCPA space
872 */
873 if (linkback || (lli_len - lli_current > 1)) {
874 /*
875 * If the channel is expected to use only soft_lli don't
876 * allocate a lcla. This is to avoid a HW issue that exists
877 * in some controller during a peripheral to memory transfer
878 * that uses linked lists.
879 */
880 if (!(chan->phy_chan->use_soft_lli &&
881 chan->dma_cfg.dir == DMA_DEV_TO_MEM))
882 curr_lcla = d40_lcla_alloc_one(chan, desc);
883
884 first_lcla = curr_lcla;
885 }
886
887 /*
888 * For linkback, we normally load the LCPA in the loop since we need to
889 * link it to the second LCLA and not the first. However, if we
890 * couldn't even get a first LCLA, then we have to run in LCPA and
891 * reload manually.
892 */
893 if (!linkback || curr_lcla == -EINVAL) {
894 unsigned int flags = 0;
895
896 if (curr_lcla == -EINVAL)
897 flags |= LLI_TERM_INT;
898
899 d40_log_lli_lcpa_write(chan->lcpa,
900 &lli->dst[lli_current],
901 &lli->src[lli_current],
902 curr_lcla,
903 flags);
904 lli_current++;
905 }
906
907 if (curr_lcla < 0)
908 goto set_current;
909
910 for (; lli_current < lli_len; lli_current++) {
911 unsigned int lcla_offset = chan->phy_chan->num * 1024 +
912 8 * curr_lcla * 2;
913 struct d40_log_lli *lcla = pool->base + lcla_offset;
914 unsigned int flags = 0;
915 int next_lcla;
916
917 if (lli_current + 1 < lli_len)
918 next_lcla = d40_lcla_alloc_one(chan, desc);
919 else
920 next_lcla = linkback ? first_lcla : -EINVAL;
921
922 if (cyclic || next_lcla == -EINVAL)
923 flags |= LLI_TERM_INT;
924
925 if (linkback && curr_lcla == first_lcla) {
926 /* First link goes in both LCPA and LCLA */
927 d40_log_lli_lcpa_write(chan->lcpa,
928 &lli->dst[lli_current],
929 &lli->src[lli_current],
930 next_lcla, flags);
931 }
932
933 /*
934 * One unused LCLA in the cyclic case if the very first
935 * next_lcla fails...
936 */
937 d40_log_lli_lcla_write(lcla,
938 &lli->dst[lli_current],
939 &lli->src[lli_current],
940 next_lcla, flags);
941
942 /*
943 * Cache maintenance is not needed if lcla is
944 * mapped in esram
945 */
946 if (!use_esram_lcla) {
947 dma_sync_single_range_for_device(chan->base->dev,
948 pool->dma_addr, lcla_offset,
949 2 * sizeof(struct d40_log_lli),
950 DMA_TO_DEVICE);
951 }
952 curr_lcla = next_lcla;
953
954 if (curr_lcla == -EINVAL || curr_lcla == first_lcla) {
955 lli_current++;
956 break;
957 }
958 }
959 set_current:
960 desc->lli_current = lli_current;
961}
962
963static void d40_desc_load(struct d40_chan *d40c, struct d40_desc *d40d)
964{
965 if (chan_is_physical(d40c)) {
966 d40_phy_lli_load(d40c, d40d);
967 d40d->lli_current = d40d->lli_len;
968 } else
969 d40_log_lli_to_lcxa(d40c, d40d);
970}
971
972static struct d40_desc *d40_first_active_get(struct d40_chan *d40c)
973{
974 return list_first_entry_or_null(&d40c->active, struct d40_desc, node);
975}
976
977/* remove desc from current queue and add it to the pending_queue */
978static void d40_desc_queue(struct d40_chan *d40c, struct d40_desc *desc)
979{
980 d40_desc_remove(desc);
981 desc->is_in_client_list = false;
982 list_add_tail(&desc->node, &d40c->pending_queue);
983}
984
985static struct d40_desc *d40_first_pending(struct d40_chan *d40c)
986{
987 return list_first_entry_or_null(&d40c->pending_queue, struct d40_desc,
988 node);
989}
990
991static struct d40_desc *d40_first_queued(struct d40_chan *d40c)
992{
993 return list_first_entry_or_null(&d40c->queue, struct d40_desc, node);
994}
995
996static struct d40_desc *d40_first_done(struct d40_chan *d40c)
997{
998 return list_first_entry_or_null(&d40c->done, struct d40_desc, node);
999}
1000
1001static int d40_psize_2_burst_size(bool is_log, int psize)
1002{
1003 if (is_log) {
1004 if (psize == STEDMA40_PSIZE_LOG_1)
1005 return 1;
1006 } else {
1007 if (psize == STEDMA40_PSIZE_PHY_1)
1008 return 1;
1009 }
1010
1011 return 2 << psize;
1012}
1013
1014/*
1015 * The dma only supports transmitting packages up to
1016 * STEDMA40_MAX_SEG_SIZE * data_width, where data_width is stored in Bytes.
1017 *
1018 * Calculate the total number of dma elements required to send the entire sg list.
1019 */
1020static int d40_size_2_dmalen(int size, u32 data_width1, u32 data_width2)
1021{
1022 int dmalen;
1023 u32 max_w = max(data_width1, data_width2);
1024 u32 min_w = min(data_width1, data_width2);
1025 u32 seg_max = ALIGN(STEDMA40_MAX_SEG_SIZE * min_w, max_w);
1026
1027 if (seg_max > STEDMA40_MAX_SEG_SIZE)
1028 seg_max -= max_w;
1029
1030 if (!IS_ALIGNED(size, max_w))
1031 return -EINVAL;
1032
1033 if (size <= seg_max)
1034 dmalen = 1;
1035 else {
1036 dmalen = size / seg_max;
1037 if (dmalen * seg_max < size)
1038 dmalen++;
1039 }
1040 return dmalen;
1041}
1042
1043static int d40_sg_2_dmalen(struct scatterlist *sgl, int sg_len,
1044 u32 data_width1, u32 data_width2)
1045{
1046 struct scatterlist *sg;
1047 int i;
1048 int len = 0;
1049 int ret;
1050
1051 for_each_sg(sgl, sg, sg_len, i) {
1052 ret = d40_size_2_dmalen(sg_dma_len(sg),
1053 data_width1, data_width2);
1054 if (ret < 0)
1055 return ret;
1056 len += ret;
1057 }
1058 return len;
1059}
1060
1061static int __d40_execute_command_phy(struct d40_chan *d40c,
1062 enum d40_command command)
1063{
1064 u32 status;
1065 int i;
1066 void __iomem *active_reg;
1067 int ret = 0;
1068 unsigned long flags;
1069 u32 wmask;
1070
1071 if (command == D40_DMA_STOP) {
1072 ret = __d40_execute_command_phy(d40c, D40_DMA_SUSPEND_REQ);
1073 if (ret)
1074 return ret;
1075 }
1076
1077 spin_lock_irqsave(&d40c->base->execmd_lock, flags);
1078
1079 if (d40c->phy_chan->num % 2 == 0)
1080 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1081 else
1082 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1083
1084 if (command == D40_DMA_SUSPEND_REQ) {
1085 status = (readl(active_reg) &
1086 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1087 D40_CHAN_POS(d40c->phy_chan->num);
1088
1089 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
1090 goto unlock;
1091 }
1092
1093 wmask = 0xffffffff & ~(D40_CHAN_POS_MASK(d40c->phy_chan->num));
1094 writel(wmask | (command << D40_CHAN_POS(d40c->phy_chan->num)),
1095 active_reg);
1096
1097 if (command == D40_DMA_SUSPEND_REQ) {
1098
1099 for (i = 0 ; i < D40_SUSPEND_MAX_IT; i++) {
1100 status = (readl(active_reg) &
1101 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1102 D40_CHAN_POS(d40c->phy_chan->num);
1103
1104 cpu_relax();
1105 /*
1106 * Reduce the number of bus accesses while
1107 * waiting for the DMA to suspend.
1108 */
1109 udelay(3);
1110
1111 if (status == D40_DMA_STOP ||
1112 status == D40_DMA_SUSPENDED)
1113 break;
1114 }
1115
1116 if (i == D40_SUSPEND_MAX_IT) {
1117 chan_err(d40c,
1118 "unable to suspend the chl %d (log: %d) status %x\n",
1119 d40c->phy_chan->num, d40c->log_num,
1120 status);
1121 dump_stack();
1122 ret = -EBUSY;
1123 }
1124
1125 }
1126 unlock:
1127 spin_unlock_irqrestore(&d40c->base->execmd_lock, flags);
1128 return ret;
1129}
1130
1131static void d40_term_all(struct d40_chan *d40c)
1132{
1133 struct d40_desc *d40d;
1134 struct d40_desc *_d;
1135
1136 /* Release completed descriptors */
1137 while ((d40d = d40_first_done(d40c))) {
1138 d40_desc_remove(d40d);
1139 d40_desc_free(d40c, d40d);
1140 }
1141
1142 /* Release active descriptors */
1143 while ((d40d = d40_first_active_get(d40c))) {
1144 d40_desc_remove(d40d);
1145 d40_desc_free(d40c, d40d);
1146 }
1147
1148 /* Release queued descriptors waiting for transfer */
1149 while ((d40d = d40_first_queued(d40c))) {
1150 d40_desc_remove(d40d);
1151 d40_desc_free(d40c, d40d);
1152 }
1153
1154 /* Release pending descriptors */
1155 while ((d40d = d40_first_pending(d40c))) {
1156 d40_desc_remove(d40d);
1157 d40_desc_free(d40c, d40d);
1158 }
1159
1160 /* Release client owned descriptors */
1161 if (!list_empty(&d40c->client))
1162 list_for_each_entry_safe(d40d, _d, &d40c->client, node) {
1163 d40_desc_remove(d40d);
1164 d40_desc_free(d40c, d40d);
1165 }
1166
1167 /* Release descriptors in prepare queue */
1168 if (!list_empty(&d40c->prepare_queue))
1169 list_for_each_entry_safe(d40d, _d,
1170 &d40c->prepare_queue, node) {
1171 d40_desc_remove(d40d);
1172 d40_desc_free(d40c, d40d);
1173 }
1174
1175 d40c->pending_tx = 0;
1176}
1177
1178static void __d40_config_set_event(struct d40_chan *d40c,
1179 enum d40_events event_type, u32 event,
1180 int reg)
1181{
1182 void __iomem *addr = chan_base(d40c) + reg;
1183 int tries;
1184 u32 status;
1185
1186 switch (event_type) {
1187
1188 case D40_DEACTIVATE_EVENTLINE:
1189
1190 writel((D40_DEACTIVATE_EVENTLINE << D40_EVENTLINE_POS(event))
1191 | ~D40_EVENTLINE_MASK(event), addr);
1192 break;
1193
1194 case D40_SUSPEND_REQ_EVENTLINE:
1195 status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1196 D40_EVENTLINE_POS(event);
1197
1198 if (status == D40_DEACTIVATE_EVENTLINE ||
1199 status == D40_SUSPEND_REQ_EVENTLINE)
1200 break;
1201
1202 writel((D40_SUSPEND_REQ_EVENTLINE << D40_EVENTLINE_POS(event))
1203 | ~D40_EVENTLINE_MASK(event), addr);
1204
1205 for (tries = 0 ; tries < D40_SUSPEND_MAX_IT; tries++) {
1206
1207 status = (readl(addr) & D40_EVENTLINE_MASK(event)) >>
1208 D40_EVENTLINE_POS(event);
1209
1210 cpu_relax();
1211 /*
1212 * Reduce the number of bus accesses while
1213 * waiting for the DMA to suspend.
1214 */
1215 udelay(3);
1216
1217 if (status == D40_DEACTIVATE_EVENTLINE)
1218 break;
1219 }
1220
1221 if (tries == D40_SUSPEND_MAX_IT) {
1222 chan_err(d40c,
1223 "unable to stop the event_line chl %d (log: %d)"
1224 "status %x\n", d40c->phy_chan->num,
1225 d40c->log_num, status);
1226 }
1227 break;
1228
1229 case D40_ACTIVATE_EVENTLINE:
1230 /*
1231 * The hardware sometimes doesn't register the enable when src and dst
1232 * event lines are active on the same logical channel. Retry to ensure
1233 * it does. Usually only one retry is sufficient.
