Merge branch 'clockevents/fixes' of git://git.linaro.org/people/daniel.lezcano/linux...
[linux-drm-fsl-dcu.git] / drivers / dma / cppi41.c
1 #include <linux/dmaengine.h>
2 #include <linux/dma-mapping.h>
3 #include <linux/platform_device.h>
4 #include <linux/module.h>
5 #include <linux/of.h>
6 #include <linux/slab.h>
7 #include <linux/of_dma.h>
8 #include <linux/of_irq.h>
9 #include <linux/dmapool.h>
10 #include <linux/interrupt.h>
11 #include <linux/of_address.h>
12 #include <linux/pm_runtime.h>
13 #include "dmaengine.h"
14
15 #define DESC_TYPE       27
16 #define DESC_TYPE_HOST  0x10
17 #define DESC_TYPE_TEARD 0x13
18
19 #define TD_DESC_IS_RX   (1 << 16)
20 #define TD_DESC_DMA_NUM 10
21
22 #define DESC_LENGTH_BITS_NUM    21
23
24 #define DESC_TYPE_USB   (5 << 26)
25 #define DESC_PD_COMPLETE        (1 << 31)
26
27 /* DMA engine */
28 #define DMA_TDFDQ       4
29 #define DMA_TXGCR(x)    (0x800 + (x) * 0x20)
30 #define DMA_RXGCR(x)    (0x808 + (x) * 0x20)
31 #define RXHPCRA0                4
32
33 #define GCR_CHAN_ENABLE         (1 << 31)
34 #define GCR_TEARDOWN            (1 << 30)
35 #define GCR_STARV_RETRY         (1 << 24)
36 #define GCR_DESC_TYPE_HOST      (1 << 14)
37
38 /* DMA scheduler */
39 #define DMA_SCHED_CTRL          0
40 #define DMA_SCHED_CTRL_EN       (1 << 31)
41 #define DMA_SCHED_WORD(x)       ((x) * 4 + 0x800)
42
43 #define SCHED_ENTRY0_CHAN(x)    ((x) << 0)
44 #define SCHED_ENTRY0_IS_RX      (1 << 7)
45
46 #define SCHED_ENTRY1_CHAN(x)    ((x) << 8)
47 #define SCHED_ENTRY1_IS_RX      (1 << 15)
48
49 #define SCHED_ENTRY2_CHAN(x)    ((x) << 16)
50 #define SCHED_ENTRY2_IS_RX      (1 << 23)
51
52 #define SCHED_ENTRY3_CHAN(x)    ((x) << 24)
53 #define SCHED_ENTRY3_IS_RX      (1 << 31)
54
55 /* Queue manager */
56 /* 4 KiB of memory for descriptors, 2 for each endpoint */
57 #define ALLOC_DECS_NUM          128
58 #define DESCS_AREAS             1
59 #define TOTAL_DESCS_NUM         (ALLOC_DECS_NUM * DESCS_AREAS)
60 #define QMGR_SCRATCH_SIZE       (TOTAL_DESCS_NUM * 4)
61
62 #define QMGR_LRAM0_BASE         0x80
63 #define QMGR_LRAM_SIZE          0x84
64 #define QMGR_LRAM1_BASE         0x88
65 #define QMGR_MEMBASE(x)         (0x1000 + (x) * 0x10)
66 #define QMGR_MEMCTRL(x)         (0x1004 + (x) * 0x10)
67 #define QMGR_MEMCTRL_IDX_SH     16
68 #define QMGR_MEMCTRL_DESC_SH    8
69
70 #define QMGR_NUM_PEND   5
71 #define QMGR_PEND(x)    (0x90 + (x) * 4)
72
73 #define QMGR_PENDING_SLOT_Q(x)  (x / 32)
74 #define QMGR_PENDING_BIT_Q(x)   (x % 32)
75
76 #define QMGR_QUEUE_A(n) (0x2000 + (n) * 0x10)
77 #define QMGR_QUEUE_B(n) (0x2004 + (n) * 0x10)
78 #define QMGR_QUEUE_C(n) (0x2008 + (n) * 0x10)
79 #define QMGR_QUEUE_D(n) (0x200c + (n) * 0x10)
80
81 /* Glue layer specific */
82 /* USBSS  / USB AM335x */
83 #define USBSS_IRQ_STATUS        0x28
84 #define USBSS_IRQ_ENABLER       0x2c
85 #define USBSS_IRQ_CLEARR        0x30
86
87 #define USBSS_IRQ_PD_COMP       (1 <<  2)
88
89 struct cppi41_channel {
90         struct dma_chan chan;
91         struct dma_async_tx_descriptor txd;
92         struct cppi41_dd *cdd;
93         struct cppi41_desc *desc;
94         dma_addr_t desc_phys;
95         void __iomem *gcr_reg;
96         int is_tx;
97         u32 residue;
98
99         unsigned int q_num;
100         unsigned int q_comp_num;
101         unsigned int port_num;
102
103         unsigned td_retry;
104         unsigned td_queued:1;
105         unsigned td_seen:1;
106         unsigned td_desc_seen:1;
107 };
108
109 struct cppi41_desc {
110         u32 pd0;
111         u32 pd1;
112         u32 pd2;
113         u32 pd3;
114         u32 pd4;
115         u32 pd5;
116         u32 pd6;
117         u32 pd7;
118 } __aligned(32);
119
120 struct chan_queues {
121         u16 submit;
122         u16 complete;
123 };
124
125 struct cppi41_dd {
126         struct dma_device ddev;
127
128         void *qmgr_scratch;
129         dma_addr_t scratch_phys;
130
131         struct cppi41_desc *cd;
132         dma_addr_t descs_phys;
133         u32 first_td_desc;
134         struct cppi41_channel *chan_busy[ALLOC_DECS_NUM];
135
136         void __iomem *usbss_mem;
137         void __iomem *ctrl_mem;
138         void __iomem *sched_mem;
139         void __iomem *qmgr_mem;
140         unsigned int irq;
141         const struct chan_queues *queues_rx;
142         const struct chan_queues *queues_tx;
143         struct chan_queues td_queue;
144
145         /* context for suspend/resume */
146         unsigned int dma_tdfdq;
147 };
148
149 #define FIST_COMPLETION_QUEUE   93
150 static struct chan_queues usb_queues_tx[] = {
151         /* USB0 ENDP 1 */
152         [ 0] = { .submit = 32, .complete =  93},
153         [ 1] = { .submit = 34, .complete =  94},
154         [ 2] = { .submit = 36, .complete =  95},
