Merge branch 'stable/xen-pcifront-0.8.2' of git://git.kernel.org/pub/scm/linux/kernel...
[platform/adaptation/renesas_rcar/renesas_kernel.git] / arch / arm / mach-tegra / dma.c
1 /*
2  * arch/arm/mach-tegra/dma.c
3  *
4  * System DMA driver for NVIDIA Tegra SoCs
5  *
6  * Copyright (c) 2008-2009, NVIDIA Corporation.
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12  *
13  * This program is distributed in the hope that it will be useful, but WITHOUT
14  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
15  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
16  * more details.
17  *
18  * You should have received a copy of the GNU General Public License along
19  * with this program; if not, write to the Free Software Foundation, Inc.,
20  * 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
21  */
22
23 #include <linux/io.h>
24 #include <linux/interrupt.h>
25 #include <linux/module.h>
26 #include <linux/spinlock.h>
27 #include <linux/err.h>
28 #include <linux/irq.h>
29 #include <linux/delay.h>
30 #include <mach/dma.h>
31 #include <mach/irqs.h>
32 #include <mach/iomap.h>
33
34 #define APB_DMA_GEN                             0x000
35 #define GEN_ENABLE                              (1<<31)
36
37 #define APB_DMA_CNTRL                           0x010
38
39 #define APB_DMA_IRQ_MASK                        0x01c
40
41 #define APB_DMA_IRQ_MASK_SET                    0x020
42
43 #define APB_DMA_CHAN_CSR                        0x000
44 #define CSR_ENB                                 (1<<31)
45 #define CSR_IE_EOC                              (1<<30)
46 #define CSR_HOLD                                (1<<29)
47 #define CSR_DIR                                 (1<<28)
48 #define CSR_ONCE                                (1<<27)
49 #define CSR_FLOW                                (1<<21)
50 #define CSR_REQ_SEL_SHIFT                       16
51 #define CSR_REQ_SEL_MASK                        (0x1F<<CSR_REQ_SEL_SHIFT)
52 #define CSR_REQ_SEL_INVALID                     (31<<CSR_REQ_SEL_SHIFT)
53 #define CSR_WCOUNT_SHIFT                        2
54 #define CSR_WCOUNT_MASK                         0xFFFC
55
56 #define APB_DMA_CHAN_STA                                0x004
57 #define STA_BUSY                                (1<<31)
58 #define STA_ISE_EOC                             (1<<30)
59 #define STA_HALT                                (1<<29)
60 #define STA_PING_PONG                           (1<<28)
61 #define STA_COUNT_SHIFT                         2
62 #define STA_COUNT_MASK                          0xFFFC
63
64 #define APB_DMA_CHAN_AHB_PTR                            0x010
65
66 #define APB_DMA_CHAN_AHB_SEQ                            0x014
67 #define AHB_SEQ_INTR_ENB                        (1<<31)
68 #define AHB_SEQ_BUS_WIDTH_SHIFT                 28
69 #define AHB_SEQ_BUS_WIDTH_MASK                  (0x7<<AHB_SEQ_BUS_WIDTH_SHIFT)
70 #define AHB_SEQ_BUS_WIDTH_8                     (0<<AHB_SEQ_BUS_WIDTH_SHIFT)
71 #define AHB_SEQ_BUS_WIDTH_16                    (1<<AHB_SEQ_BUS_WIDTH_SHIFT)
72 #define AHB_SEQ_BUS_WIDTH_32                    (2<<AHB_SEQ_BUS_WIDTH_SHIFT)
73 #define AHB_SEQ_BUS_WIDTH_64                    (3<<AHB_SEQ_BUS_WIDTH_SHIFT)
74 #define AHB_SEQ_BUS_WIDTH_128                   (4<<AHB_SEQ_BUS_WIDTH_SHIFT)
75 #define AHB_SEQ_DATA_SWAP                       (1<<27)
76 #define AHB_SEQ_BURST_MASK                      (0x7<<24)
77 #define AHB_SEQ_BURST_1                         (4<<24)