1234 */
1235 tries = 100;
1236 while (--tries) {
1237 writel((D40_ACTIVATE_EVENTLINE <<
1238 D40_EVENTLINE_POS(event)) |
1239 ~D40_EVENTLINE_MASK(event), addr);
1240
1241 if (readl(addr) & D40_EVENTLINE_MASK(event))
1242 break;
1243 }
1244
1245 if (tries != 99)
1246 dev_dbg(chan2dev(d40c),
1247 "[%s] workaround enable S%cLNK (%d tries)\n",
1248 __func__, reg == D40_CHAN_REG_SSLNK ? 'S' : 'D',
1249 100 - tries);
1250
1251 WARN_ON(!tries);
1252 break;
1253
1254 case D40_ROUND_EVENTLINE:
1255 BUG();
1256 break;
1257
1258 }
1259}
1260
1261static void d40_config_set_event(struct d40_chan *d40c,
1262 enum d40_events event_type)
1263{
1264 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
1265
1266 /* Enable event line connected to device (or memcpy) */
1267 if ((d40c->dma_cfg.dir == DMA_DEV_TO_MEM) ||
1268 (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
1269 __d40_config_set_event(d40c, event_type, event,
1270 D40_CHAN_REG_SSLNK);
1271
1272 if (d40c->dma_cfg.dir != DMA_DEV_TO_MEM)
1273 __d40_config_set_event(d40c, event_type, event,
1274 D40_CHAN_REG_SDLNK);
1275}
1276
1277static u32 d40_chan_has_events(struct d40_chan *d40c)
1278{
1279 void __iomem *chanbase = chan_base(d40c);
1280 u32 val;
1281
1282 val = readl(chanbase + D40_CHAN_REG_SSLNK);
1283 val |= readl(chanbase + D40_CHAN_REG_SDLNK);
1284
1285 return val;
1286}
1287
1288static int
1289__d40_execute_command_log(struct d40_chan *d40c, enum d40_command command)
1290{
1291 unsigned long flags;
1292 int ret = 0;
1293 u32 active_status;
1294 void __iomem *active_reg;
1295
1296 if (d40c->phy_chan->num % 2 == 0)
1297 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
1298 else
1299 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
1300
1301
1302 spin_lock_irqsave(&d40c->phy_chan->lock, flags);
1303
1304 switch (command) {
1305 case D40_DMA_STOP:
1306 case D40_DMA_SUSPEND_REQ:
1307
1308 active_status = (readl(active_reg) &
1309 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
1310 D40_CHAN_POS(d40c->phy_chan->num);
1311
1312 if (active_status == D40_DMA_RUN)
1313 d40_config_set_event(d40c, D40_SUSPEND_REQ_EVENTLINE);
1314 else
1315 d40_config_set_event(d40c, D40_DEACTIVATE_EVENTLINE);
1316
1317 if (!d40_chan_has_events(d40c) && (command == D40_DMA_STOP))
1318 ret = __d40_execute_command_phy(d40c, command);
1319
1320 break;
1321
1322 case D40_DMA_RUN:
1323
1324 d40_config_set_event(d40c, D40_ACTIVATE_EVENTLINE);
1325 ret = __d40_execute_command_phy(d40c, command);
1326 break;
1327
1328 case D40_DMA_SUSPENDED:
1329 BUG();
1330 break;
1331 }
1332
1333 spin_unlock_irqrestore(&d40c->phy_chan->lock, flags);
1334 return ret;
1335}
1336
1337static int d40_channel_execute_command(struct d40_chan *d40c,
1338 enum d40_command command)
1339{
1340 if (chan_is_logical(d40c))
1341 return __d40_execute_command_log(d40c, command);
1342 else
1343 return __d40_execute_command_phy(d40c, command);
1344}
1345
1346static u32 d40_get_prmo(struct d40_chan *d40c)
1347{
1348 static const unsigned int phy_map[] = {
1349 [STEDMA40_PCHAN_BASIC_MODE]
1350 = D40_DREG_PRMO_PCHAN_BASIC,
1351 [STEDMA40_PCHAN_MODULO_MODE]
1352 = D40_DREG_PRMO_PCHAN_MODULO,
1353 [STEDMA40_PCHAN_DOUBLE_DST_MODE]
1354 = D40_DREG_PRMO_PCHAN_DOUBLE_DST,
1355 };
1356 static const unsigned int log_map[] = {
1357 [STEDMA40_LCHAN_SRC_PHY_DST_LOG]
1358 = D40_DREG_PRMO_LCHAN_SRC_PHY_DST_LOG,
1359 [STEDMA40_LCHAN_SRC_LOG_DST_PHY]
1360 = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_PHY,
1361 [STEDMA40_LCHAN_SRC_LOG_DST_LOG]
1362 = D40_DREG_PRMO_LCHAN_SRC_LOG_DST_LOG,
1363 };
1364
1365 if (chan_is_physical(d40c))
1366 return phy_map[d40c->dma_cfg.mode_opt];
1367 else
1368 return log_map[d40c->dma_cfg.mode_opt];
1369}
1370
1371static void d40_config_write(struct d40_chan *d40c)
1372{
1373 u32 addr_base;
1374 u32 var;
1375
1376 /* Odd addresses are even addresses + 4 */
1377 addr_base = (d40c->phy_chan->num % 2) * 4;
1378 /* Setup channel mode to logical or physical */
1379 var = ((u32)(chan_is_logical(d40c)) + 1) <<
1380 D40_CHAN_POS(d40c->phy_chan->num);
1381 writel(var, d40c->base->virtbase + D40_DREG_PRMSE + addr_base);
1382
1383 /* Setup operational mode option register */
1384 var = d40_get_prmo(d40c) << D40_CHAN_POS(d40c->phy_chan->num);
1385
1386 writel(var, d40c->base->virtbase + D40_DREG_PRMOE + addr_base);
1387
1388 if (chan_is_logical(d40c)) {
1389 int lidx = (d40c->phy_chan->num << D40_SREG_ELEM_LOG_LIDX_POS)
1390 & D40_SREG_ELEM_LOG_LIDX_MASK;
1391 void __iomem *chanbase = chan_base(d40c);
1392
1393 /* Set default config for CFG reg */
1394 writel(d40c->src_def_cfg, chanbase + D40_CHAN_REG_SSCFG);
1395 writel(d40c->dst_def_cfg, chanbase + D40_CHAN_REG_SDCFG);
1396
1397 /* Set LIDX for lcla */
1398 writel(lidx, chanbase + D40_CHAN_REG_SSELT);
1399 writel(lidx, chanbase + D40_CHAN_REG_SDELT);
1400
1401 /* Clear LNK which will be used by d40_chan_has_events() */
1402 writel(0, chanbase + D40_CHAN_REG_SSLNK);
1403 writel(0, chanbase + D40_CHAN_REG_SDLNK);
1404 }
1405}
1406
1407static u32 d40_residue(struct d40_chan *d40c)
1408{
1409 u32 num_elt;
1410
1411 if (chan_is_logical(d40c))
1412 num_elt = (readl(&d40c->lcpa->lcsp2) & D40_MEM_LCSP2_ECNT_MASK)
1413 >> D40_MEM_LCSP2_ECNT_POS;
1414 else {
1415 u32 val = readl(chan_base(d40c) + D40_CHAN_REG_SDELT);
1416 num_elt = (val & D40_SREG_ELEM_PHY_ECNT_MASK)
1417 >> D40_SREG_ELEM_PHY_ECNT_POS;
1418 }
1419
1420 return num_elt * d40c->dma_cfg.dst_info.data_width;
1421}
1422
1423static bool d40_tx_is_linked(struct d40_chan *d40c)
1424{
1425 bool is_link;
1426
1427 if (chan_is_logical(d40c))
1428 is_link = readl(&d40c->lcpa->lcsp3) & D40_MEM_LCSP3_DLOS_MASK;
1429 else
1430 is_link = readl(chan_base(d40c) + D40_CHAN_REG_SDLNK)
1431 & D40_SREG_LNK_PHYS_LNK_MASK;
1432
1433 return is_link;
1434}
1435
1436static int d40_pause(struct dma_chan *chan)
1437{
1438 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
1439 int res = 0;
1440 unsigned long flags;
1441
1442 if (d40c->phy_chan == NULL) {
1443 chan_err(d40c, "Channel is not allocated!\n");
1444 return -EINVAL;
1445 }
1446
1447 if (!d40c->busy)
1448 return 0;
1449
1450 spin_lock_irqsave(&d40c->lock, flags);
1451 pm_runtime_get_sync(d40c->base->dev);
1452
1453 res = d40_channel_execute_command(d40c, D40_DMA_SUSPEND_REQ);
1454
1455 pm_runtime_mark_last_busy(d40c->base->dev);
1456 pm_runtime_put_autosuspend(d40c->base->dev);
1457 spin_unlock_irqrestore(&d40c->lock, flags);
1458 return res;
1459}
1460
1461static int d40_resume(struct dma_chan *chan)
1462{
1463 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
1464 int res = 0;
1465 unsigned long flags;
1466
1467 if (d40c->phy_chan == NULL) {
1468 chan_err(d40c, "Channel is not allocated!\n");
1469 return -EINVAL;
1470 }
1471
1472 if (!d40c->busy)
1473 return 0;
1474
1475 spin_lock_irqsave(&d40c->lock, flags);
1476 pm_runtime_get_sync(d40c->base->dev);
1477
1478 /* If bytes left to transfer or linked tx resume job */
1479 if (d40_residue(d40c) || d40_tx_is_linked(d40c))
1480 res = d40_channel_execute_command(d40c, D40_DMA_RUN);
1481
1482 pm_runtime_mark_last_busy(d40c->base->dev);
1483 pm_runtime_put_autosuspend(d40c->base->dev);
1484 spin_unlock_irqrestore(&d40c->lock, flags);
1485 return res;
1486}
1487
1488static dma_cookie_t d40_tx_submit(struct dma_async_tx_descriptor *tx)
1489{
1490 struct d40_chan *d40c = container_of(tx->chan,
1491 struct d40_chan,
1492 chan);
1493 struct d40_desc *d40d = container_of(tx, struct d40_desc, txd);
1494 unsigned long flags;
1495 dma_cookie_t cookie;
1496
1497 spin_lock_irqsave(&d40c->lock, flags);
1498 cookie = dma_cookie_assign(tx);
1499 d40_desc_queue(d40c, d40d);
1500 spin_unlock_irqrestore(&d40c->lock, flags);
1501
1502 return cookie;
1503}
1504
1505static int d40_start(struct d40_chan *d40c)
1506{
1507 return d40_channel_execute_command(d40c, D40_DMA_RUN);
1508}
1509
1510static struct d40_desc *d40_queue_start(struct d40_chan *d40c)
1511{
1512 struct d40_desc *d40d;
1513 int err;
1514
1515 /* Start queued jobs, if any */
1516 d40d = d40_first_queued(d40c);
1517
1518 if (d40d != NULL) {
1519 if (!d40c->busy) {
1520 d40c->busy = true;
1521 pm_runtime_get_sync(d40c->base->dev);
1522 }
1523
1524 /* Remove from queue */
1525 d40_desc_remove(d40d);
1526
1527 /* Add to active queue */
1528 d40_desc_submit(d40c, d40d);
1529
1530 /* Initiate DMA job */
1531 d40_desc_load(d40c, d40d);
1532
1533 /* Start dma job */
1534 err = d40_start(d40c);
1535
1536 if (err)
1537 return NULL;
1538 }
1539
1540 return d40d;
1541}
1542
1543/* called from interrupt context */
1544static void dma_tc_handle(struct d40_chan *d40c)
1545{
1546 struct d40_desc *d40d;
1547
1548 /* Get first active entry from list */
1549 d40d = d40_first_active_get(d40c);
1550
1551 if (d40d == NULL)
1552 return;
1553
1554 if (d40d->cyclic) {
1555 /*
1556 * If this was a paritially loaded list, we need to reloaded
1557 * it, and only when the list is completed. We need to check
1558 * for done because the interrupt will hit for every link, and
1559 * not just the last one.
1560 */
1561 if (d40d->lli_current < d40d->lli_len
1562 && !d40_tx_is_linked(d40c)
1563 && !d40_residue(d40c)) {
1564 d40_lcla_free_all(d40c, d40d);
1565 d40_desc_load(d40c, d40d);
1566 (void) d40_start(d40c);
1567
1568 if (d40d->lli_current == d40d->lli_len)
1569 d40d->lli_current = 0;
1570 }
1571 } else {
1572 d40_lcla_free_all(d40c, d40d);
1573
1574 if (d40d->lli_current < d40d->lli_len) {
1575 d40_desc_load(d40c, d40d);
1576 /* Start dma job */
1577 (void) d40_start(d40c);
1578 return;
1579 }
1580
1581 if (d40_queue_start(d40c) == NULL) {
1582 d40c->busy = false;
1583
1584 pm_runtime_mark_last_busy(d40c->base->dev);
1585 pm_runtime_put_autosuspend(d40c->base->dev);
1586 }
1587
1588 d40_desc_remove(d40d);
1589 d40_desc_done(d40c, d40d);
1590 }
1591
1592 d40c->pending_tx++;
1593 tasklet_schedule(&d40c->tasklet);
1594
1595}
1596
1597static void dma_tasklet(struct tasklet_struct *t)
1598{
1599 struct d40_chan *d40c = from_tasklet(d40c, t, tasklet);
1600 struct d40_desc *d40d;
1601 unsigned long flags;
1602 bool callback_active;
1603 struct dmaengine_desc_callback cb;
1604
1605 spin_lock_irqsave(&d40c->lock, flags);
1606
1607 /* Get first entry from the done list */
1608 d40d = d40_first_done(d40c);
1609 if (d40d == NULL) {
1610 /* Check if we have reached here for cyclic job */
1611 d40d = d40_first_active_get(d40c);
1612 if (d40d == NULL || !d40d->cyclic)
1613 goto check_pending_tx;
1614 }
1615
1616 if (!d40d->cyclic)
1617 dma_cookie_complete(&d40d->txd);
1618
1619 /*
1620 * If terminating a channel pending_tx is set to zero.