155         [ 3] = { .submit = 38, .complete =  96},
156         [ 4] = { .submit = 40, .complete =  97},
157         [ 5] = { .submit = 42, .complete =  98},
158         [ 6] = { .submit = 44, .complete =  99},
159         [ 7] = { .submit = 46, .complete = 100},
160         [ 8] = { .submit = 48, .complete = 101},
161         [ 9] = { .submit = 50, .complete = 102},
162         [10] = { .submit = 52, .complete = 103},
163         [11] = { .submit = 54, .complete = 104},
164         [12] = { .submit = 56, .complete = 105},
165         [13] = { .submit = 58, .complete = 106},
166         [14] = { .submit = 60, .complete = 107},
167
168         /* USB1 ENDP1 */
169         [15] = { .submit = 62, .complete = 125},
170         [16] = { .submit = 64, .complete = 126},
171         [17] = { .submit = 66, .complete = 127},
172         [18] = { .submit = 68, .complete = 128},
173         [19] = { .submit = 70, .complete = 129},
174         [20] = { .submit = 72, .complete = 130},
175         [21] = { .submit = 74, .complete = 131},
176         [22] = { .submit = 76, .complete = 132},
177         [23] = { .submit = 78, .complete = 133},
178         [24] = { .submit = 80, .complete = 134},
179         [25] = { .submit = 82, .complete = 135},
180         [26] = { .submit = 84, .complete = 136},
181         [27] = { .submit = 86, .complete = 137},
182         [28] = { .submit = 88, .complete = 138},
183         [29] = { .submit = 90, .complete = 139},
184 };
185
186 static const struct chan_queues usb_queues_rx[] = {
187         /* USB0 ENDP 1 */
188         [ 0] = { .submit =  1, .complete = 109},
189         [ 1] = { .submit =  2, .complete = 110},
190         [ 2] = { .submit =  3, .complete = 111},
191         [ 3] = { .submit =  4, .complete = 112},
192         [ 4] = { .submit =  5, .complete = 113},
193         [ 5] = { .submit =  6, .complete = 114},
194         [ 6] = { .submit =  7, .complete = 115},
195         [ 7] = { .submit =  8, .complete = 116},
196         [ 8] = { .submit =  9, .complete = 117},
197         [ 9] = { .submit = 10, .complete = 118},
198         [10] = { .submit = 11, .complete = 119},
199         [11] = { .submit = 12, .complete = 120},
200         [12] = { .submit = 13, .complete = 121},
201         [13] = { .submit = 14, .complete = 122},
202         [14] = { .submit = 15, .complete = 123},
203
204         /* USB1 ENDP 1 */
205         [15] = { .submit = 16, .complete = 141},
206         [16] = { .submit = 17, .complete = 142},
207         [17] = { .submit = 18, .complete = 143},
208         [18] = { .submit = 19, .complete = 144},
209         [19] = { .submit = 20, .complete = 145},
210         [20] = { .submit = 21, .complete = 146},
211         [21] = { .submit = 22, .complete = 147},
212         [22] = { .submit = 23, .complete = 148},
213         [23] = { .submit = 24, .complete = 149},
214         [24] = { .submit = 25, .complete = 150},
215         [25] = { .submit = 26, .complete = 151},
216         [26] = { .submit = 27, .complete = 152},
217         [27] = { .submit = 28, .complete = 153},
218         [28] = { .submit = 29, .complete = 154},
219         [29] = { .submit = 30, .complete = 155},
220 };
221
222 struct cppi_glue_infos {
223         irqreturn_t (*isr)(int irq, void *data);
224         const struct chan_queues *queues_rx;
225         const struct chan_queues *queues_tx;
226         struct chan_queues td_queue;
227 };
228
229 static struct cppi41_channel *to_cpp41_chan(struct dma_chan *c)
230 {
231         return container_of(c, struct cppi41_channel, chan);
232 }
233
234 static struct cppi41_channel *desc_to_chan(struct cppi41_dd *cdd, u32 desc)
235 {
236         struct cppi41_channel *c;
237         u32 descs_size;
238         u32 desc_num;
239
240         descs_size = sizeof(struct cppi41_desc) * ALLOC_DECS_NUM;
241
242         if (!((desc >= cdd->descs_phys) &&
243                         (desc < (cdd->descs_phys + descs_size)))) {
244                 return NULL;
245         }
246
247         desc_num = (desc - cdd->descs_phys) / sizeof(struct cppi41_desc);
248         BUG_ON(desc_num >= ALLOC_DECS_NUM);
249         c = cdd->chan_busy[desc_num];
250         cdd->chan_busy[desc_num] = NULL;
251         return c;
252 }
253
254 static void cppi_writel(u32 val, void *__iomem *mem)
255 {
256         __raw_writel(val, mem);
257 }
258
259 static u32 cppi_readl(void *__iomem *mem)
260 {
261         return __raw_readl(mem);
262 }
263
264 static u32 pd_trans_len(u32 val)
265 {
266         return val & ((1 << (DESC_LENGTH_BITS_NUM + 1)) - 1);
267 }
268
269 static u32 cppi41_pop_desc(struct cppi41_dd *cdd, unsigned queue_num)
270 {
271         u32 desc;
272
273         desc = cppi_readl(cdd->qmgr_mem + QMGR_QUEUE_D(queue_num));