78 #define AHB_SEQ_BURST_4                         (5<<24)
79 #define AHB_SEQ_BURST_8                         (6<<24)
80 #define AHB_SEQ_DBL_BUF                         (1<<19)
81 #define AHB_SEQ_WRAP_SHIFT                      16
82 #define AHB_SEQ_WRAP_MASK                       (0x7<<AHB_SEQ_WRAP_SHIFT)
83
84 #define APB_DMA_CHAN_APB_PTR                            0x018
85
86 #define APB_DMA_CHAN_APB_SEQ                            0x01c
87 #define APB_SEQ_BUS_WIDTH_SHIFT                 28
88 #define APB_SEQ_BUS_WIDTH_MASK                  (0x7<<APB_SEQ_BUS_WIDTH_SHIFT)
89 #define APB_SEQ_BUS_WIDTH_8                     (0<<APB_SEQ_BUS_WIDTH_SHIFT)
90 #define APB_SEQ_BUS_WIDTH_16                    (1<<APB_SEQ_BUS_WIDTH_SHIFT)
91 #define APB_SEQ_BUS_WIDTH_32                    (2<<APB_SEQ_BUS_WIDTH_SHIFT)
92 #define APB_SEQ_BUS_WIDTH_64                    (3<<APB_SEQ_BUS_WIDTH_SHIFT)
93 #define APB_SEQ_BUS_WIDTH_128                   (4<<APB_SEQ_BUS_WIDTH_SHIFT)
94 #define APB_SEQ_DATA_SWAP                       (1<<27)
95 #define APB_SEQ_WRAP_SHIFT                      16
96 #define APB_SEQ_WRAP_MASK                       (0x7<<APB_SEQ_WRAP_SHIFT)
97
98 #define TEGRA_SYSTEM_DMA_CH_NR                  16
99 #define TEGRA_SYSTEM_DMA_AVP_CH_NUM             4
100 #define TEGRA_SYSTEM_DMA_CH_MIN                 0
101 #define TEGRA_SYSTEM_DMA_CH_MAX \
102         (TEGRA_SYSTEM_DMA_CH_NR - TEGRA_SYSTEM_DMA_AVP_CH_NUM - 1)
103
104 #define NV_DMA_MAX_TRASFER_SIZE 0x10000
105
106 const unsigned int ahb_addr_wrap_table[8] = {
107         0, 32, 64, 128, 256, 512, 1024, 2048
108 };
109
110 const unsigned int apb_addr_wrap_table[8] = {0, 1, 2, 4, 8, 16, 32, 64};
111
112 const unsigned int bus_width_table[5] = {8, 16, 32, 64, 128};
113
114 #define TEGRA_DMA_NAME_SIZE 16
115 struct tegra_dma_channel {
116         struct list_head        list;
117         int                     id;
118         spinlock_t              lock;
119         char                    name[TEGRA_DMA_NAME_SIZE];
120         void  __iomem           *addr;
121         int                     mode;
122         int                     irq;
123
124         /* Register shadow */
125         u32                     csr;
126         u32                     ahb_seq;
127         u32                     ahb_ptr;
128         u32                     apb_seq;
129         u32                     apb_ptr;
130 };
131
132 #define  NV_DMA_MAX_CHANNELS  32
133
134 static DECLARE_BITMAP(channel_usage, NV_DMA_MAX_CHANNELS);
135 static struct tegra_dma_channel dma_channels[NV_DMA_MAX_CHANNELS];
136
137 static void tegra_dma_update_hw(struct tegra_dma_channel *ch,
138         struct tegra_dma_req *req);
139 static void tegra_dma_update_hw_partial(struct tegra_dma_channel *ch,
140         struct tegra_dma_req *req);
141 static void tegra_dma_init_hw(struct tegra_dma_channel *ch);
142 static void tegra_dma_stop(struct tegra_dma_channel *ch);
143
144 void tegra_dma_flush(struct tegra_dma_channel *ch)
145 {
146 }
147 EXPORT_SYMBOL(tegra_dma_flush);
148
149 void tegra_dma_dequeue(struct tegra_dma_channel *ch)
150 {
151         struct tegra_dma_req *req;
152
153         req = list_entry(ch->list.next, typeof(*req), node);
154
155         tegra_dma_dequeue_req(ch, req);
156         return;
157 }
158
159 void tegra_dma_stop(struct tegra_dma_channel *ch)
160 {
161         unsigned int csr;
162         unsigned int status;
163
164         csr = ch->csr;