1621 * This prevents any finished active jobs to return to the client.
1622 */
1623 if (d40c->pending_tx == 0) {
1624 spin_unlock_irqrestore(&d40c->lock, flags);
1625 return;
1626 }
1627
1628 /* Callback to client */
1629 callback_active = !!(d40d->txd.flags & DMA_PREP_INTERRUPT);
1630 dmaengine_desc_get_callback(&d40d->txd, &cb);
1631
1632 if (!d40d->cyclic) {
1633 if (async_tx_test_ack(&d40d->txd)) {
1634 d40_desc_remove(d40d);
1635 d40_desc_free(d40c, d40d);
1636 } else if (!d40d->is_in_client_list) {
1637 d40_desc_remove(d40d);
1638 d40_lcla_free_all(d40c, d40d);
1639 list_add_tail(&d40d->node, &d40c->client);
1640 d40d->is_in_client_list = true;
1641 }
1642 }
1643
1644 d40c->pending_tx--;
1645
1646 if (d40c->pending_tx)
1647 tasklet_schedule(&d40c->tasklet);
1648
1649 spin_unlock_irqrestore(&d40c->lock, flags);
1650
1651 if (callback_active)
1652 dmaengine_desc_callback_invoke(&cb, NULL);
1653
1654 return;
1655 check_pending_tx:
1656 /* Rescue maneuver if receiving double interrupts */
1657 if (d40c->pending_tx > 0)
1658 d40c->pending_tx--;
1659 spin_unlock_irqrestore(&d40c->lock, flags);
1660}
1661
1662static irqreturn_t d40_handle_interrupt(int irq, void *data)
1663{
1664 int i;
1665 u32 idx;
1666 u32 row;
1667 long chan = -1;
1668 struct d40_chan *d40c;
1669 struct d40_base *base = data;
1670 u32 *regs = base->regs_interrupt;
1671 struct d40_interrupt_lookup *il = base->gen_dmac.il;
1672 u32 il_size = base->gen_dmac.il_size;
1673
1674 spin_lock(&base->interrupt_lock);
1675
1676 /* Read interrupt status of both logical and physical channels */
1677 for (i = 0; i < il_size; i++)
1678 regs[i] = readl(base->virtbase + il[i].src);
1679
1680 for (;;) {
1681
1682 chan = find_next_bit((unsigned long *)regs,
1683 BITS_PER_LONG * il_size, chan + 1);
1684
1685 /* No more set bits found? */
1686 if (chan == BITS_PER_LONG * il_size)
1687 break;
1688
1689 row = chan / BITS_PER_LONG;
1690 idx = chan & (BITS_PER_LONG - 1);
1691
1692 if (il[row].offset == D40_PHY_CHAN)
1693 d40c = base->lookup_phy_chans[idx];
1694 else
1695 d40c = base->lookup_log_chans[il[row].offset + idx];
1696
1697 if (!d40c) {
1698 /*
1699 * No error because this can happen if something else
1700 * in the system is using the channel.
1701 */
1702 continue;
1703 }
1704
1705 /* ACK interrupt */
1706 writel(BIT(idx), base->virtbase + il[row].clr);
1707
1708 spin_lock(&d40c->lock);
1709
1710 if (!il[row].is_error)
1711 dma_tc_handle(d40c);
1712 else
1713 d40_err(base->dev, "IRQ chan: %ld offset %d idx %d\n",
1714 chan, il[row].offset, idx);
1715
1716 spin_unlock(&d40c->lock);
1717 }
1718
1719 spin_unlock(&base->interrupt_lock);
1720
1721 return IRQ_HANDLED;
1722}
1723
1724static int d40_validate_conf(struct d40_chan *d40c,
1725 struct stedma40_chan_cfg *conf)
1726{
1727 int res = 0;
1728 bool is_log = conf->mode == STEDMA40_MODE_LOGICAL;
1729
1730 if (!conf->dir) {
1731 chan_err(d40c, "Invalid direction.\n");
1732 res = -EINVAL;
1733 }
1734
1735 if ((is_log && conf->dev_type > d40c->base->num_log_chans) ||
1736 (!is_log && conf->dev_type > d40c->base->num_phy_chans) ||
1737 (conf->dev_type < 0)) {
1738 chan_err(d40c, "Invalid device type (%d)\n", conf->dev_type);
1739 res = -EINVAL;
1740 }
1741
1742 if (conf->dir == DMA_DEV_TO_DEV) {
1743 /*
1744 * DMAC HW supports it. Will be added to this driver,
1745 * in case any dma client requires it.
1746 */
1747 chan_err(d40c, "periph to periph not supported\n");
1748 res = -EINVAL;
1749 }
1750
1751 if (d40_psize_2_burst_size(is_log, conf->src_info.psize) *
1752 conf->src_info.data_width !=
1753 d40_psize_2_burst_size(is_log, conf->dst_info.psize) *
1754 conf->dst_info.data_width) {
1755 /*
1756 * The DMAC hardware only supports
1757 * src (burst x width) == dst (burst x width)
1758 */
1759
1760 chan_err(d40c, "src (burst x width) != dst (burst x width)\n");
1761 res = -EINVAL;
1762 }
1763
1764 return res;
1765}
1766
1767static bool d40_alloc_mask_set(struct d40_phy_res *phy,
1768 bool is_src, int log_event_line, bool is_log,
1769 bool *first_user)
1770{
1771 unsigned long flags;
1772 spin_lock_irqsave(&phy->lock, flags);
1773
1774 *first_user = ((phy->allocated_src | phy->allocated_dst)
1775 == D40_ALLOC_FREE);
1776
1777 if (!is_log) {
1778 /* Physical interrupts are masked per physical full channel */
1779 if (phy->allocated_src == D40_ALLOC_FREE &&
1780 phy->allocated_dst == D40_ALLOC_FREE) {
1781 phy->allocated_dst = D40_ALLOC_PHY;
1782 phy->allocated_src = D40_ALLOC_PHY;
1783 goto found_unlock;
1784 } else
1785 goto not_found_unlock;
1786 }
1787
1788 /* Logical channel */
1789 if (is_src) {
1790 if (phy->allocated_src == D40_ALLOC_PHY)
1791 goto not_found_unlock;
1792
1793 if (phy->allocated_src == D40_ALLOC_FREE)
1794 phy->allocated_src = D40_ALLOC_LOG_FREE;
1795
1796 if (!(phy->allocated_src & BIT(log_event_line))) {
1797 phy->allocated_src |= BIT(log_event_line);
1798 goto found_unlock;
1799 } else
1800 goto not_found_unlock;
1801 } else {
1802 if (phy->allocated_dst == D40_ALLOC_PHY)
1803 goto not_found_unlock;
1804
1805 if (phy->allocated_dst == D40_ALLOC_FREE)
1806 phy->allocated_dst = D40_ALLOC_LOG_FREE;
1807
1808 if (!(phy->allocated_dst & BIT(log_event_line))) {
1809 phy->allocated_dst |= BIT(log_event_line);
1810 goto found_unlock;
1811 }
1812 }
1813 not_found_unlock:
1814 spin_unlock_irqrestore(&phy->lock, flags);
1815 return false;
1816 found_unlock:
1817 spin_unlock_irqrestore(&phy->lock, flags);
1818 return true;
1819}
1820
1821static bool d40_alloc_mask_free(struct d40_phy_res *phy, bool is_src,
1822 int log_event_line)
1823{
1824 unsigned long flags;
1825 bool is_free = false;
1826
1827 spin_lock_irqsave(&phy->lock, flags);
1828 if (!log_event_line) {
1829 phy->allocated_dst = D40_ALLOC_FREE;
1830 phy->allocated_src = D40_ALLOC_FREE;
1831 is_free = true;
1832 goto unlock;
1833 }
1834
1835 /* Logical channel */
1836 if (is_src) {
1837 phy->allocated_src &= ~BIT(log_event_line);
1838 if (phy->allocated_src == D40_ALLOC_LOG_FREE)
1839 phy->allocated_src = D40_ALLOC_FREE;
1840 } else {
1841 phy->allocated_dst &= ~BIT(log_event_line);
1842 if (phy->allocated_dst == D40_ALLOC_LOG_FREE)
1843 phy->allocated_dst = D40_ALLOC_FREE;
1844 }
1845
1846 is_free = ((phy->allocated_src | phy->allocated_dst) ==
1847 D40_ALLOC_FREE);
1848 unlock:
1849 spin_unlock_irqrestore(&phy->lock, flags);
1850
1851 return is_free;
1852}
1853
1854static int d40_allocate_channel(struct d40_chan *d40c, bool *first_phy_user)
1855{
1856 int dev_type = d40c->dma_cfg.dev_type;
1857 int event_group;
1858 int event_line;
1859 struct d40_phy_res *phys;
1860 int i;
1861 int j;
1862 int log_num;
1863 int num_phy_chans;
1864 bool is_src;
1865 bool is_log = d40c->dma_cfg.mode == STEDMA40_MODE_LOGICAL;
1866
1867 phys = d40c->base->phy_res;
1868 num_phy_chans = d40c->base->num_phy_chans;
1869
1870 if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
1871 log_num = 2 * dev_type;
1872 is_src = true;
1873 } else if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
1874 d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1875 /* dst event lines are used for logical memcpy */
1876 log_num = 2 * dev_type + 1;
1877 is_src = false;
1878 } else
1879 return -EINVAL;
1880
1881 event_group = D40_TYPE_TO_GROUP(dev_type);
1882 event_line = D40_TYPE_TO_EVENT(dev_type);
1883
1884 if (!is_log) {
1885 if (d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
1886 /* Find physical half channel */
1887 if (d40c->dma_cfg.use_fixed_channel) {
1888 i = d40c->dma_cfg.phy_channel;
1889 if (d40_alloc_mask_set(&phys[i], is_src,
1890 0, is_log,
1891 first_phy_user))
1892 goto found_phy;
1893 } else {
1894 for (i = 0; i < num_phy_chans; i++) {
1895 if (d40_alloc_mask_set(&phys[i], is_src,
1896 0, is_log,
1897 first_phy_user))
1898 goto found_phy;
1899 }
1900 }
1901 } else
1902 for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1903 int phy_num = j + event_group * 2;
1904 for (i = phy_num; i < phy_num + 2; i++) {
1905 if (d40_alloc_mask_set(&phys[i],
1906 is_src,
1907 0,
1908 is_log,
1909 first_phy_user))
1910 goto found_phy;
1911 }
1912 }
1913 return -EINVAL;
1914found_phy:
1915 d40c->phy_chan = &phys[i];
1916 d40c->log_num = D40_PHY_CHAN;
1917 goto out;
1918 }
1919 if (dev_type == -1)
1920 return -EINVAL;
1921
1922 /* Find logical channel */
1923 for (j = 0; j < d40c->base->num_phy_chans; j += 8) {
1924 int phy_num = j + event_group * 2;
1925
1926 if (d40c->dma_cfg.use_fixed_channel) {
1927 i = d40c->dma_cfg.phy_channel;
1928
1929 if ((i != phy_num) && (i != phy_num + 1)) {
1930 dev_err(chan2dev(d40c),
1931 "invalid fixed phy channel %d\n", i);
1932 return -EINVAL;
1933 }
1934
1935 if (d40_alloc_mask_set(&phys[i], is_src, event_line,
1936 is_log, first_phy_user))
1937 goto found_log;
1938
1939 dev_err(chan2dev(d40c),
1940 "could not allocate fixed phy channel %d\n", i);
1941 return -EINVAL;
1942 }
1943
1944 /*
1945 * Spread logical channels across all available physical rather
1946 * than pack every logical channel at the first available phy
1947 * channels.