274         desc &= ~0x1f;
275         return desc;
276 }
277
278 static irqreturn_t cppi41_irq(int irq, void *data)
279 {
280         struct cppi41_dd *cdd = data;
281         struct cppi41_channel *c;
282         u32 status;
283         int i;
284
285         status = cppi_readl(cdd->usbss_mem + USBSS_IRQ_STATUS);
286         if (!(status & USBSS_IRQ_PD_COMP))
287                 return IRQ_NONE;
288         cppi_writel(status, cdd->usbss_mem + USBSS_IRQ_STATUS);
289
290         for (i = QMGR_PENDING_SLOT_Q(FIST_COMPLETION_QUEUE); i < QMGR_NUM_PEND;
291                         i++) {
292                 u32 val;
293                 u32 q_num;
294
295                 val = cppi_readl(cdd->qmgr_mem + QMGR_PEND(i));
296                 if (i == QMGR_PENDING_SLOT_Q(FIST_COMPLETION_QUEUE) && val) {
297                         u32 mask;
298                         /* set corresponding bit for completetion Q 93 */
299                         mask = 1 << QMGR_PENDING_BIT_Q(FIST_COMPLETION_QUEUE);
300                         /* not set all bits for queues less than Q 93 */
301                         mask--;
302                         /* now invert and keep only Q 93+ set */
303                         val &= ~mask;
304                 }
305
306                 if (val)
307                         __iormb();
308
309                 while (val) {
310                         u32 desc;
311
312                         q_num = __fls(val);
313                         val &= ~(1 << q_num);
314                         q_num += 32 * i;
315                         desc = cppi41_pop_desc(cdd, q_num);
316                         c = desc_to_chan(cdd, desc);
317                         if (WARN_ON(!c)) {
318                                 pr_err("%s() q %d desc %08x\n", __func__,
319                                                 q_num, desc);
320                                 continue;
321                         }
322                         c->residue = pd_trans_len(c->desc->pd6) -
323                                 pd_trans_len(c->desc->pd0);
324
325                         dma_cookie_complete(&c->txd);
326                         c->txd.callback(c->txd.callback_param);
327                 }
328         }
329         return IRQ_HANDLED;
330 }
331
332 static dma_cookie_t cppi41_tx_submit(struct dma_async_tx_descriptor *tx)
333 {
334         dma_cookie_t cookie;
335
336         cookie = dma_cookie_assign(tx);
337
338         return cookie;
339 }
340
341 static int cppi41_dma_alloc_chan_resources(struct dma_chan *chan)
342 {
343         struct cppi41_channel *c = to_cpp41_chan(chan);
344
345         dma_cookie_init(chan);
346         dma_async_tx_descriptor_init(&c->txd, chan);
347         c->txd.tx_submit = cppi41_tx_submit;
348
349         if (!c->is_tx)
350                 cppi_writel(c->q_num, c->gcr_reg + RXHPCRA0);
351
352         return 0;
353 }
354
355 static void cppi41_dma_free_chan_resources(struct dma_chan *chan)
356 {
357 }
358
359 static enum dma_status cppi41_dma_tx_status(struct dma_chan *chan,
360         dma_cookie_t cookie, struct dma_tx_state *txstate)
361 {
362         struct cppi41_channel *c = to_cpp41_chan(chan);
363         enum dma_status ret;
364
365         /* lock */
366         ret = dma_cookie_status(chan, cookie, txstate);
367         if (txstate && ret == DMA_COMPLETE)
368                 txstate->residue = c->residue;
369         /* unlock */
370
371         return ret;
372 }
373
374 static void push_desc_queue(struct cppi41_channel *c)
375 {
376         struct cppi41_dd *cdd = c->cdd;
377         u32 desc_num;
378         u32 desc_phys;
379         u32 reg;
380
381         desc_phys = lower_32_bits(c->desc_phys);
382         desc_num = (desc_phys - cdd->descs_phys) / sizeof(struct cppi41_desc);
383         WARN_ON(cdd->chan_busy[desc_num]);
384         cdd->chan_busy[desc_num] = c;
385
386         reg = (sizeof(struct cppi41_desc) - 24) / 4;
387         reg |= desc_phys;
388         cppi_writel(reg, cdd->qmgr_mem + QMGR_QUEUE_D(c->q_num));
389 }
390
391 static void cppi41_dma_issue_pending(struct dma_chan *chan)
392 {
393         struct cppi41_channel *c = to_cpp41_chan(chan);
394         u32 reg;
395
396         c->residue = 0;
397
398         reg = GCR_CHAN_ENABLE;
399         if (!c->is_tx) {
400                 reg |= GCR_STARV_RETRY;
401                 reg |= GCR_DESC_TYPE_HOST;
402                 reg |= c->q_comp_num;
403         }
404
405         cppi_writel(reg, c->gcr_reg);
406
407         /*
408          * We don't use writel() but __raw_writel() so we have to make sure
409          * that the DMA descriptor in coherent memory made to the main memory
410          * before starting the dma engine.