165         csr &= ~CSR_IE_EOC;
166         writel(csr, ch->addr + APB_DMA_CHAN_CSR);
167
168         csr &= ~CSR_ENB;
169         writel(csr, ch->addr + APB_DMA_CHAN_CSR);
170
171         status = readl(ch->addr + APB_DMA_CHAN_STA);
172         if (status & STA_ISE_EOC)
173                 writel(status, ch->addr + APB_DMA_CHAN_STA);
174 }
175
176 int tegra_dma_cancel(struct tegra_dma_channel *ch)
177 {
178         unsigned int csr;
179         unsigned long irq_flags;
180
181         spin_lock_irqsave(&ch->lock, irq_flags);
182         while (!list_empty(&ch->list))
183                 list_del(ch->list.next);
184
185         csr = ch->csr;
186         csr &= ~CSR_REQ_SEL_MASK;
187         csr |= CSR_REQ_SEL_INVALID;
188
189         /* Set the enable as that is not shadowed */
190         csr |= CSR_ENB;
191         writel(csr, ch->addr + APB_DMA_CHAN_CSR);
192
193         tegra_dma_stop(ch);
194
195         spin_unlock_irqrestore(&ch->lock, irq_flags);
196         return 0;
197 }
198
199 int tegra_dma_dequeue_req(struct tegra_dma_channel *ch,
200         struct tegra_dma_req *_req)
201 {
202         unsigned int csr;
203         unsigned int status;
204         struct tegra_dma_req *req = NULL;
205         int found = 0;
206         unsigned long irq_flags;
207         int to_transfer;
208         int req_transfer_count;
209
210         spin_lock_irqsave(&ch->lock, irq_flags);
211         list_for_each_entry(req, &ch->list, node) {
212                 if (req == _req) {
213                         list_del(&req->node);
214                         found = 1;
215                         break;
216                 }
217         }
218         if (!found) {
219                 spin_unlock_irqrestore(&ch->lock, irq_flags);
220                 return 0;
221         }
222
223         /* STOP the DMA and get the transfer count.
224          * Getting the transfer count is tricky.
225          *  - Change the source selector to invalid to stop the DMA from
226          *    FIFO to memory.
227          *  - Read the status register to know the number of pending
228          *    bytes to be transfered.
229          *  - Finally stop or program the DMA to the next buffer in the
230          *    list.
231          */
232         csr = ch->csr;
233         csr &= ~CSR_REQ_SEL_MASK;
234         csr |= CSR_REQ_SEL_INVALID;
235
236         /* Set the enable as that is not shadowed */
237         csr |= CSR_ENB;
238         writel(csr, ch->addr + APB_DMA_CHAN_CSR);
239
240         /* Get the transfer count */
241         status = readl(ch->addr + APB_DMA_CHAN_STA);
242         to_transfer = (status & STA_COUNT_MASK) >> STA_COUNT_SHIFT;
243         req_transfer_count = (ch->csr & CSR_WCOUNT_MASK) >> CSR_WCOUNT_SHIFT;
244         req_transfer_count += 1;
245         to_transfer += 1;
246
247         req->bytes_transferred = req_transfer_count;
248
249         if (status & STA_BUSY)
250                 req->bytes_transferred -= to_transfer;
251
252         /* In continous transfer mode, DMA only tracks the count of the
253          * half DMA buffer. So, if the DMA already finished half the DMA
254          * then add the half buffer to the completed count.
255          *
256          *      FIXME: There can be a race here. What if the req to
257          *      dequue happens at the same time as the DMA just moved to
258          *      the new buffer and SW didn't yet received the interrupt?