1948 */
1949 if (is_src) {
1950 for (i = phy_num; i < phy_num + 2; i++) {
1951 if (d40_alloc_mask_set(&phys[i], is_src,
1952 event_line, is_log,
1953 first_phy_user))
1954 goto found_log;
1955 }
1956 } else {
1957 for (i = phy_num + 1; i >= phy_num; i--) {
1958 if (d40_alloc_mask_set(&phys[i], is_src,
1959 event_line, is_log,
1960 first_phy_user))
1961 goto found_log;
1962 }
1963 }
1964 }
1965 return -EINVAL;
1966
1967found_log:
1968 d40c->phy_chan = &phys[i];
1969 d40c->log_num = log_num;
1970out:
1971
1972 if (is_log)
1973 d40c->base->lookup_log_chans[d40c->log_num] = d40c;
1974 else
1975 d40c->base->lookup_phy_chans[d40c->phy_chan->num] = d40c;
1976
1977 return 0;
1978
1979}
1980
1981static int d40_config_memcpy(struct d40_chan *d40c)
1982{
1983 dma_cap_mask_t cap = d40c->chan.device->cap_mask;
1984
1985 if (dma_has_cap(DMA_MEMCPY, cap) && !dma_has_cap(DMA_SLAVE, cap)) {
1986 d40c->dma_cfg = dma40_memcpy_conf_log;
1987 d40c->dma_cfg.dev_type = dma40_memcpy_channels[d40c->chan.chan_id];
1988
1989 d40_log_cfg(&d40c->dma_cfg,
1990 &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
1991
1992 } else if (dma_has_cap(DMA_MEMCPY, cap) &&
1993 dma_has_cap(DMA_SLAVE, cap)) {
1994 d40c->dma_cfg = dma40_memcpy_conf_phy;
1995
1996 /* Generate interrupt at end of transfer or relink. */
1997 d40c->dst_def_cfg |= BIT(D40_SREG_CFG_TIM_POS);
1998
1999 /* Generate interrupt on error. */
2000 d40c->src_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
2001 d40c->dst_def_cfg |= BIT(D40_SREG_CFG_EIM_POS);
2002
2003 } else {
2004 chan_err(d40c, "No memcpy\n");
2005 return -EINVAL;
2006 }
2007
2008 return 0;
2009}
2010
2011static int d40_free_dma(struct d40_chan *d40c)
2012{
2013
2014 int res = 0;
2015 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2016 struct d40_phy_res *phy = d40c->phy_chan;
2017 bool is_src;
2018
2019 /* Terminate all queued and active transfers */
2020 d40_term_all(d40c);
2021
2022 if (phy == NULL) {
2023 chan_err(d40c, "phy == null\n");
2024 return -EINVAL;
2025 }
2026
2027 if (phy->allocated_src == D40_ALLOC_FREE &&
2028 phy->allocated_dst == D40_ALLOC_FREE) {
2029 chan_err(d40c, "channel already free\n");
2030 return -EINVAL;
2031 }
2032
2033 if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2034 d40c->dma_cfg.dir == DMA_MEM_TO_MEM)
2035 is_src = false;
2036 else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2037 is_src = true;
2038 else {
2039 chan_err(d40c, "Unknown direction\n");
2040 return -EINVAL;
2041 }
2042
2043 pm_runtime_get_sync(d40c->base->dev);
2044 res = d40_channel_execute_command(d40c, D40_DMA_STOP);
2045 if (res) {
2046 chan_err(d40c, "stop failed\n");
2047 goto mark_last_busy;
2048 }
2049
2050 d40_alloc_mask_free(phy, is_src, chan_is_logical(d40c) ? event : 0);
2051
2052 if (chan_is_logical(d40c))
2053 d40c->base->lookup_log_chans[d40c->log_num] = NULL;
2054 else
2055 d40c->base->lookup_phy_chans[phy->num] = NULL;
2056
2057 if (d40c->busy) {
2058 pm_runtime_mark_last_busy(d40c->base->dev);
2059 pm_runtime_put_autosuspend(d40c->base->dev);
2060 }
2061
2062 d40c->busy = false;
2063 d40c->phy_chan = NULL;
2064 d40c->configured = false;
2065 mark_last_busy:
2066 pm_runtime_mark_last_busy(d40c->base->dev);
2067 pm_runtime_put_autosuspend(d40c->base->dev);
2068 return res;
2069}
2070
2071static bool d40_is_paused(struct d40_chan *d40c)
2072{
2073 void __iomem *chanbase = chan_base(d40c);
2074 bool is_paused = false;
2075 unsigned long flags;
2076 void __iomem *active_reg;
2077 u32 status;
2078 u32 event = D40_TYPE_TO_EVENT(d40c->dma_cfg.dev_type);
2079
2080 spin_lock_irqsave(&d40c->lock, flags);
2081
2082 if (chan_is_physical(d40c)) {
2083 if (d40c->phy_chan->num % 2 == 0)
2084 active_reg = d40c->base->virtbase + D40_DREG_ACTIVE;
2085 else
2086 active_reg = d40c->base->virtbase + D40_DREG_ACTIVO;
2087
2088 status = (readl(active_reg) &
2089 D40_CHAN_POS_MASK(d40c->phy_chan->num)) >>
2090 D40_CHAN_POS(d40c->phy_chan->num);
2091 if (status == D40_DMA_SUSPENDED || status == D40_DMA_STOP)
2092 is_paused = true;
2093 goto unlock;
2094 }
2095
2096 if (d40c->dma_cfg.dir == DMA_MEM_TO_DEV ||
2097 d40c->dma_cfg.dir == DMA_MEM_TO_MEM) {
2098 status = readl(chanbase + D40_CHAN_REG_SDLNK);
2099 } else if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM) {
2100 status = readl(chanbase + D40_CHAN_REG_SSLNK);
2101 } else {
2102 chan_err(d40c, "Unknown direction\n");
2103 goto unlock;
2104 }
2105
2106 status = (status & D40_EVENTLINE_MASK(event)) >>
2107 D40_EVENTLINE_POS(event);
2108
2109 if (status != D40_DMA_RUN)
2110 is_paused = true;
2111 unlock:
2112 spin_unlock_irqrestore(&d40c->lock, flags);
2113 return is_paused;
2114
2115}
2116
2117static u32 stedma40_residue(struct dma_chan *chan)
2118{
2119 struct d40_chan *d40c =
2120 container_of(chan, struct d40_chan, chan);
2121 u32 bytes_left;
2122 unsigned long flags;
2123
2124 spin_lock_irqsave(&d40c->lock, flags);
2125 bytes_left = d40_residue(d40c);
2126 spin_unlock_irqrestore(&d40c->lock, flags);
2127
2128 return bytes_left;
2129}
2130
2131static int
2132d40_prep_sg_log(struct d40_chan *chan, struct d40_desc *desc,
2133 struct scatterlist *sg_src, struct scatterlist *sg_dst,
2134 unsigned int sg_len, dma_addr_t src_dev_addr,
2135 dma_addr_t dst_dev_addr)
2136{
2137 struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2138 struct stedma40_half_channel_info *src_info = &cfg->src_info;
2139 struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2140 int ret;
2141
2142 ret = d40_log_sg_to_lli(sg_src, sg_len,
2143 src_dev_addr,
2144 desc->lli_log.src,
2145 chan->log_def.lcsp1,
2146 src_info->data_width,
2147 dst_info->data_width);
2148
2149 ret = d40_log_sg_to_lli(sg_dst, sg_len,
2150 dst_dev_addr,
2151 desc->lli_log.dst,
2152 chan->log_def.lcsp3,
2153 dst_info->data_width,
2154 src_info->data_width);
2155
2156 return ret < 0 ? ret : 0;
2157}
2158
2159static int
2160d40_prep_sg_phy(struct d40_chan *chan, struct d40_desc *desc,
2161 struct scatterlist *sg_src, struct scatterlist *sg_dst,
2162 unsigned int sg_len, dma_addr_t src_dev_addr,
2163 dma_addr_t dst_dev_addr)
2164{
2165 struct stedma40_chan_cfg *cfg = &chan->dma_cfg;
2166 struct stedma40_half_channel_info *src_info = &cfg->src_info;
2167 struct stedma40_half_channel_info *dst_info = &cfg->dst_info;
2168 unsigned long flags = 0;
2169 int ret;
2170
2171 if (desc->cyclic)
2172 flags |= LLI_CYCLIC | LLI_TERM_INT;
2173
2174 ret = d40_phy_sg_to_lli(sg_src, sg_len, src_dev_addr,
2175 desc->lli_phy.src,
2176 virt_to_phys(desc->lli_phy.src),
2177 chan->src_def_cfg,
2178 src_info, dst_info, flags);
2179
2180 ret = d40_phy_sg_to_lli(sg_dst, sg_len, dst_dev_addr,
2181 desc->lli_phy.dst,
2182 virt_to_phys(desc->lli_phy.dst),
2183 chan->dst_def_cfg,
2184 dst_info, src_info, flags);
2185
2186 dma_sync_single_for_device(chan->base->dev, desc->lli_pool.dma_addr,
2187 desc->lli_pool.size, DMA_TO_DEVICE);
2188
2189 return ret < 0 ? ret : 0;
2190}
2191
2192static struct d40_desc *
2193d40_prep_desc(struct d40_chan *chan, struct scatterlist *sg,
2194 unsigned int sg_len, unsigned long dma_flags)
2195{
2196 struct stedma40_chan_cfg *cfg;
2197 struct d40_desc *desc;
2198 int ret;
2199
2200 desc = d40_desc_get(chan);
2201 if (!desc)
2202 return NULL;
2203
2204 cfg = &chan->dma_cfg;
2205 desc->lli_len = d40_sg_2_dmalen(sg, sg_len, cfg->src_info.data_width,
2206 cfg->dst_info.data_width);
2207 if (desc->lli_len < 0) {
2208 chan_err(chan, "Unaligned size\n");
2209 goto free_desc;
2210 }
2211
2212 ret = d40_pool_lli_alloc(chan, desc, desc->lli_len);
2213 if (ret < 0) {
2214 chan_err(chan, "Could not allocate lli\n");
2215 goto free_desc;
2216 }
2217
2218 desc->lli_current = 0;
2219 desc->txd.flags = dma_flags;
2220 desc->txd.tx_submit = d40_tx_submit;
2221
2222 dma_async_tx_descriptor_init(&desc->txd, &chan->chan);
2223
2224 return desc;
2225 free_desc:
2226 d40_desc_free(chan, desc);
2227 return NULL;
2228}
2229
2230static struct dma_async_tx_descriptor *
2231d40_prep_sg(struct dma_chan *dchan, struct scatterlist *sg_src,
2232 struct scatterlist *sg_dst, unsigned int sg_len,
2233 enum dma_transfer_direction direction, unsigned long dma_flags)
2234{
2235 struct d40_chan *chan = container_of(dchan, struct d40_chan, chan);
2236 dma_addr_t src_dev_addr;
2237 dma_addr_t dst_dev_addr;
2238 struct d40_desc *desc;
2239 unsigned long flags;
2240 int ret;
2241
2242 if (!chan->phy_chan) {
2243 chan_err(chan, "Cannot prepare unallocated channel\n");
2244 return NULL;
2245 }
2246
2247 d40_set_runtime_config_write(dchan, &chan->slave_config, direction);
2248
2249 spin_lock_irqsave(&chan->lock, flags);
2250
2251 desc = d40_prep_desc(chan, sg_src, sg_len, dma_flags);
2252 if (desc == NULL)
2253 goto unlock;
2254
2255 if (sg_next(&sg_src[sg_len - 1]) == sg_src)
2256 desc->cyclic = true;
2257
2258 src_dev_addr = 0;
2259 dst_dev_addr = 0;
2260 if (direction == DMA_DEV_TO_MEM)
2261 src_dev_addr = chan->runtime_addr;
2262 else if (direction == DMA_MEM_TO_DEV)
2263 dst_dev_addr = chan->runtime_addr;
2264
2265 if (chan_is_logical(chan))
2266 ret = d40_prep_sg_log(chan, desc, sg_src, sg_dst,
2267 sg_len, src_dev_addr, dst_dev_addr);
2268 else
2269 ret = d40_prep_sg_phy(chan, desc, sg_src, sg_dst,
2270 sg_len, src_dev_addr, dst_dev_addr);
2271
2272 if (ret) {
2273 chan_err(chan, "Failed to prepare %s sg job: %d\n",
2274 chan_is_logical(chan) ? "log" : "phy", ret);
2275 goto free_desc;
2276 }
2277
2278 /*
2279 * add descriptor to the prepare queue in order to be able
2280 * to free them later in terminate_all
2281 */
2282 list_add_tail(&desc->node, &chan->prepare_queue);
2283
2284 spin_unlock_irqrestore(&chan->lock, flags);
2285
2286 return &desc->txd;
2287 free_desc:
2288 d40_desc_free(chan, desc);
2289 unlock:
2290 spin_unlock_irqrestore(&chan->lock, flags);
2291 return NULL;
2292}
2293
2294static bool stedma40_filter(struct dma_chan *chan, void *data)
2295{
2296 struct stedma40_chan_cfg *info = data;
2297 struct d40_chan *d40c =
2298 container_of(chan, struct d40_chan, chan);
2299 int err;
2300
2301 if (data) {
2302 err = d40_validate_conf(d40c, info);
2303 if (!err)
2304 d40c->dma_cfg = *info;
2305 } else
2306 err = d40_config_memcpy(d40c);
2307
2308 if (!err)
2309 d40c->configured = true;
2310
2311 return err == 0;
2312}
2313
2314static void __d40_set_prio_rt(struct d40_chan *d40c, int dev_type, bool src)
2315{
2316 bool realtime = d40c->dma_cfg.realtime;
2317 bool highprio = d40c->dma_cfg.high_priority;
2318 u32 rtreg;
2319 u32 event = D40_TYPE_TO_EVENT(dev_type);
2320 u32 group = D40_TYPE_TO_GROUP(dev_type);
2321 u32 bit = BIT(event);
2322 u32 prioreg;
2323 struct d40_gen_dmac *dmac = &d40c->base->gen_dmac;
2324
2325 rtreg = realtime ? dmac->realtime_en : dmac->realtime_clear;
2326 /*
2327 * Due to a hardware bug, in some cases a logical channel triggered by
2328 * a high priority destination event line can generate extra packet
2329 * transactions.
2330 *
2331 * The workaround is to not set the high priority level for the
2332 * destination event lines that trigger logical channels.