411          */
412         __iowmb();
413         push_desc_queue(c);
414 }
415
416 static u32 get_host_pd0(u32 length)
417 {
418         u32 reg;
419
420         reg = DESC_TYPE_HOST << DESC_TYPE;
421         reg |= length;
422
423         return reg;
424 }
425
426 static u32 get_host_pd1(struct cppi41_channel *c)
427 {
428         u32 reg;
429
430         reg = 0;
431
432         return reg;
433 }
434
435 static u32 get_host_pd2(struct cppi41_channel *c)
436 {
437         u32 reg;
438
439         reg = DESC_TYPE_USB;
440         reg |= c->q_comp_num;
441
442         return reg;
443 }
444
445 static u32 get_host_pd3(u32 length)
446 {
447         u32 reg;
448
449         /* PD3 = packet size */
450         reg = length;
451
452         return reg;
453 }
454
455 static u32 get_host_pd6(u32 length)
456 {
457         u32 reg;
458
459         /* PD6 buffer size */
460         reg = DESC_PD_COMPLETE;
461         reg |= length;
462
463         return reg;
464 }
465
466 static u32 get_host_pd4_or_7(u32 addr)
467 {
468         u32 reg;
469
470         reg = addr;
471
472         return reg;
473 }
474
475 static u32 get_host_pd5(void)
476 {
477         u32 reg;
478
479         reg = 0;
480
481         return reg;
482 }
483
484 static struct dma_async_tx_descriptor *cppi41_dma_prep_slave_sg(
485         struct dma_chan *chan, struct scatterlist *sgl, unsigned sg_len,
486         enum dma_transfer_direction dir, unsigned long tx_flags, void *context)
487 {
488         struct cppi41_channel *c = to_cpp41_chan(chan);
489         struct cppi41_desc *d;
490         struct scatterlist *sg;
491         unsigned int i;
492         unsigned int num;
493
494         num = 0;
495         d = c->desc;
496         for_each_sg(sgl, sg, sg_len, i) {
497                 u32 addr;
498                 u32 len;
499
500                 /* We need to use more than one desc once musb supports sg */
501                 BUG_ON(num > 0);
502                 addr = lower_32_bits(sg_dma_address(sg));
503                 len = sg_dma_len(sg);
504
505                 d->pd0 = get_host_pd0(len);
506                 d->pd1 = get_host_pd1(c);
507                 d->pd2 = get_host_pd2(c);
508                 d->pd3 = get_host_pd3(len);
509                 d->pd4 = get_host_pd4_or_7(addr);
510                 d->pd5 = get_host_pd5();
511                 d->pd6 = get_host_pd6(len);
512                 d->pd7 = get_host_pd4_or_7(addr);
513
514                 d++;
515         }
516
517         return &c->txd;
518 }
519
520 static int cpp41_cfg_chan(struct cppi41_channel *c,
521                 struct dma_slave_config *cfg)
522 {
523         return 0;
524 }
525
526 static void cppi41_compute_td_desc(struct cppi41_desc *d)
527 {
528         d->pd0 = DESC_TYPE_TEARD << DESC_TYPE;
529 }
530
531 static int cppi41_tear_down_chan(struct cppi41_channel *c)
532 {
533         struct cppi41_dd *cdd = c->cdd;
534         struct cppi41_desc *td;
535         u32 reg;
536         u32 desc_phys;
537         u32 td_desc_phys;
538
539         td = cdd->cd;
540         td += cdd->first_td_desc;
541
542         td_desc_phys = cdd->descs_phys;
543         td_desc_phys += cdd->first_td_desc * sizeof(struct cppi41_desc);
544
545         if (!c->td_queued) {
546                 cppi41_compute_td_desc(td);
547                 __iowmb();
548
549                 reg = (sizeof(struct cppi41_desc) - 24) / 4;
550                 reg |= td_desc_phys;
551                 cppi_writel(reg, cdd->qmgr_mem +
552                                 QMGR_QUEUE_D(cdd->td_queue.submit));
553
554                 reg = GCR_CHAN_ENABLE;
555                 if (!c->is_tx) {
556                         reg |= GCR_STARV_RETRY;
557                         reg |= GCR_DESC_TYPE_HOST;
558                         reg |= c->q_comp_num;
559                 }
560                 reg |= GCR_TEARDOWN;
561                 cppi_writel(reg, c->gcr_reg);
562                 c->td_queued = 1;
563                 c->td_retry = 100;
564         }
565
566         if (!c->td_seen || !c->td_desc_seen) {
567
568                 desc_phys = cppi41_pop_desc(cdd, cdd->td_queue.complete);
569                 if (!desc_phys)
570                         desc_phys = cppi41_pop_desc(cdd, c->q_comp_num);
571
572                 if (desc_phys == c->desc_phys) {
573                         c->td_desc_seen = 1;
574
575                 } else if (desc_phys == td_desc_phys) {
576                         u32 pd0;
577
578                         __iormb();
579                         pd0 = td->pd0;
580                         WARN_ON((pd0 >> DESC_TYPE) != DESC_TYPE_TEARD);
581                         WARN_ON(!c->is_tx && !(pd0 & TD_DESC_IS_RX));
582                         WARN_ON((pd0 & 0x1f) != c->port_num);
583                         c->td_seen = 1;
584                 } else if (desc_phys) {
585                         WARN_ON_ONCE(1);
586                 }
587         }
588         c->td_retry--;
589         /*
590          * If the TX descriptor / channel is in use, the caller needs to poke
591          * his TD bit multiple times. After that he hardware releases the
592          * transfer descriptor followed by TD descriptor. Waiting seems not to
593          * cause any difference.