259          */
260         if (ch->mode & TEGRA_DMA_MODE_CONTINOUS)
261                 if (req->buffer_status == TEGRA_DMA_REQ_BUF_STATUS_HALF_FULL)
262                         req->bytes_transferred += req_transfer_count;
263
264         req->bytes_transferred *= 4;
265
266         tegra_dma_stop(ch);
267         if (!list_empty(&ch->list)) {
268                 /* if the list is not empty, queue the next request */
269                 struct tegra_dma_req *next_req;
270                 next_req = list_entry(ch->list.next,
271                         typeof(*next_req), node);
272                 tegra_dma_update_hw(ch, next_req);
273         }
274         req->status = -TEGRA_DMA_REQ_ERROR_ABORTED;
275
276         spin_unlock_irqrestore(&ch->lock, irq_flags);
277
278         /* Callback should be called without any lock */
279         req->complete(req);
280         return 0;
281 }
282 EXPORT_SYMBOL(tegra_dma_dequeue_req);
283
284 bool tegra_dma_is_empty(struct tegra_dma_channel *ch)
285 {
286         unsigned long irq_flags;
287         bool is_empty;
288
289         spin_lock_irqsave(&ch->lock, irq_flags);
290         if (list_empty(&ch->list))
291                 is_empty = true;
292         else
293                 is_empty = false;
294         spin_unlock_irqrestore(&ch->lock, irq_flags);
295         return is_empty;
296 }
297 EXPORT_SYMBOL(tegra_dma_is_empty);
298
299 bool tegra_dma_is_req_inflight(struct tegra_dma_channel *ch,
300         struct tegra_dma_req *_req)
301 {
302         unsigned long irq_flags;
303         struct tegra_dma_req *req;
304
305         spin_lock_irqsave(&ch->lock, irq_flags);
306         list_for_each_entry(req, &ch->list, node) {
307                 if (req == _req) {
308                         spin_unlock_irqrestore(&ch->lock, irq_flags);
309                         return true;
310                 }
311         }
312         spin_unlock_irqrestore(&ch->lock, irq_flags);
313         return false;
314 }
315 EXPORT_SYMBOL(tegra_dma_is_req_inflight);
316
317 int tegra_dma_enqueue_req(struct tegra_dma_channel *ch,
318         struct tegra_dma_req *req)
319 {
320         unsigned long irq_flags;
321         int start_dma = 0;
322
323         if (req->size > NV_DMA_MAX_TRASFER_SIZE ||
324                 req->source_addr & 0x3 || req->dest_addr & 0x3) {
325                 pr_err("Invalid DMA request for channel %d\n", ch->id);
326                 return -EINVAL;
327         }
328
329         spin_lock_irqsave(&ch->lock, irq_flags);
330
331         req->bytes_transferred = 0;
332         req->status = 0;
333         req->buffer_status = 0;
334         if (list_empty(&ch->list))
335                 start_dma = 1;
336
337         list_add_tail(&req->node, &ch->list);
338
339         if (start_dma)
340                 tegra_dma_update_hw(ch, req);
341
342         spin_unlock_irqrestore(&ch->lock, irq_flags);
343
344         return 0;
345 }
346 EXPORT_SYMBOL(tegra_dma_enqueue_req);
347
348 struct tegra_dma_channel *tegra_dma_allocate_channel(int mode)
349 {
350         int channel;
351         struct tegra_dma_channel *ch;
352
353         /* first channel is the shared channel */
354         if (mode & TEGRA_DMA_SHARED) {
355                 channel = TEGRA_SYSTEM_DMA_CH_MIN;
356         } else {
357                 channel = find_first_zero_bit(channel_usage,
358                         ARRAY_SIZE(dma_channels));
359                 if (channel >= ARRAY_SIZE(dma_channels))
360                         return NULL;
361         }
362         __set_bit(channel, channel_usage);
363         ch = &dma_channels[channel];
364         ch->mode = mode;
365         return ch;
366 }
367 EXPORT_SYMBOL(tegra_dma_allocate_channel);
368
369 void tegra_dma_free_channel(struct tegra_dma_channel *ch)
370 {
371         if (ch->mode & TEGRA_DMA_SHARED)
372                 return;
373         tegra_dma_cancel(ch);
374         __clear_bit(ch->id, channel_usage);
375 }
376 EXPORT_SYMBOL(tegra_dma_free_channel);
377
378 static void tegra_dma_update_hw_partial(struct tegra_dma_channel *ch,
379         struct tegra_dma_req *req)
380 {
381         if (req->to_memory) {
382                 ch->apb_ptr = req->source_addr;