2333 */
2334 if (!src && chan_is_logical(d40c))
2335 highprio = false;
2336
2337 prioreg = highprio ? dmac->high_prio_en : dmac->high_prio_clear;
2338
2339 /* Destination event lines are stored in the upper halfword */
2340 if (!src)
2341 bit <<= 16;
2342
2343 writel(bit, d40c->base->virtbase + prioreg + group * 4);
2344 writel(bit, d40c->base->virtbase + rtreg + group * 4);
2345}
2346
2347static void d40_set_prio_realtime(struct d40_chan *d40c)
2348{
2349 if (d40c->base->rev < 3)
2350 return;
2351
2352 if ((d40c->dma_cfg.dir == DMA_DEV_TO_MEM) ||
2353 (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2354 __d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, true);
2355
2356 if ((d40c->dma_cfg.dir == DMA_MEM_TO_DEV) ||
2357 (d40c->dma_cfg.dir == DMA_DEV_TO_DEV))
2358 __d40_set_prio_rt(d40c, d40c->dma_cfg.dev_type, false);
2359}
2360
2361#define D40_DT_FLAGS_MODE(flags) ((flags >> 0) & 0x1)
2362#define D40_DT_FLAGS_DIR(flags) ((flags >> 1) & 0x1)
2363#define D40_DT_FLAGS_BIG_ENDIAN(flags) ((flags >> 2) & 0x1)
2364#define D40_DT_FLAGS_FIXED_CHAN(flags) ((flags >> 3) & 0x1)
2365#define D40_DT_FLAGS_HIGH_PRIO(flags) ((flags >> 4) & 0x1)
2366
2367static struct dma_chan *d40_xlate(struct of_phandle_args *dma_spec,
2368 struct of_dma *ofdma)
2369{
2370 struct stedma40_chan_cfg cfg;
2371 dma_cap_mask_t cap;
2372 u32 flags;
2373
2374 memset(&cfg, 0, sizeof(struct stedma40_chan_cfg));
2375
2376 dma_cap_zero(cap);
2377 dma_cap_set(DMA_SLAVE, cap);
2378
2379 cfg.dev_type = dma_spec->args[0];
2380 flags = dma_spec->args[2];
2381
2382 switch (D40_DT_FLAGS_MODE(flags)) {
2383 case 0: cfg.mode = STEDMA40_MODE_LOGICAL; break;
2384 case 1: cfg.mode = STEDMA40_MODE_PHYSICAL; break;
2385 }
2386
2387 switch (D40_DT_FLAGS_DIR(flags)) {
2388 case 0:
2389 cfg.dir = DMA_MEM_TO_DEV;
2390 cfg.dst_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2391 break;
2392 case 1:
2393 cfg.dir = DMA_DEV_TO_MEM;
2394 cfg.src_info.big_endian = D40_DT_FLAGS_BIG_ENDIAN(flags);
2395 break;
2396 }
2397
2398 if (D40_DT_FLAGS_FIXED_CHAN(flags)) {
2399 cfg.phy_channel = dma_spec->args[1];
2400 cfg.use_fixed_channel = true;
2401 }
2402
2403 if (D40_DT_FLAGS_HIGH_PRIO(flags))
2404 cfg.high_priority = true;
2405
2406 return dma_request_channel(cap, stedma40_filter, &cfg);
2407}
2408
2409/* DMA ENGINE functions */
2410static int d40_alloc_chan_resources(struct dma_chan *chan)
2411{
2412 int err;
2413 unsigned long flags;
2414 struct d40_chan *d40c =
2415 container_of(chan, struct d40_chan, chan);
2416 bool is_free_phy;
2417 spin_lock_irqsave(&d40c->lock, flags);
2418
2419 dma_cookie_init(chan);
2420
2421 /* If no dma configuration is set use default configuration (memcpy) */
2422 if (!d40c->configured) {
2423 err = d40_config_memcpy(d40c);
2424 if (err) {
2425 chan_err(d40c, "Failed to configure memcpy channel\n");
2426 goto mark_last_busy;
2427 }
2428 }
2429
2430 err = d40_allocate_channel(d40c, &is_free_phy);
2431 if (err) {
2432 chan_err(d40c, "Failed to allocate channel\n");
2433 d40c->configured = false;
2434 goto mark_last_busy;
2435 }
2436
2437 pm_runtime_get_sync(d40c->base->dev);
2438
2439 d40_set_prio_realtime(d40c);
2440
2441 if (chan_is_logical(d40c)) {
2442 if (d40c->dma_cfg.dir == DMA_DEV_TO_MEM)
2443 d40c->lcpa = d40c->base->lcpa_base +
2444 d40c->dma_cfg.dev_type * D40_LCPA_CHAN_SIZE;
2445 else
2446 d40c->lcpa = d40c->base->lcpa_base +
2447 d40c->dma_cfg.dev_type *
2448 D40_LCPA_CHAN_SIZE + D40_LCPA_CHAN_DST_DELTA;
2449
2450 /* Unmask the Global Interrupt Mask. */
2451 d40c->src_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2452 d40c->dst_def_cfg |= BIT(D40_SREG_CFG_LOG_GIM_POS);
2453 }
2454
2455 dev_dbg(chan2dev(d40c), "allocated %s channel (phy %d%s)\n",
2456 chan_is_logical(d40c) ? "logical" : "physical",
2457 d40c->phy_chan->num,
2458 d40c->dma_cfg.use_fixed_channel ? ", fixed" : "");
2459
2460
2461 /*
2462 * Only write channel configuration to the DMA if the physical
2463 * resource is free. In case of multiple logical channels
2464 * on the same physical resource, only the first write is necessary.
2465 */
2466 if (is_free_phy)
2467 d40_config_write(d40c);
2468 mark_last_busy:
2469 pm_runtime_mark_last_busy(d40c->base->dev);
2470 pm_runtime_put_autosuspend(d40c->base->dev);
2471 spin_unlock_irqrestore(&d40c->lock, flags);
2472 return err;
2473}
2474
2475static void d40_free_chan_resources(struct dma_chan *chan)
2476{
2477 struct d40_chan *d40c =
2478 container_of(chan, struct d40_chan, chan);
2479 int err;
2480 unsigned long flags;
2481
2482 if (d40c->phy_chan == NULL) {
2483 chan_err(d40c, "Cannot free unallocated channel\n");
2484 return;
2485 }
2486
2487 spin_lock_irqsave(&d40c->lock, flags);
2488
2489 err = d40_free_dma(d40c);
2490
2491 if (err)
2492 chan_err(d40c, "Failed to free channel\n");
2493 spin_unlock_irqrestore(&d40c->lock, flags);
2494}
2495
2496static struct dma_async_tx_descriptor *d40_prep_memcpy(struct dma_chan *chan,
2497 dma_addr_t dst,
2498 dma_addr_t src,
2499 size_t size,
2500 unsigned long dma_flags)
2501{
2502 struct scatterlist dst_sg;
2503 struct scatterlist src_sg;
2504
2505 sg_init_table(&dst_sg, 1);
2506 sg_init_table(&src_sg, 1);
2507
2508 sg_dma_address(&dst_sg) = dst;
2509 sg_dma_address(&src_sg) = src;
2510
2511 sg_dma_len(&dst_sg) = size;
2512 sg_dma_len(&src_sg) = size;
2513
2514 return d40_prep_sg(chan, &src_sg, &dst_sg, 1,
2515 DMA_MEM_TO_MEM, dma_flags);
2516}
2517
2518static struct dma_async_tx_descriptor *
2519d40_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
2520 unsigned int sg_len, enum dma_transfer_direction direction,
2521 unsigned long dma_flags, void *context)
2522{
2523 if (!is_slave_direction(direction))
2524 return NULL;
2525
2526 return d40_prep_sg(chan, sgl, sgl, sg_len, direction, dma_flags);
2527}
2528
2529static struct dma_async_tx_descriptor *
2530dma40_prep_dma_cyclic(struct dma_chan *chan, dma_addr_t dma_addr,
2531 size_t buf_len, size_t period_len,
2532 enum dma_transfer_direction direction, unsigned long flags)
2533{
2534 unsigned int periods = buf_len / period_len;
2535 struct dma_async_tx_descriptor *txd;
2536 struct scatterlist *sg;
2537 int i;
2538
2539 sg = kcalloc(periods + 1, sizeof(struct scatterlist), GFP_NOWAIT);
2540 if (!sg)
2541 return NULL;
2542
2543 for (i = 0; i < periods; i++) {
2544 sg_dma_address(&sg[i]) = dma_addr;
2545 sg_dma_len(&sg[i]) = period_len;
2546 dma_addr += period_len;
2547 }
2548
2549 sg_chain(sg, periods + 1, sg);
2550
2551 txd = d40_prep_sg(chan, sg, sg, periods, direction,
2552 DMA_PREP_INTERRUPT);
2553
2554 kfree(sg);
2555
2556 return txd;
2557}
2558
2559static enum dma_status d40_tx_status(struct dma_chan *chan,
2560 dma_cookie_t cookie,
2561 struct dma_tx_state *txstate)
2562{
2563 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2564 enum dma_status ret;
2565
2566 if (d40c->phy_chan == NULL) {
2567 chan_err(d40c, "Cannot read status of unallocated channel\n");
2568 return -EINVAL;
2569 }
2570
2571 ret = dma_cookie_status(chan, cookie, txstate);
2572 if (ret != DMA_COMPLETE && txstate)
2573 dma_set_residue(txstate, stedma40_residue(chan));
2574
2575 if (d40_is_paused(d40c))
2576 ret = DMA_PAUSED;
2577
2578 return ret;
2579}
2580
2581static void d40_issue_pending(struct dma_chan *chan)
2582{
2583 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2584 unsigned long flags;
2585
2586 if (d40c->phy_chan == NULL) {
2587 chan_err(d40c, "Channel is not allocated!\n");
2588 return;
2589 }
2590
2591 spin_lock_irqsave(&d40c->lock, flags);
2592
2593 list_splice_tail_init(&d40c->pending_queue, &d40c->queue);
2594
2595 /* Busy means that queued jobs are already being processed */
2596 if (!d40c->busy)
2597 (void) d40_queue_start(d40c);
2598
2599 spin_unlock_irqrestore(&d40c->lock, flags);
2600}
2601
2602static int d40_terminate_all(struct dma_chan *chan)
2603{
2604 unsigned long flags;
2605 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2606 int ret;
2607
2608 if (d40c->phy_chan == NULL) {
2609 chan_err(d40c, "Channel is not allocated!\n");
2610 return -EINVAL;
2611 }
2612
2613 spin_lock_irqsave(&d40c->lock, flags);
2614
2615 pm_runtime_get_sync(d40c->base->dev);
2616 ret = d40_channel_execute_command(d40c, D40_DMA_STOP);
2617 if (ret)
2618 chan_err(d40c, "Failed to stop channel\n");
2619
2620 d40_term_all(d40c);
2621 pm_runtime_mark_last_busy(d40c->base->dev);
2622 pm_runtime_put_autosuspend(d40c->base->dev);
2623 if (d40c->busy) {
2624 pm_runtime_mark_last_busy(d40c->base->dev);
2625 pm_runtime_put_autosuspend(d40c->base->dev);
2626 }
2627 d40c->busy = false;
2628
2629 spin_unlock_irqrestore(&d40c->lock, flags);
2630 return 0;
2631}
2632
2633static int
2634dma40_config_to_halfchannel(struct d40_chan *d40c,
2635 struct stedma40_half_channel_info *info,
2636 u32 maxburst)
2637{
2638 int psize;
2639
2640 if (chan_is_logical(d40c)) {
2641 if (maxburst >= 16)
2642 psize = STEDMA40_PSIZE_LOG_16;
2643 else if (maxburst >= 8)
2644 psize = STEDMA40_PSIZE_LOG_8;
2645 else if (maxburst >= 4)
2646 psize = STEDMA40_PSIZE_LOG_4;
2647 else
2648 psize = STEDMA40_PSIZE_LOG_1;
2649 } else {
2650 if (maxburst >= 16)
2651 psize = STEDMA40_PSIZE_PHY_16;
2652 else if (maxburst >= 8)
2653 psize = STEDMA40_PSIZE_PHY_8;
2654 else if (maxburst >= 4)
2655 psize = STEDMA40_PSIZE_PHY_4;
2656 else
2657 psize = STEDMA40_PSIZE_PHY_1;
2658 }
2659
2660 info->psize = psize;
2661 info->flow_ctrl = STEDMA40_NO_FLOW_CTRL;
2662
2663 return 0;
2664}
2665
2666static int d40_set_runtime_config(struct dma_chan *chan,
2667 struct dma_slave_config *config)
2668{
2669 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2670
2671 memcpy(&d40c->slave_config, config, sizeof(*config));
2672
2673 return 0;
2674}
2675
2676/* Runtime reconfiguration extension */
2677static int d40_set_runtime_config_write(struct dma_chan *chan,
2678 struct dma_slave_config *config,
2679 enum dma_transfer_direction direction)
2680{
2681 struct d40_chan *d40c = container_of(chan, struct d40_chan, chan);
2682 struct stedma40_chan_cfg *cfg = &d40c->dma_cfg;
2683 enum dma_slave_buswidth src_addr_width, dst_addr_width;
2684 dma_addr_t config_addr;
2685 u32 src_maxburst, dst_maxburst;
2686 int ret;
2687
2688 if (d40c->phy_chan == NULL) {
2689 chan_err(d40c, "Channel is not allocated!\n");
2690 return -EINVAL;
2691 }
2692
2693 src_addr_width = config->src_addr_width;
2694 src_maxburst = config->src_maxburst;
2695 dst_addr_width = config->dst_addr_width;
2696 dst_maxburst = config->dst_maxburst;
2697
2698 if (direction == DMA_DEV_TO_MEM) {
2699 config_addr = config->src_addr;
2700
2701 if (cfg->dir != DMA_DEV_TO_MEM)
2702 dev_dbg(d40c->base->dev,
2703 "channel was not configured for peripheral "
2704 "to memory transfer (%d) overriding\n",
2705 cfg->dir);
2706 cfg->dir = DMA_DEV_TO_MEM;
2707
2708 /* Configure the memory side */
2709 if (dst_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2710 dst_addr_width = src_addr_width;
2711 if (dst_maxburst == 0)
2712 dst_maxburst = src_maxburst;
2713
2714 } else if (direction == DMA_MEM_TO_DEV) {
2715 config_addr = config->dst_addr;
2716
2717 if (cfg->dir != DMA_MEM_TO_DEV)
2718 dev_dbg(d40c->base->dev,
2719 "channel was not configured for memory "
2720 "to peripheral transfer (%d) overriding\n",
2721 cfg->dir);
2722 cfg->dir = DMA_MEM_TO_DEV;
2723
2724 /* Configure the memory side */
2725 if (src_addr_width == DMA_SLAVE_BUSWIDTH_UNDEFINED)
2726 src_addr_width = dst_addr_width;