594          * RX seems to be thrown out right away. However once the TearDown
595          * descriptor gets through we are done. If we have seens the transfer
596          * descriptor before the TD we fetch it from enqueue, it has to be
597          * there waiting for us.
598          */
599         if (!c->td_seen && c->td_retry)
600                 return -EAGAIN;
601
602         WARN_ON(!c->td_retry);
603         if (!c->td_desc_seen) {
604                 desc_phys = cppi41_pop_desc(cdd, c->q_num);
605                 WARN_ON(!desc_phys);
606         }
607
608         c->td_queued = 0;
609         c->td_seen = 0;
610         c->td_desc_seen = 0;
611         cppi_writel(0, c->gcr_reg);
612         return 0;
613 }
614
615 static int cppi41_stop_chan(struct dma_chan *chan)
616 {
617         struct cppi41_channel *c = to_cpp41_chan(chan);
618         struct cppi41_dd *cdd = c->cdd;
619         u32 desc_num;
620         u32 desc_phys;
621         int ret;
622
623         ret = cppi41_tear_down_chan(c);
624         if (ret)
625                 return ret;
626
627         desc_phys = lower_32_bits(c->desc_phys);
628         desc_num = (desc_phys - cdd->descs_phys) / sizeof(struct cppi41_desc);
629         WARN_ON(!cdd->chan_busy[desc_num]);
630         cdd->chan_busy[desc_num] = NULL;
631
632         return 0;
633 }
634
635 static int cppi41_dma_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
636         unsigned long arg)
637 {
638         struct cppi41_channel *c = to_cpp41_chan(chan);
639         int ret;
640
641         switch (cmd) {
642         case DMA_SLAVE_CONFIG:
643                 ret = cpp41_cfg_chan(c, (struct dma_slave_config *) arg);
644                 break;
645
646         case DMA_TERMINATE_ALL:
647                 ret = cppi41_stop_chan(chan);
648                 break;
649
650         default:
651                 ret = -ENXIO;
652                 break;
653         }
654         return ret;
655 }
656
657 static void cleanup_chans(struct cppi41_dd *cdd)
658 {
659         while (!list_empty(&cdd->ddev.channels)) {
660                 struct cppi41_channel *cchan;
661
662                 cchan = list_first_entry(&cdd->ddev.channels,
663                                 struct cppi41_channel, chan.device_node);
664                 list_del(&cchan->chan.device_node);
665                 kfree(cchan);
666         }
667 }
668
669 static int cppi41_add_chans(struct device *dev, struct cppi41_dd *cdd)
670 {
671         struct cppi41_channel *cchan;
672         int i;
673         int ret;
674         u32 n_chans;
675
676         ret = of_property_read_u32(dev->of_node, "#dma-channels",
677                         &n_chans);
678         if (ret)
679                 return ret;
680         /*
681          * The channels can only be used as TX or as RX. So we add twice
682          * that much dma channels because USB can only do RX or TX.