383                 ch->ahb_ptr = req->dest_addr;
384         } else {
385                 ch->apb_ptr = req->dest_addr;
386                 ch->ahb_ptr = req->source_addr;
387         }
388         writel(ch->apb_ptr, ch->addr + APB_DMA_CHAN_APB_PTR);
389         writel(ch->ahb_ptr, ch->addr + APB_DMA_CHAN_AHB_PTR);
390
391         req->status = TEGRA_DMA_REQ_INFLIGHT;
392         return;
393 }
394
395 static void tegra_dma_update_hw(struct tegra_dma_channel *ch,
396         struct tegra_dma_req *req)
397 {
398         int ahb_addr_wrap;
399         int apb_addr_wrap;
400         int ahb_bus_width;
401         int apb_bus_width;
402         int index;
403         unsigned long csr;
404
405
406         ch->csr |= CSR_FLOW;
407         ch->csr &= ~CSR_REQ_SEL_MASK;
408         ch->csr |= req->req_sel << CSR_REQ_SEL_SHIFT;
409         ch->ahb_seq &= ~AHB_SEQ_BURST_MASK;
410         ch->ahb_seq |= AHB_SEQ_BURST_1;
411
412         /* One shot mode is always single buffered,
413          * continuous mode is always double buffered
414          * */
415         if (ch->mode & TEGRA_DMA_MODE_ONESHOT) {
416                 ch->csr |= CSR_ONCE;
417                 ch->ahb_seq &= ~AHB_SEQ_DBL_BUF;
418                 ch->csr &= ~CSR_WCOUNT_MASK;
419                 ch->csr |= ((req->size>>2) - 1) << CSR_WCOUNT_SHIFT;
420         } else {
421                 ch->csr &= ~CSR_ONCE;
422                 ch->ahb_seq |= AHB_SEQ_DBL_BUF;
423
424                 /* In double buffered mode, we set the size to half the
425                  * requested size and interrupt when half the buffer
426                  * is full */
427                 ch->csr &= ~CSR_WCOUNT_MASK;
428                 ch->csr |= ((req->size>>3) - 1) << CSR_WCOUNT_SHIFT;
429         }
430
431         if (req->to_memory) {
432                 ch->csr &= ~CSR_DIR;
433                 ch->apb_ptr = req->source_addr;
434                 ch->ahb_ptr = req->dest_addr;
435
436                 apb_addr_wrap = req->source_wrap;
437                 ahb_addr_wrap = req->dest_wrap;
438                 apb_bus_width = req->source_bus_width;
439                 ahb_bus_width = req->dest_bus_width;
440
441         } else {
442                 ch->csr |= CSR_DIR;
443                 ch->apb_ptr = req->dest_addr;
444                 ch->ahb_ptr = req->source_addr;
445
446                 apb_addr_wrap = req->dest_wrap;
447                 ahb_addr_wrap = req->source_wrap;
448                 apb_bus_width = req->dest_bus_width;
449                 ahb_bus_width = req->source_bus_width;
450         }
451
452         apb_addr_wrap >>= 2;
453         ahb_addr_wrap >>= 2;
454
455         /* set address wrap for APB size */
456         index = 0;
457         do  {
458                 if (apb_addr_wrap_table[index] == apb_addr_wrap)
459                         break;
460                 index++;
461         } while (index < ARRAY_SIZE(apb_addr_wrap_table));
462         BUG_ON(index == ARRAY_SIZE(apb_addr_wrap_table));
463         ch->apb_seq &= ~APB_SEQ_WRAP_MASK;
464         ch->apb_seq |= index << APB_SEQ_WRAP_SHIFT;
465
466         /* set address wrap for AHB size */
467         index = 0;
468         do  {
469                 if (ahb_addr_wrap_table[index] == ahb_addr_wrap)
470                         break;
471                 index++;
472         } while (index < ARRAY_SIZE(ahb_addr_wrap_table));
473         BUG_ON(index == ARRAY_SIZE(ahb_addr_wrap_table));
474         ch->ahb_seq &= ~AHB_SEQ_WRAP_MASK;
475         ch->ahb_seq |= index << AHB_SEQ_WRAP_SHIFT;
476
477         for (index = 0; index < ARRAY_SIZE(bus_width_table); index++) {
478                 if (bus_width_table[index] == ahb_bus_width)
479                         break;
480         }
481         BUG_ON(index == ARRAY_SIZE(bus_width_table));
482         ch->ahb_seq &= ~AHB_SEQ_BUS_WIDTH_MASK;
483         ch->ahb_seq |= index << AHB_SEQ_BUS_WIDTH_SHIFT;
484
485         for (index = 0; index < ARRAY_SIZE(bus_width_table); index++) {
486                 if (bus_width_table[index] == apb_bus_width)
487                         break;
488         }
489         BUG_ON(index == ARRAY_SIZE(bus_width_table));
490         ch->apb_seq &= ~APB_SEQ_BUS_WIDTH_MASK;