2727 if (src_maxburst == 0)
2728 src_maxburst = dst_maxburst;
2729 } else {
2730 dev_err(d40c->base->dev,
2731 "unrecognized channel direction %d\n",
2732 direction);
2733 return -EINVAL;
2734 }
2735
2736 if (config_addr <= 0) {
2737 dev_err(d40c->base->dev, "no address supplied\n");
2738 return -EINVAL;
2739 }
2740
2741 if (src_maxburst * src_addr_width != dst_maxburst * dst_addr_width) {
2742 dev_err(d40c->base->dev,
2743 "src/dst width/maxburst mismatch: %d*%d != %d*%d\n",
2744 src_maxburst,
2745 src_addr_width,
2746 dst_maxburst,
2747 dst_addr_width);
2748 return -EINVAL;
2749 }
2750
2751 if (src_maxburst > 16) {
2752 src_maxburst = 16;
2753 dst_maxburst = src_maxburst * src_addr_width / dst_addr_width;
2754 } else if (dst_maxburst > 16) {
2755 dst_maxburst = 16;
2756 src_maxburst = dst_maxburst * dst_addr_width / src_addr_width;
2757 }
2758
2759 /* Only valid widths are; 1, 2, 4 and 8. */
2760 if (src_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2761 src_addr_width > DMA_SLAVE_BUSWIDTH_8_BYTES ||
2762 dst_addr_width <= DMA_SLAVE_BUSWIDTH_UNDEFINED ||
2763 dst_addr_width > DMA_SLAVE_BUSWIDTH_8_BYTES ||
2764 !is_power_of_2(src_addr_width) ||
2765 !is_power_of_2(dst_addr_width))
2766 return -EINVAL;
2767
2768 cfg->src_info.data_width = src_addr_width;
2769 cfg->dst_info.data_width = dst_addr_width;
2770
2771 ret = dma40_config_to_halfchannel(d40c, &cfg->src_info,
2772 src_maxburst);
2773 if (ret)
2774 return ret;
2775
2776 ret = dma40_config_to_halfchannel(d40c, &cfg->dst_info,
2777 dst_maxburst);
2778 if (ret)
2779 return ret;
2780
2781 /* Fill in register values */
2782 if (chan_is_logical(d40c))
2783 d40_log_cfg(cfg, &d40c->log_def.lcsp1, &d40c->log_def.lcsp3);
2784 else
2785 d40_phy_cfg(cfg, &d40c->src_def_cfg, &d40c->dst_def_cfg);
2786
2787 /* These settings will take precedence later */
2788 d40c->runtime_addr = config_addr;
2789 d40c->runtime_direction = direction;
2790 dev_dbg(d40c->base->dev,
2791 "configured channel %s for %s, data width %d/%d, "
2792 "maxburst %d/%d elements, LE, no flow control\n",
2793 dma_chan_name(chan),
2794 (direction == DMA_DEV_TO_MEM) ? "RX" : "TX",
2795 src_addr_width, dst_addr_width,
2796 src_maxburst, dst_maxburst);
2797
2798 return 0;
2799}
2800
2801/* Initialization functions */
2802
2803static void __init d40_chan_init(struct d40_base *base, struct dma_device *dma,
2804 struct d40_chan *chans, int offset,
2805 int num_chans)
2806{
2807 int i = 0;
2808 struct d40_chan *d40c;
2809
2810 INIT_LIST_HEAD(&dma->channels);
2811
2812 for (i = offset; i < offset + num_chans; i++) {
2813 d40c = &chans[i];
2814 d40c->base = base;
2815 d40c->chan.device = dma;
2816
2817 spin_lock_init(&d40c->lock);
2818
2819 d40c->log_num = D40_PHY_CHAN;
2820
2821 INIT_LIST_HEAD(&d40c->done);
2822 INIT_LIST_HEAD(&d40c->active);
2823 INIT_LIST_HEAD(&d40c->queue);
2824 INIT_LIST_HEAD(&d40c->pending_queue);
2825 INIT_LIST_HEAD(&d40c->client);
2826 INIT_LIST_HEAD(&d40c->prepare_queue);
2827
2828 tasklet_setup(&d40c->tasklet, dma_tasklet);
2829
2830 list_add_tail(&d40c->chan.device_node,
2831 &dma->channels);
2832 }
2833}
2834
2835static void d40_ops_init(struct d40_base *base, struct dma_device *dev)
2836{
2837 if (dma_has_cap(DMA_SLAVE, dev->cap_mask)) {
2838 dev->device_prep_slave_sg = d40_prep_slave_sg;
2839 dev->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
2840 }
2841
2842 if (dma_has_cap(DMA_MEMCPY, dev->cap_mask)) {
2843 dev->device_prep_dma_memcpy = d40_prep_memcpy;
2844 dev->directions = BIT(DMA_MEM_TO_MEM);
2845 /*
2846 * This controller can only access address at even
2847 * 32bit boundaries, i.e. 2^2
2848 */
2849 dev->copy_align = DMAENGINE_ALIGN_4_BYTES;
2850 }
2851
2852 if (dma_has_cap(DMA_CYCLIC, dev->cap_mask))
2853 dev->device_prep_dma_cyclic = dma40_prep_dma_cyclic;
2854
2855 dev->device_alloc_chan_resources = d40_alloc_chan_resources;
2856 dev->device_free_chan_resources = d40_free_chan_resources;
2857 dev->device_issue_pending = d40_issue_pending;
2858 dev->device_tx_status = d40_tx_status;
2859 dev->device_config = d40_set_runtime_config;
2860 dev->device_pause = d40_pause;
2861 dev->device_resume = d40_resume;
2862 dev->device_terminate_all = d40_terminate_all;
2863 dev->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
2864 dev->dev = base->dev;
2865}
2866
2867static int __init d40_dmaengine_init(struct d40_base *base,
2868 int num_reserved_chans)
2869{
2870 int err ;
2871
2872 d40_chan_init(base, &base->dma_slave, base->log_chans,
2873 0, base->num_log_chans);
2874
2875 dma_cap_zero(base->dma_slave.cap_mask);
2876 dma_cap_set(DMA_SLAVE, base->dma_slave.cap_mask);
2877 dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2878
2879 d40_ops_init(base, &base->dma_slave);
2880
2881 err = dmaenginem_async_device_register(&base->dma_slave);
2882
2883 if (err) {
2884 d40_err(base->dev, "Failed to register slave channels\n");
2885 goto exit;
2886 }
2887
2888 d40_chan_init(base, &base->dma_memcpy, base->log_chans,
2889 base->num_log_chans, base->num_memcpy_chans);
2890
2891 dma_cap_zero(base->dma_memcpy.cap_mask);
2892 dma_cap_set(DMA_MEMCPY, base->dma_memcpy.cap_mask);
2893
2894 d40_ops_init(base, &base->dma_memcpy);
2895
2896 err = dmaenginem_async_device_register(&base->dma_memcpy);
2897
2898 if (err) {
2899 d40_err(base->dev,
2900 "Failed to register memcpy only channels\n");
2901 goto exit;
2902 }
2903
2904 d40_chan_init(base, &base->dma_both, base->phy_chans,
2905 0, num_reserved_chans);
2906
2907 dma_cap_zero(base->dma_both.cap_mask);
2908 dma_cap_set(DMA_SLAVE, base->dma_both.cap_mask);
2909 dma_cap_set(DMA_MEMCPY, base->dma_both.cap_mask);
2910 dma_cap_set(DMA_CYCLIC, base->dma_slave.cap_mask);
2911
2912 d40_ops_init(base, &base->dma_both);
2913 err = dmaenginem_async_device_register(&base->dma_both);
2914
2915 if (err) {
2916 d40_err(base->dev,
2917 "Failed to register logical and physical capable channels\n");
2918 goto exit;
2919 }
2920 return 0;
2921 exit:
2922 return err;
2923}
2924
2925/* Suspend resume functionality */
2926#ifdef CONFIG_PM_SLEEP
2927static int dma40_suspend(struct device *dev)
2928{
2929 struct d40_base *base = dev_get_drvdata(dev);
2930 int ret;
2931
2932 ret = pm_runtime_force_suspend(dev);
2933 if (ret)
2934 return ret;
2935
2936 if (base->lcpa_regulator)
2937 ret = regulator_disable(base->lcpa_regulator);
2938 return ret;
2939}
2940
2941static int dma40_resume(struct device *dev)
2942{
2943 struct d40_base *base = dev_get_drvdata(dev);
2944 int ret = 0;
2945
2946 if (base->lcpa_regulator) {
2947 ret = regulator_enable(base->lcpa_regulator);
2948 if (ret)
2949 return ret;
2950 }
2951
2952 return pm_runtime_force_resume(dev);
2953}
2954#endif
2955
2956#ifdef CONFIG_PM
2957static void dma40_backup(void __iomem *baseaddr, u32 *backup,
2958 u32 *regaddr, int num, bool save)
2959{
2960 int i;
2961
2962 for (i = 0; i < num; i++) {
2963 void __iomem *addr = baseaddr + regaddr[i];
2964
2965 if (save)
2966 backup[i] = readl_relaxed(addr);
2967 else
2968 writel_relaxed(backup[i], addr);
2969 }
2970}
2971
2972static void d40_save_restore_registers(struct d40_base *base, bool save)
2973{
2974 int i;
2975
2976 /* Save/Restore channel specific registers */
2977 for (i = 0; i < base->num_phy_chans; i++) {
2978 void __iomem *addr;
2979 int idx;
2980
2981 if (base->phy_res[i].reserved)
2982 continue;
2983
2984 addr = base->virtbase + D40_DREG_PCBASE + i * D40_DREG_PCDELTA;
2985 idx = i * ARRAY_SIZE(d40_backup_regs_chan);
2986
2987 dma40_backup(addr, &base->reg_val_backup_chan[idx],
2988 d40_backup_regs_chan,
2989 ARRAY_SIZE(d40_backup_regs_chan),
2990 save);
2991 }
2992
2993 /* Save/Restore global registers */
2994 dma40_backup(base->virtbase, base->reg_val_backup,
2995 d40_backup_regs, ARRAY_SIZE(d40_backup_regs),
2996 save);
2997
2998 /* Save/Restore registers only existing on dma40 v3 and later */
2999 if (base->gen_dmac.backup)
3000 dma40_backup(base->virtbase, base->reg_val_backup_v4,
3001 base->gen_dmac.backup,
3002 base->gen_dmac.backup_size,
3003 save);
3004}
3005
3006static int dma40_runtime_suspend(struct device *dev)
3007{
3008 struct d40_base *base = dev_get_drvdata(dev);
3009
3010 d40_save_restore_registers(base, true);
3011
3012 /* Don't disable/enable clocks for v1 due to HW bugs */
3013 if (base->rev != 1)
3014 writel_relaxed(base->gcc_pwr_off_mask,
3015 base->virtbase + D40_DREG_GCC);
3016
3017 return 0;
3018}
3019
3020static int dma40_runtime_resume(struct device *dev)
3021{
3022 struct d40_base *base = dev_get_drvdata(dev);
3023
3024 d40_save_restore_registers(base, false);
3025
3026 writel_relaxed(D40_DREG_GCC_ENABLE_ALL,
3027 base->virtbase + D40_DREG_GCC);
3028 return 0;
3029}
3030#endif
3031
3032static const struct dev_pm_ops dma40_pm_ops = {
3033 SET_LATE_SYSTEM_SLEEP_PM_OPS(dma40_suspend, dma40_resume)
3034 SET_RUNTIME_PM_OPS(dma40_runtime_suspend,
3035 dma40_runtime_resume,
3036 NULL)
3037};
3038
3039/* Initialization functions. */
3040
3041static int __init d40_phy_res_init(struct d40_base *base)
3042{
3043 int i;
3044 int num_phy_chans_avail = 0;
3045 u32 val[2];
3046 int odd_even_bit = -2;
3047 int gcc = D40_DREG_GCC_ENA;
3048
3049 val[0] = readl(base->virtbase + D40_DREG_PRSME);
3050 val[1] = readl(base->virtbase + D40_DREG_PRSMO);
3051
3052 for (i = 0; i < base->num_phy_chans; i++) {
3053 base->phy_res[i].num = i;
3054 odd_even_bit += 2 * ((i % 2) == 0);
3055 if (((val[i % 2] >> odd_even_bit) & 3) == 1) {
3056 /* Mark security only channels as occupied */
3057 base->phy_res[i].allocated_src = D40_ALLOC_PHY;
3058 base->phy_res[i].allocated_dst = D40_ALLOC_PHY;
3059 base->phy_res[i].reserved = true;
3060 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3061 D40_DREG_GCC_SRC);
3062 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(i),
3063 D40_DREG_GCC_DST);
3064
3065
3066 } else {
3067 base->phy_res[i].allocated_src = D40_ALLOC_FREE;
3068 base->phy_res[i].allocated_dst = D40_ALLOC_FREE;
3069 base->phy_res[i].reserved = false;
3070 num_phy_chans_avail++;
3071 }
3072 spin_lock_init(&base->phy_res[i].lock);
3073 }
3074
3075 /* Mark disabled channels as occupied */
3076 for (i = 0; base->plat_data->disabled_channels[i] != -1; i++) {
3077 int chan = base->plat_data->disabled_channels[i];
3078
3079 base->phy_res[chan].allocated_src = D40_ALLOC_PHY;
3080 base->phy_res[chan].allocated_dst = D40_ALLOC_PHY;
3081 base->phy_res[chan].reserved = true;
3082 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3083 D40_DREG_GCC_SRC);
3084 gcc |= D40_DREG_GCC_EVTGRP_ENA(D40_PHYS_TO_GROUP(chan),
3085 D40_DREG_GCC_DST);
3086 num_phy_chans_avail--;
3087 }
3088
3089 /* Mark soft_lli channels */
3090 for (i = 0; i < base->plat_data->num_of_soft_lli_chans; i++) {
3091 int chan = base->plat_data->soft_lli_chans[i];
3092
3093 base->phy_res[chan].use_soft_lli = true;
3094 }
3095
3096 dev_info(base->dev, "%d of %d physical DMA channels available\n",
3097 num_phy_chans_avail, base->num_phy_chans);
3098
3099 /* Verify settings extended vs standard */
3100 val[0] = readl(base->virtbase + D40_DREG_PRTYP);
3101
3102 for (i = 0; i < base->num_phy_chans; i++) {
3103
3104 if (base->phy_res[i].allocated_src == D40_ALLOC_FREE &&
3105 (val[0] & 0x3) != 1)
3106 dev_info(base->dev,
3107 "[%s] INFO: channel %d is misconfigured (%d)\n",
3108 __func__, i, val[0] & 0x3);
3109
3110 val[0] = val[0] >> 2;
3111 }
3112
3113 /*
3114 * To keep things simple, Enable all clocks initially.