683          */
684         n_chans *= 2;
685
686         for (i = 0; i < n_chans; i++) {
687                 cchan = kzalloc(sizeof(*cchan), GFP_KERNEL);
688                 if (!cchan)
689                         goto err;
690
691                 cchan->cdd = cdd;
692                 if (i & 1) {
693                         cchan->gcr_reg = cdd->ctrl_mem + DMA_TXGCR(i >> 1);
694                         cchan->is_tx = 1;
695                 } else {
696                         cchan->gcr_reg = cdd->ctrl_mem + DMA_RXGCR(i >> 1);
697                         cchan->is_tx = 0;
698                 }
699                 cchan->port_num = i >> 1;
700                 cchan->desc = &cdd->cd[i];
701                 cchan->desc_phys = cdd->descs_phys;
702                 cchan->desc_phys += i * sizeof(struct cppi41_desc);
703                 cchan->chan.device = &cdd->ddev;
704                 list_add_tail(&cchan->chan.device_node, &cdd->ddev.channels);
705         }
706         cdd->first_td_desc = n_chans;
707
708         return 0;
709 err:
710         cleanup_chans(cdd);
711         return -ENOMEM;
712 }
713
714 static void purge_descs(struct device *dev, struct cppi41_dd *cdd)
715 {
716         unsigned int mem_decs;
717         int i;
718
719         mem_decs = ALLOC_DECS_NUM * sizeof(struct cppi41_desc);
720
721         for (i = 0; i < DESCS_AREAS; i++) {
722
723                 cppi_writel(0, cdd->qmgr_mem + QMGR_MEMBASE(i));
724                 cppi_writel(0, cdd->qmgr_mem + QMGR_MEMCTRL(i));
725
726                 dma_free_coherent(dev, mem_decs, cdd->cd,
727                                 cdd->descs_phys);
728         }
729 }
730
731 static void disable_sched(struct cppi41_dd *cdd)
732 {
733         cppi_writel(0, cdd->sched_mem + DMA_SCHED_CTRL);
734 }
735
736 static void deinit_cppi41(struct device *dev, struct cppi41_dd *cdd)
737 {
738         disable_sched(cdd);
739
740         purge_descs(dev, cdd);
741
742         cppi_writel(0, cdd->qmgr_mem + QMGR_LRAM0_BASE);
743         cppi_writel(0, cdd->qmgr_mem + QMGR_LRAM0_BASE);
744         dma_free_coherent(dev, QMGR_SCRATCH_SIZE, cdd->qmgr_scratch,
745                         cdd->scratch_phys);
746 }
747
748 static int init_descs(struct device *dev, struct cppi41_dd *cdd)
749 {
750         unsigned int desc_size;
751         unsigned int mem_decs;
752         int i;
753         u32 reg;
754         u32 idx;
755
756         BUILD_BUG_ON(sizeof(struct cppi41_desc) &
757                         (sizeof(struct cppi41_desc) - 1));
758         BUILD_BUG_ON(sizeof(struct cppi41_desc) < 32);
759         BUILD_BUG_ON(ALLOC_DECS_NUM < 32);
760
761         desc_size = sizeof(struct cppi41_desc);
762         mem_decs = ALLOC_DECS_NUM * desc_size;
763
764         idx = 0;
765         for (i = 0; i < DESCS_AREAS; i++) {
766
767                 reg = idx << QMGR_MEMCTRL_IDX_SH;
768                 reg |= (ilog2(desc_size) - 5) << QMGR_MEMCTRL_DESC_SH;
769                 reg |= ilog2(ALLOC_DECS_NUM) - 5;
770
771                 BUILD_BUG_ON(DESCS_AREAS != 1);
772                 cdd->cd = dma_alloc_coherent(dev, mem_decs,
773                                 &cdd->descs_phys, GFP_KERNEL);
774                 if (!cdd->cd)
775                         return -ENOMEM;
776
777                 cppi_writel(cdd->descs_phys, cdd->qmgr_mem + QMGR_MEMBASE(i));
778                 cppi_writel(reg, cdd->qmgr_mem + QMGR_MEMCTRL(i));
779
780                 idx += ALLOC_DECS_NUM;
781         }
782         return 0;
783 }
784
785 static void init_sched(struct cppi41_dd *cdd)
786 {
787         unsigned ch;
788         unsigned word;
789         u32 reg;
790
791         word = 0;
792         cppi_writel(0, cdd->sched_mem + DMA_SCHED_CTRL);
793         for (ch = 0; ch < 15 * 2; ch += 2) {
794
795                 reg = SCHED_ENTRY0_CHAN(ch);
796                 reg |= SCHED_ENTRY1_CHAN(ch) | SCHED_ENTRY1_IS_RX;
797
798                 reg |= SCHED_ENTRY2_CHAN(ch + 1);
799                 reg |= SCHED_ENTRY3_CHAN(ch + 1) | SCHED_ENTRY3_IS_RX;
800                 cppi_writel(reg, cdd->sched_mem + DMA_SCHED_WORD(word));
801                 word++;
802         }
803         reg = 15 * 2 * 2 - 1;
804         reg |= DMA_SCHED_CTRL_EN;
805         cppi_writel(reg, cdd->sched_mem + DMA_SCHED_CTRL);
806 }
807
808 static int init_cppi41(struct device *dev, struct cppi41_dd *cdd)
809 {
810         int ret;
811
812         BUILD_BUG_ON(QMGR_SCRATCH_SIZE > ((1 << 14) - 1));
813         cdd->qmgr_scratch = dma_alloc_coherent(dev, QMGR_SCRATCH_SIZE,
814                         &cdd->scratch_phys, GFP_KERNEL);
815         if (!cdd->qmgr_scratch)
816                 return -ENOMEM;