491         ch->apb_seq |= index << APB_SEQ_BUS_WIDTH_SHIFT;
492
493         ch->csr |= CSR_IE_EOC;
494
495         /* update hw registers with the shadow */
496         writel(ch->csr, ch->addr + APB_DMA_CHAN_CSR);
497         writel(ch->apb_seq, ch->addr + APB_DMA_CHAN_APB_SEQ);
498         writel(ch->apb_ptr, ch->addr + APB_DMA_CHAN_APB_PTR);
499         writel(ch->ahb_seq, ch->addr + APB_DMA_CHAN_AHB_SEQ);
500         writel(ch->ahb_ptr, ch->addr + APB_DMA_CHAN_AHB_PTR);
501
502         csr = ch->csr | CSR_ENB;
503         writel(csr, ch->addr + APB_DMA_CHAN_CSR);
504
505         req->status = TEGRA_DMA_REQ_INFLIGHT;
506 }
507
508 static void tegra_dma_init_hw(struct tegra_dma_channel *ch)
509 {
510         /* One shot with an interrupt to CPU after transfer */
511         ch->csr = CSR_ONCE | CSR_IE_EOC;
512         ch->ahb_seq = AHB_SEQ_BUS_WIDTH_32 | AHB_SEQ_INTR_ENB;
513         ch->apb_seq = APB_SEQ_BUS_WIDTH_32 | 1 << APB_SEQ_WRAP_SHIFT;
514 }
515
516 static void handle_oneshot_dma(struct tegra_dma_channel *ch)
517 {
518         struct tegra_dma_req *req;
519
520         spin_lock(&ch->lock);
521         if (list_empty(&ch->list)) {
522                 spin_unlock(&ch->lock);
523                 return;
524         }
525
526         req = list_entry(ch->list.next, typeof(*req), node);
527         if (req) {
528                 int bytes_transferred;
529
530                 bytes_transferred =
531                         (ch->csr & CSR_WCOUNT_MASK) >> CSR_WCOUNT_SHIFT;
532                 bytes_transferred += 1;
533                 bytes_transferred <<= 2;
534
535                 list_del(&req->node);
536                 req->bytes_transferred = bytes_transferred;
537                 req->status = TEGRA_DMA_REQ_SUCCESS;
538
539                 spin_unlock(&ch->lock);
540                 /* Callback should be called without any lock */
541                 pr_debug("%s: transferred %d bytes\n", __func__,
542                         req->bytes_transferred);
543                 req->complete(req);
544                 spin_lock(&ch->lock);
545         }
546
547         if (!list_empty(&ch->list)) {
548                 req = list_entry(ch->list.next, typeof(*req), node);
549                 /* the complete function we just called may have enqueued
550                    another req, in which case dma has already started */
551                 if (req->status != TEGRA_DMA_REQ_INFLIGHT)
552                         tegra_dma_update_hw(ch, req);
553         }
554         spin_unlock(&ch->lock);
555 }
556
557 static void handle_continuous_dma(struct tegra_dma_channel *ch)
558 {
559         struct tegra_dma_req *req;
560
561         spin_lock(&ch->lock);
562         if (list_empty(&ch->list)) {
563                 spin_unlock(&ch->lock);
564                 return;
565         }
566
567         req = list_entry(ch->list.next, typeof(*req), node);
568         if (req) {
569                 if (req->buffer_status == TEGRA_DMA_REQ_BUF_STATUS_EMPTY) {
570                         /* Load the next request into the hardware, if available
571                          * */
572                         if (!list_is_last(&req->node, &ch->list)) {
573                                 struct tegra_dma_req *next_req;
574
575                                 next_req = list_entry(req->node.next,
576                                         typeof(*next_req), node);
577                                 tegra_dma_update_hw_partial(ch, next_req);
578                         }
579                         req->buffer_status = TEGRA_DMA_REQ_BUF_STATUS_HALF_FULL;
580                         req->status = TEGRA_DMA_REQ_SUCCESS;
581                         /* DMA lock is NOT held when callback is called */
582                         spin_unlock(&ch->lock);
583                         if (likely(req->threshold))
584                                 req->threshold(req);
585                         return;
586
587                 } else if (req->buffer_status ==
588                         TEGRA_DMA_REQ_BUF_STATUS_HALF_FULL) {
589                         /* Callback when the buffer is completely full (i.e on
590                          * the second  interrupt */
591                         int bytes_transferred;