3115 * The clocks will get managed later post channel allocation.
3116 * The clocks for the event lines on which reserved channels exists
3117 * are not managed here.
3118 */
3119 writel(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3120 base->gcc_pwr_off_mask = gcc;
3121
3122 return num_phy_chans_avail;
3123}
3124
3125/* Called from the registered devm action */
3126static void d40_drop_kmem_cache_action(void *d)
3127{
3128 struct kmem_cache *desc_slab = d;
3129
3130 kmem_cache_destroy(desc_slab);
3131}
3132
3133static int __init d40_hw_detect_init(struct platform_device *pdev,
3134 struct d40_base **retbase)
3135{
3136 struct stedma40_platform_data *plat_data = dev_get_platdata(&pdev->dev);
3137 struct device *dev = &pdev->dev;
3138 struct clk *clk;
3139 void __iomem *virtbase;
3140 struct d40_base *base;
3141 int num_log_chans;
3142 int num_phy_chans;
3143 int num_memcpy_chans;
3144 int i;
3145 u32 pid;
3146 u32 cid;
3147 u8 rev;
3148 int ret;
3149
3150 clk = devm_clk_get_enabled(dev, NULL);
3151 if (IS_ERR(clk))
3152 return PTR_ERR(clk);
3153
3154 /* Get IO for DMAC base address */
3155 virtbase = devm_platform_ioremap_resource_byname(pdev, "base");
3156 if (IS_ERR(virtbase))
3157 return PTR_ERR(virtbase);
3158
3159 /* This is just a regular AMBA PrimeCell ID actually */
3160 for (pid = 0, i = 0; i < 4; i++)
3161 pid |= (readl(virtbase + SZ_4K - 0x20 + 4 * i)
3162 & 255) << (i * 8);
3163 for (cid = 0, i = 0; i < 4; i++)
3164 cid |= (readl(virtbase + SZ_4K - 0x10 + 4 * i)
3165 & 255) << (i * 8);
3166
3167 if (cid != AMBA_CID) {
3168 d40_err(dev, "Unknown hardware! No PrimeCell ID\n");
3169 return -EINVAL;
3170 }
3171 if (AMBA_MANF_BITS(pid) != AMBA_VENDOR_ST) {
3172 d40_err(dev, "Unknown designer! Got %x wanted %x\n",
3173 AMBA_MANF_BITS(pid),
3174 AMBA_VENDOR_ST);
3175 return -EINVAL;
3176 }
3177 /*
3178 * HW revision:
3179 * DB8500ed has revision 0
3180 * ? has revision 1
3181 * DB8500v1 has revision 2
3182 * DB8500v2 has revision 3
3183 * AP9540v1 has revision 4
3184 * DB8540v1 has revision 4
3185 */
3186 rev = AMBA_REV_BITS(pid);
3187 if (rev < 2) {
3188 d40_err(dev, "hardware revision: %d is not supported", rev);
3189 return -EINVAL;
3190 }
3191
3192 /* The number of physical channels on this HW */
3193 if (plat_data->num_of_phy_chans)
3194 num_phy_chans = plat_data->num_of_phy_chans;
3195 else
3196 num_phy_chans = 4 * (readl(virtbase + D40_DREG_ICFG) & 0x7) + 4;
3197
3198 /* The number of channels used for memcpy */
3199 if (plat_data->num_of_memcpy_chans)
3200 num_memcpy_chans = plat_data->num_of_memcpy_chans;
3201 else
3202 num_memcpy_chans = ARRAY_SIZE(dma40_memcpy_channels);
3203
3204 num_log_chans = num_phy_chans * D40_MAX_LOG_CHAN_PER_PHY;
3205
3206 dev_info(dev,
3207 "hardware rev: %d with %d physical and %d logical channels\n",
3208 rev, num_phy_chans, num_log_chans);
3209
3210 base = devm_kzalloc(dev,
3211 ALIGN(sizeof(struct d40_base), 4) +
3212 (num_phy_chans + num_log_chans + num_memcpy_chans) *
3213 sizeof(struct d40_chan), GFP_KERNEL);
3214
3215 if (!base)
3216 return -ENOMEM;
3217
3218 base->rev = rev;
3219 base->clk = clk;
3220 base->num_memcpy_chans = num_memcpy_chans;
3221 base->num_phy_chans = num_phy_chans;
3222 base->num_log_chans = num_log_chans;
3223 base->virtbase = virtbase;
3224 base->plat_data = plat_data;
3225 base->dev = dev;
3226 base->phy_chans = ((void *)base) + ALIGN(sizeof(struct d40_base), 4);
3227 base->log_chans = &base->phy_chans[num_phy_chans];
3228
3229 if (base->plat_data->num_of_phy_chans == 14) {
3230 base->gen_dmac.backup = d40_backup_regs_v4b;
3231 base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4B;
3232 base->gen_dmac.interrupt_en = D40_DREG_CPCMIS;
3233 base->gen_dmac.interrupt_clear = D40_DREG_CPCICR;
3234 base->gen_dmac.realtime_en = D40_DREG_CRSEG1;
3235 base->gen_dmac.realtime_clear = D40_DREG_CRCEG1;
3236 base->gen_dmac.high_prio_en = D40_DREG_CPSEG1;
3237 base->gen_dmac.high_prio_clear = D40_DREG_CPCEG1;
3238 base->gen_dmac.il = il_v4b;
3239 base->gen_dmac.il_size = ARRAY_SIZE(il_v4b);
3240 base->gen_dmac.init_reg = dma_init_reg_v4b;
3241 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4b);
3242 } else {
3243 if (base->rev >= 3) {
3244 base->gen_dmac.backup = d40_backup_regs_v4a;
3245 base->gen_dmac.backup_size = BACKUP_REGS_SZ_V4A;
3246 }
3247 base->gen_dmac.interrupt_en = D40_DREG_PCMIS;
3248 base->gen_dmac.interrupt_clear = D40_DREG_PCICR;
3249 base->gen_dmac.realtime_en = D40_DREG_RSEG1;
3250 base->gen_dmac.realtime_clear = D40_DREG_RCEG1;
3251 base->gen_dmac.high_prio_en = D40_DREG_PSEG1;
3252 base->gen_dmac.high_prio_clear = D40_DREG_PCEG1;
3253 base->gen_dmac.il = il_v4a;
3254 base->gen_dmac.il_size = ARRAY_SIZE(il_v4a);
3255 base->gen_dmac.init_reg = dma_init_reg_v4a;
3256 base->gen_dmac.init_reg_size = ARRAY_SIZE(dma_init_reg_v4a);
3257 }
3258
3259 base->phy_res = devm_kcalloc(dev, num_phy_chans,
3260 sizeof(*base->phy_res),
3261 GFP_KERNEL);
3262 if (!base->phy_res)
3263 return -ENOMEM;
3264
3265 base->lookup_phy_chans = devm_kcalloc(dev, num_phy_chans,
3266 sizeof(*base->lookup_phy_chans),
3267 GFP_KERNEL);
3268 if (!base->lookup_phy_chans)
3269 return -ENOMEM;
3270
3271 base->lookup_log_chans = devm_kcalloc(dev, num_log_chans,
3272 sizeof(*base->lookup_log_chans),
3273 GFP_KERNEL);
3274 if (!base->lookup_log_chans)
3275 return -ENOMEM;
3276
3277 base->reg_val_backup_chan = devm_kmalloc_array(dev, base->num_phy_chans,
3278 sizeof(d40_backup_regs_chan),
3279 GFP_KERNEL);
3280 if (!base->reg_val_backup_chan)
3281 return -ENOMEM;
3282
3283 base->lcla_pool.alloc_map = devm_kcalloc(dev, num_phy_chans
3284 * D40_LCLA_LINK_PER_EVENT_GRP,
3285 sizeof(*base->lcla_pool.alloc_map),
3286 GFP_KERNEL);
3287 if (!base->lcla_pool.alloc_map)
3288 return -ENOMEM;
3289
3290 base->regs_interrupt = devm_kmalloc_array(dev, base->gen_dmac.il_size,
3291 sizeof(*base->regs_interrupt),
3292 GFP_KERNEL);
3293 if (!base->regs_interrupt)
3294 return -ENOMEM;
3295
3296 base->desc_slab = kmem_cache_create(D40_NAME, sizeof(struct d40_desc),
3297 0, SLAB_HWCACHE_ALIGN,
3298 NULL);
3299 if (!base->desc_slab)
3300 return -ENOMEM;
3301
3302 ret = devm_add_action_or_reset(dev, d40_drop_kmem_cache_action,
3303 base->desc_slab);
3304 if (ret)
3305 return ret;
3306
3307 *retbase = base;
3308
3309 return 0;
3310}
3311
3312static void __init d40_hw_init(struct d40_base *base)
3313{
3314
3315 int i;
3316 u32 prmseo[2] = {0, 0};
3317 u32 activeo[2] = {0xFFFFFFFF, 0xFFFFFFFF};
3318 u32 pcmis = 0;
3319 u32 pcicr = 0;
3320 struct d40_reg_val *dma_init_reg = base->gen_dmac.init_reg;
3321 u32 reg_size = base->gen_dmac.init_reg_size;
3322
3323 for (i = 0; i < reg_size; i++)
3324 writel(dma_init_reg[i].val,
3325 base->virtbase + dma_init_reg[i].reg);
3326
3327 /* Configure all our dma channels to default settings */
3328 for (i = 0; i < base->num_phy_chans; i++) {
3329
3330 activeo[i % 2] = activeo[i % 2] << 2;
3331
3332 if (base->phy_res[base->num_phy_chans - i - 1].allocated_src
3333 == D40_ALLOC_PHY) {
3334 activeo[i % 2] |= 3;
3335 continue;
3336 }
3337
3338 /* Enable interrupt # */
3339 pcmis = (pcmis << 1) | 1;
3340
3341 /* Clear interrupt # */
3342 pcicr = (pcicr << 1) | 1;
3343
3344 /* Set channel to physical mode */
3345 prmseo[i % 2] = prmseo[i % 2] << 2;
3346 prmseo[i % 2] |= 1;
3347
3348 }
3349
3350 writel(prmseo[1], base->virtbase + D40_DREG_PRMSE);
3351 writel(prmseo[0], base->virtbase + D40_DREG_PRMSO);
3352 writel(activeo[1], base->virtbase + D40_DREG_ACTIVE);
3353 writel(activeo[0], base->virtbase + D40_DREG_ACTIVO);
3354
3355 /* Write which interrupt to enable */
3356 writel(pcmis, base->virtbase + base->gen_dmac.interrupt_en);
3357
3358 /* Write which interrupt to clear */
3359 writel(pcicr, base->virtbase + base->gen_dmac.interrupt_clear);
3360
3361 /* These are __initdata and cannot be accessed after init */
3362 base->gen_dmac.init_reg = NULL;
3363 base->gen_dmac.init_reg_size = 0;
3364}
3365
3366static int __init d40_lcla_allocate(struct d40_base *base)
3367{
3368 struct d40_lcla_pool *pool = &base->lcla_pool;
3369 unsigned long *page_list;
3370 int i, j;
3371 int ret;
3372
3373 /*
3374 * This is somewhat ugly. We need 8192 bytes that are 18 bit aligned,
3375 * To full fill this hardware requirement without wasting 256 kb
3376 * we allocate pages until we get an aligned one.