817
818         cppi_writel(cdd->scratch_phys, cdd->qmgr_mem + QMGR_LRAM0_BASE);
819         cppi_writel(QMGR_SCRATCH_SIZE, cdd->qmgr_mem + QMGR_LRAM_SIZE);
820         cppi_writel(0, cdd->qmgr_mem + QMGR_LRAM1_BASE);
821
822         ret = init_descs(dev, cdd);
823         if (ret)
824                 goto err_td;
825
826         cppi_writel(cdd->td_queue.submit, cdd->ctrl_mem + DMA_TDFDQ);
827         init_sched(cdd);
828         return 0;
829 err_td:
830         deinit_cppi41(dev, cdd);
831         return ret;
832 }
833
834 static struct platform_driver cpp41_dma_driver;
835 /*
836  * The param format is:
837  * X Y
838  * X: Port
839  * Y: 0 = RX else TX
840  */
841 #define INFO_PORT       0
842 #define INFO_IS_TX      1
843
844 static bool cpp41_dma_filter_fn(struct dma_chan *chan, void *param)
845 {
846         struct cppi41_channel *cchan;
847         struct cppi41_dd *cdd;
848         const struct chan_queues *queues;
849         u32 *num = param;
850
851         if (chan->device->dev->driver != &cpp41_dma_driver.driver)
852                 return false;
853
854         cchan = to_cpp41_chan(chan);
855
856         if (cchan->port_num != num[INFO_PORT])
857                 return false;
858
859         if (cchan->is_tx && !num[INFO_IS_TX])
860                 return false;
861         cdd = cchan->cdd;
862         if (cchan->is_tx)
863                 queues = cdd->queues_tx;
864         else
865                 queues = cdd->queues_rx;
866
867         BUILD_BUG_ON(ARRAY_SIZE(usb_queues_rx) != ARRAY_SIZE(usb_queues_tx));
868         if (WARN_ON(cchan->port_num > ARRAY_SIZE(usb_queues_rx)))
869                 return false;
870
871         cchan->q_num = queues[cchan->port_num].submit;
872         cchan->q_comp_num = queues[cchan->port_num].complete;
873         return true;
874 }
875
876 static struct of_dma_filter_info cpp41_dma_info = {
877         .filter_fn = cpp41_dma_filter_fn,
878 };
879
880 static struct dma_chan *cppi41_dma_xlate(struct of_phandle_args *dma_spec,
881                 struct of_dma *ofdma)
882 {
883         int count = dma_spec->args_count;
884         struct of_dma_filter_info *info = ofdma->of_dma_data;
885
886         if (!info || !info->filter_fn)
887                 return NULL;
888
889         if (count != 2)
890                 return NULL;
891
892         return dma_request_channel(info->dma_cap, info->filter_fn,
893                         &dma_spec->args[0]);
894 }
895
896 static const struct cppi_glue_infos usb_infos = {
897         .isr = cppi41_irq,
898         .queues_rx = usb_queues_rx,
899         .queues_tx = usb_queues_tx,
900         .td_queue = { .submit = 31, .complete = 0 },
901 };
902
903 static const struct of_device_id cppi41_dma_ids[] = {
904         { .compatible = "ti,am3359-cppi41", .data = &usb_infos},
905         {},
906 };
907 MODULE_DEVICE_TABLE(of, cppi41_dma_ids);
908
909 static const struct cppi_glue_infos *get_glue_info(struct device *dev)
910 {
911         const struct of_device_id *of_id;
912
913         of_id = of_match_node(cppi41_dma_ids, dev->of_node);
914         if (!of_id)
915                 return NULL;
916         return of_id->data;
917 }
918
919 static int cppi41_dma_probe(struct platform_device *pdev)
920 {
921         struct cppi41_dd *cdd;
922         struct device *dev = &pdev->dev;
923         const struct cppi_glue_infos *glue_info;
924         int irq;
925         int ret;
926
927         glue_info = get_glue_info(dev);
928         if (!glue_info)
929                 return -EINVAL;
930
931         cdd = kzalloc(sizeof(*cdd), GFP_KERNEL);
932         if (!cdd)
933                 return -ENOMEM;
934
935         dma_cap_set(DMA_SLAVE, cdd->ddev.cap_mask);
936         cdd->ddev.device_alloc_chan_resources = cppi41_dma_alloc_chan_resources;
937         cdd->ddev.device_free_chan_resources = cppi41_dma_free_chan_resources;
938         cdd->ddev.device_tx_status = cppi41_dma_tx_status;
939         cdd->ddev.device_issue_pending = cppi41_dma_issue_pending;
940         cdd->ddev.device_prep_slave_sg = cppi41_dma_prep_slave_sg;
941         cdd->ddev.device_control = cppi41_dma_control;
942         cdd->ddev.dev = dev;
943         INIT_LIST_HEAD(&cdd->ddev.channels);
944         cpp41_dma_info.dma_cap = cdd->ddev.cap_mask;
945
946         cdd->usbss_mem = of_iomap(dev->of_node, 0);
947         cdd->ctrl_mem = of_iomap(dev->of_node, 1);
948         cdd->sched_mem = of_iomap(dev->of_node, 2);
949         cdd->qmgr_mem = of_iomap(dev->of_node, 3);
950
951         if (!cdd->usbss_mem || !cdd->ctrl_mem || !cdd->sched_mem ||
952                         !cdd->qmgr_mem) {