3377 */
3378 page_list = kmalloc_array(MAX_LCLA_ALLOC_ATTEMPTS,
3379 sizeof(*page_list),
3380 GFP_KERNEL);
3381 if (!page_list)
3382 return -ENOMEM;
3383
3384 /* Calculating how many pages that are required */
3385 base->lcla_pool.pages = SZ_1K * base->num_phy_chans / PAGE_SIZE;
3386
3387 for (i = 0; i < MAX_LCLA_ALLOC_ATTEMPTS; i++) {
3388 page_list[i] = __get_free_pages(GFP_KERNEL,
3389 base->lcla_pool.pages);
3390 if (!page_list[i]) {
3391
3392 d40_err(base->dev, "Failed to allocate %d pages.\n",
3393 base->lcla_pool.pages);
3394 ret = -ENOMEM;
3395
3396 for (j = 0; j < i; j++)
3397 free_pages(page_list[j], base->lcla_pool.pages);
3398 goto free_page_list;
3399 }
3400
3401 if ((virt_to_phys((void *)page_list[i]) &
3402 (LCLA_ALIGNMENT - 1)) == 0)
3403 break;
3404 }
3405
3406 for (j = 0; j < i; j++)
3407 free_pages(page_list[j], base->lcla_pool.pages);
3408
3409 if (i < MAX_LCLA_ALLOC_ATTEMPTS) {
3410 base->lcla_pool.base = (void *)page_list[i];
3411 } else {
3412 /*
3413 * After many attempts and no success with finding the correct
3414 * alignment, try with allocating a big buffer.
3415 */
3416 dev_warn(base->dev,
3417 "[%s] Failed to get %d pages @ 18 bit align.\n",
3418 __func__, base->lcla_pool.pages);
3419 base->lcla_pool.base_unaligned = kmalloc(SZ_1K *
3420 base->num_phy_chans +
3421 LCLA_ALIGNMENT,
3422 GFP_KERNEL);
3423 if (!base->lcla_pool.base_unaligned) {
3424 ret = -ENOMEM;
3425 goto free_page_list;
3426 }
3427
3428 base->lcla_pool.base = PTR_ALIGN(base->lcla_pool.base_unaligned,
3429 LCLA_ALIGNMENT);
3430 }
3431
3432 pool->dma_addr = dma_map_single(base->dev, pool->base,
3433 SZ_1K * base->num_phy_chans,
3434 DMA_TO_DEVICE);
3435 if (dma_mapping_error(base->dev, pool->dma_addr)) {
3436 pool->dma_addr = 0;
3437 ret = -ENOMEM;
3438 goto free_page_list;
3439 }
3440
3441 writel(virt_to_phys(base->lcla_pool.base),
3442 base->virtbase + D40_DREG_LCLA);
3443 ret = 0;
3444 free_page_list:
3445 kfree(page_list);
3446 return ret;
3447}
3448
3449static int __init d40_of_probe(struct device *dev,
3450 struct device_node *np)
3451{
3452 struct stedma40_platform_data *pdata;
3453 int num_phy = 0, num_memcpy = 0, num_disabled = 0;
3454 const __be32 *list;
3455
3456 pdata = devm_kzalloc(dev, sizeof(*pdata), GFP_KERNEL);
3457 if (!pdata)
3458 return -ENOMEM;
3459
3460 /* If absent this value will be obtained from h/w. */
3461 of_property_read_u32(np, "dma-channels", &num_phy);
3462 if (num_phy > 0)
3463 pdata->num_of_phy_chans = num_phy;
3464
3465 list = of_get_property(np, "memcpy-channels", &num_memcpy);
3466 num_memcpy /= sizeof(*list);
3467
3468 if (num_memcpy > D40_MEMCPY_MAX_CHANS || num_memcpy <= 0) {
3469 d40_err(dev,
3470 "Invalid number of memcpy channels specified (%d)\n",
3471 num_memcpy);
3472 return -EINVAL;
3473 }
3474 pdata->num_of_memcpy_chans = num_memcpy;
3475
3476 of_property_read_u32_array(np, "memcpy-channels",
3477 dma40_memcpy_channels,
3478 num_memcpy);
3479
3480 list = of_get_property(np, "disabled-channels", &num_disabled);
3481 num_disabled /= sizeof(*list);
3482
3483 if (num_disabled >= STEDMA40_MAX_PHYS || num_disabled < 0) {
3484 d40_err(dev,
3485 "Invalid number of disabled channels specified (%d)\n",
3486 num_disabled);
3487 return -EINVAL;
3488 }
3489
3490 of_property_read_u32_array(np, "disabled-channels",
3491 pdata->disabled_channels,
3492 num_disabled);
3493 pdata->disabled_channels[num_disabled] = -1;
3494
3495 dev->platform_data = pdata;
3496
3497 return 0;
3498}
3499
3500static int __init d40_probe(struct platform_device *pdev)
3501{
3502 struct device *dev = &pdev->dev;
3503 struct device_node *np = pdev->dev.of_node;
3504 struct device_node *np_lcpa;
3505 struct d40_base *base;
3506 struct resource *res;
3507 struct resource res_lcpa;
3508 int num_reserved_chans;
3509 u32 val;
3510 int ret;
3511
3512 if (d40_of_probe(dev, np)) {
3513 ret = -ENOMEM;
3514 goto report_failure;
3515 }
3516
3517 ret = d40_hw_detect_init(pdev, &base);
3518 if (ret)
3519 goto report_failure;
3520
3521 num_reserved_chans = d40_phy_res_init(base);
3522
3523 platform_set_drvdata(pdev, base);
3524
3525 spin_lock_init(&base->interrupt_lock);
3526 spin_lock_init(&base->execmd_lock);
3527
3528 /* Get IO for logical channel parameter address (LCPA) */
3529 np_lcpa = of_parse_phandle(np, "sram", 0);
3530 if (!np_lcpa) {
3531 dev_err(dev, "no LCPA SRAM node\n");
3532 ret = -EINVAL;
3533 goto report_failure;
3534 }
3535 /* This is no device so read the address directly from the node */
3536 ret = of_address_to_resource(np_lcpa, 0, &res_lcpa);
3537 if (ret) {
3538 dev_err(dev, "no LCPA SRAM resource\n");
3539 goto report_failure;
3540 }
3541 base->lcpa_size = resource_size(&res_lcpa);
3542 base->phy_lcpa = res_lcpa.start;
3543 dev_info(dev, "found LCPA SRAM at %pad, size %pa\n",
3544 &base->phy_lcpa, &base->lcpa_size);
3545
3546 /* We make use of ESRAM memory for this. */
3547 val = readl(base->virtbase + D40_DREG_LCPA);
3548 if (base->phy_lcpa != val && val != 0) {
3549 dev_warn(dev,
3550 "[%s] Mismatch LCPA dma 0x%x, def %08x\n",
3551 __func__, val, (u32)base->phy_lcpa);
3552 } else
3553 writel(base->phy_lcpa, base->virtbase + D40_DREG_LCPA);
3554
3555 base->lcpa_base = devm_ioremap(dev, base->phy_lcpa, base->lcpa_size);
3556 if (!base->lcpa_base) {
3557 ret = -ENOMEM;
3558 d40_err(dev, "Failed to ioremap LCPA region\n");
3559 goto report_failure;
3560 }
3561 /* If lcla has to be located in ESRAM we don't need to allocate */
3562 if (base->plat_data->use_esram_lcla) {
3563 res = platform_get_resource_byname(pdev, IORESOURCE_MEM,
3564 "lcla_esram");
3565 if (!res) {
3566 ret = -ENOENT;
3567 d40_err(dev,
3568 "No \"lcla_esram\" memory resource\n");
3569 goto report_failure;
3570 }
3571 base->lcla_pool.base = devm_ioremap(dev, res->start,
3572 resource_size(res));
3573 if (!base->lcla_pool.base) {
3574 ret = -ENOMEM;
3575 d40_err(dev, "Failed to ioremap LCLA region\n");
3576 goto report_failure;
3577 }
3578 writel(res->start, base->virtbase + D40_DREG_LCLA);
3579
3580 } else {
3581 ret = d40_lcla_allocate(base);
3582 if (ret) {
3583 d40_err(dev, "Failed to allocate LCLA area\n");
3584 goto destroy_cache;
3585 }
3586 }
3587
3588 spin_lock_init(&base->lcla_pool.lock);
3589
3590 base->irq = platform_get_irq(pdev, 0);
3591 if (base->irq < 0) {
3592 ret = base->irq;
3593 goto destroy_cache;
3594 }
3595
3596 ret = request_irq(base->irq, d40_handle_interrupt, 0, D40_NAME, base);
3597 if (ret) {
3598 d40_err(dev, "No IRQ defined\n");
3599 goto destroy_cache;
3600 }
3601
3602 if (base->plat_data->use_esram_lcla) {
3603
3604 base->lcpa_regulator = regulator_get(base->dev, "lcla_esram");
3605 if (IS_ERR(base->lcpa_regulator)) {
3606 d40_err(dev, "Failed to get lcpa_regulator\n");
3607 ret = PTR_ERR(base->lcpa_regulator);
3608 base->lcpa_regulator = NULL;
3609 goto destroy_cache;
3610 }
3611
3612 ret = regulator_enable(base->lcpa_regulator);
3613 if (ret) {
3614 d40_err(dev,
3615 "Failed to enable lcpa_regulator\n");
3616 regulator_put(base->lcpa_regulator);
3617 base->lcpa_regulator = NULL;
3618 goto destroy_cache;
3619 }
3620 }
3621
3622 writel_relaxed(D40_DREG_GCC_ENABLE_ALL, base->virtbase + D40_DREG_GCC);
3623
3624 pm_runtime_irq_safe(base->dev);
3625 pm_runtime_set_autosuspend_delay(base->dev, DMA40_AUTOSUSPEND_DELAY);
3626 pm_runtime_use_autosuspend(base->dev);
3627 pm_runtime_mark_last_busy(base->dev);
3628 pm_runtime_set_active(base->dev);
3629 pm_runtime_enable(base->dev);
3630
3631 ret = d40_dmaengine_init(base, num_reserved_chans);
3632 if (ret)
3633 goto destroy_cache;
3634
3635 ret = dma_set_max_seg_size(base->dev, STEDMA40_MAX_SEG_SIZE);
3636 if (ret) {
3637 d40_err(dev, "Failed to set dma max seg size\n");
3638 goto destroy_cache;
3639 }
3640
3641 d40_hw_init(base);
3642
3643 ret = of_dma_controller_register(np, d40_xlate, NULL);
3644 if (ret) {
3645 dev_err(dev,
3646 "could not register of_dma_controller\n");
3647 goto destroy_cache;
3648 }
3649
3650 dev_info(base->dev, "initialized\n");
3651 return 0;
3652
3653 destroy_cache:
3654 if (base->lcla_pool.dma_addr)
3655 dma_unmap_single(base->dev, base->lcla_pool.dma_addr,
3656 SZ_1K * base->num_phy_chans,
3657 DMA_TO_DEVICE);
3658
3659 if (!base->lcla_pool.base_unaligned && base->lcla_pool.base)
3660 free_pages((unsigned long)base->lcla_pool.base,
3661 base->lcla_pool.pages);
3662
3663 kfree(base->lcla_pool.base_unaligned);
3664
3665 if (base->lcpa_regulator) {
3666 regulator_disable(base->lcpa_regulator);
3667 regulator_put(base->lcpa_regulator);
3668 }
3669 pm_runtime_disable(base->dev);
3670
3671 report_failure:
3672 d40_err(dev, "probe failed\n");
3673 return ret;
3674}
3675
3676static const struct of_device_id d40_match[] = {
3677 { .compatible = "stericsson,dma40", },
3678 {}
3679};
3680
3681static struct platform_driver d40_driver = {
3682 .driver = {
3683 .name = D40_NAME,
3684 .pm = &dma40_pm_ops,
3685 .of_match_table = d40_match,
3686 },
3687};
3688
3689static int __init stedma40_init(void)
3690{
3691 return platform_driver_probe(&d40_driver, d40_probe);
3692}
3693subsys_initcall(stedma40_init);