953                 ret = -ENXIO;
954                 goto err_remap;
955         }
956
957         pm_runtime_enable(dev);
958         ret = pm_runtime_get_sync(dev);
959         if (ret < 0)
960                 goto err_get_sync;
961
962         cdd->queues_rx = glue_info->queues_rx;
963         cdd->queues_tx = glue_info->queues_tx;
964         cdd->td_queue = glue_info->td_queue;
965
966         ret = init_cppi41(dev, cdd);
967         if (ret)
968                 goto err_init_cppi;
969
970         ret = cppi41_add_chans(dev, cdd);
971         if (ret)
972                 goto err_chans;
973
974         irq = irq_of_parse_and_map(dev->of_node, 0);
975         if (!irq)
976                 goto err_irq;
977
978         cppi_writel(USBSS_IRQ_PD_COMP, cdd->usbss_mem + USBSS_IRQ_ENABLER);
979
980         ret = request_irq(irq, glue_info->isr, IRQF_SHARED,
981                         dev_name(dev), cdd);
982         if (ret)
983                 goto err_irq;
984         cdd->irq = irq;
985
986         ret = dma_async_device_register(&cdd->ddev);
987         if (ret)
988                 goto err_dma_reg;
989
990         ret = of_dma_controller_register(dev->of_node,
991                         cppi41_dma_xlate, &cpp41_dma_info);
992         if (ret)
993                 goto err_of;
994
995         platform_set_drvdata(pdev, cdd);
996         return 0;
997 err_of:
998         dma_async_device_unregister(&cdd->ddev);
999 err_dma_reg:
1000         free_irq(irq, cdd);
1001 err_irq:
1002         cppi_writel(0, cdd->usbss_mem + USBSS_IRQ_CLEARR);
1003         cleanup_chans(cdd);
1004 err_chans:
1005         deinit_cppi41(dev, cdd);
1006 err_init_cppi:
1007         pm_runtime_put(dev);
1008 err_get_sync:
1009         pm_runtime_disable(dev);
1010         iounmap(cdd->usbss_mem);
1011         iounmap(cdd->ctrl_mem);
1012         iounmap(cdd->sched_mem);
1013         iounmap(cdd->qmgr_mem);
1014 err_remap:
1015         kfree(cdd);
1016         return ret;
1017 }
1018
1019 static int cppi41_dma_remove(struct platform_device *pdev)
1020 {
1021         struct cppi41_dd *cdd = platform_get_drvdata(pdev);
1022
1023         of_dma_controller_free(pdev->dev.of_node);
1024         dma_async_device_unregister(&cdd->ddev);
1025
1026         cppi_writel(0, cdd->usbss_mem + USBSS_IRQ_CLEARR);
1027         free_irq(cdd->irq, cdd);
1028         cleanup_chans(cdd);
1029         deinit_cppi41(&pdev->dev, cdd);
1030         iounmap(cdd->usbss_mem);
1031         iounmap(cdd->ctrl_mem);
1032         iounmap(cdd->sched_mem);
1033         iounmap(cdd->qmgr_mem);
1034         pm_runtime_put(&pdev->dev);
1035         pm_runtime_disable(&pdev->dev);
1036         kfree(cdd);
1037         return 0;
1038 }
1039
1040 #ifdef CONFIG_PM_SLEEP
1041 static int cppi41_suspend(struct device *dev)
1042 {
1043         struct cppi41_dd *cdd = dev_get_drvdata(dev);
1044
1045         cdd->dma_tdfdq = cppi_readl(cdd->ctrl_mem + DMA_TDFDQ);
1046         cppi_writel(0, cdd->usbss_mem + USBSS_IRQ_CLEARR);
1047         disable_sched(cdd);
1048
1049         return 0;
1050 }
1051
1052 static int cppi41_resume(struct device *dev)
1053 {
1054         struct cppi41_dd *cdd = dev_get_drvdata(dev);
1055         struct cppi41_channel *c;
1056         int i;
1057
1058         for (i = 0; i < DESCS_AREAS; i++)
1059                 cppi_writel(cdd->descs_phys, cdd->qmgr_mem + QMGR_MEMBASE(i));
1060
1061         list_for_each_entry(c, &cdd->ddev.channels, chan.device_node)
1062                 if (!c->is_tx)
1063                         cppi_writel(c->q_num, c->gcr_reg + RXHPCRA0);
1064
1065         init_sched(cdd);
1066
1067         cppi_writel(cdd->dma_tdfdq, cdd->ctrl_mem + DMA_TDFDQ);
1068         cppi_writel(cdd->scratch_phys, cdd->qmgr_mem + QMGR_LRAM0_BASE);
1069         cppi_writel(QMGR_SCRATCH_SIZE, cdd->qmgr_mem + QMGR_LRAM_SIZE);
1070         cppi_writel(0, cdd->qmgr_mem + QMGR_LRAM1_BASE);
1071
1072         cppi_writel(USBSS_IRQ_PD_COMP, cdd->usbss_mem + USBSS_IRQ_ENABLER);
1073
1074         return 0;
1075 }
1076 #endif
1077
1078 static SIMPLE_DEV_PM_OPS(cppi41_pm_ops, cppi41_suspend, cppi41_resume);
1079
1080 static struct platform_driver cpp41_dma_driver = {
1081         .probe  = cppi41_dma_probe,
1082         .remove = cppi41_dma_remove,
1083         .driver = {
1084                 .name = "cppi41-dma-engine",
1085                 .owner = THIS_MODULE,
1086                 .pm = &cppi41_pm_ops,
1087                 .of_match_table = of_match_ptr(cppi41_dma_ids),
1088         },
1089 };
1090
1091 module_platform_driver(cpp41_dma_driver);
1092 MODULE_LICENSE("GPL");
1093 MODULE_AUTHOR("Sebastian Andrzej Siewior <bigeasy@linutronix.de>");