592
593                         bytes_transferred =
594                                 (ch->csr & CSR_WCOUNT_MASK) >> CSR_WCOUNT_SHIFT;
595                         bytes_transferred += 1;
596                         bytes_transferred <<= 3;
597
598                         req->buffer_status = TEGRA_DMA_REQ_BUF_STATUS_FULL;
599                         req->bytes_transferred = bytes_transferred;
600                         req->status = TEGRA_DMA_REQ_SUCCESS;
601                         list_del(&req->node);
602
603                         /* DMA lock is NOT held when callbak is called */
604                         spin_unlock(&ch->lock);
605                         req->complete(req);
606                         return;
607
608                 } else {
609                         BUG();
610                 }
611         }
612         spin_unlock(&ch->lock);
613 }
614
615 static irqreturn_t dma_isr(int irq, void *data)
616 {
617         struct tegra_dma_channel *ch = data;
618         unsigned long status;
619
620         status = readl(ch->addr + APB_DMA_CHAN_STA);
621         if (status & STA_ISE_EOC)
622                 writel(status, ch->addr + APB_DMA_CHAN_STA);
623         else {
624                 pr_warning("Got a spurious ISR for DMA channel %d\n", ch->id);
625                 return IRQ_HANDLED;
626         }
627         return IRQ_WAKE_THREAD;
628 }
629
630 static irqreturn_t dma_thread_fn(int irq, void *data)
631 {
632         struct tegra_dma_channel *ch = data;
633
634         if (ch->mode & TEGRA_DMA_MODE_ONESHOT)
635                 handle_oneshot_dma(ch);
636         else
637                 handle_continuous_dma(ch);
638
639
640         return IRQ_HANDLED;
641 }
642
643 int __init tegra_dma_init(void)
644 {
645         int ret = 0;
646         int i;
647         unsigned int irq;
648         void __iomem *addr;
649
650         addr = IO_ADDRESS(TEGRA_APB_DMA_BASE);
651         writel(GEN_ENABLE, addr + APB_DMA_GEN);
652         writel(0, addr + APB_DMA_CNTRL);
653         writel(0xFFFFFFFFul >> (31 - TEGRA_SYSTEM_DMA_CH_MAX),
654                addr + APB_DMA_IRQ_MASK_SET);
655
656         memset(channel_usage, 0, sizeof(channel_usage));
657         memset(dma_channels, 0, sizeof(dma_channels));
658
659         /* Reserve all the channels we are not supposed to touch */
660         for (i = 0; i < TEGRA_SYSTEM_DMA_CH_MIN; i++)
661                 __set_bit(i, channel_usage);
662
663         for (i = TEGRA_SYSTEM_DMA_CH_MIN; i <= TEGRA_SYSTEM_DMA_CH_MAX; i++) {
664                 struct tegra_dma_channel *ch = &dma_channels[i];
665
666                 __clear_bit(i, channel_usage);
667
668                 ch->id = i;
669                 snprintf(ch->name, TEGRA_DMA_NAME_SIZE, "dma_channel_%d", i);
670
671                 ch->addr = IO_ADDRESS(TEGRA_APB_DMA_CH0_BASE +
672                         TEGRA_APB_DMA_CH0_SIZE * i);
673
674                 spin_lock_init(&ch->lock);
675                 INIT_LIST_HEAD(&ch->list);
676                 tegra_dma_init_hw(ch);
677
678                 irq = INT_APB_DMA_CH0 + i;
679                 ret = request_threaded_irq(irq, dma_isr, dma_thread_fn, 0,
680                         dma_channels[i].name, ch);
681                 if (ret) {
682                         pr_err("Failed to register IRQ %d for DMA %d\n",
683                                 irq, i);
684                         goto fail;
685                 }
686                 ch->irq = irq;
687         }
688         /* mark the shared channel allocated */
689         __set_bit(TEGRA_SYSTEM_DMA_CH_MIN, channel_usage);
690
691         for (i = TEGRA_SYSTEM_DMA_CH_MAX+1; i < NV_DMA_MAX_CHANNELS; i++)
692                 __set_bit(i, channel_usage);
693
694         return ret;
695 fail:
696         writel(0, addr + APB_DMA_GEN);
697         for (i = TEGRA_SYSTEM_DMA_CH_MIN; i <= TEGRA_SYSTEM_DMA_CH_MAX; i++) {
698                 struct tegra_dma_channel *ch = &dma_channels[i];
699                 if (ch->irq)
700                         free_irq(ch->irq, ch);
701         }
702         return ret;
703 }
704
705 #ifdef CONFIG_PM
706 static u32 apb_dma[5*TEGRA_SYSTEM_DMA_CH_NR + 3];
707
708 void tegra_dma_suspend(void)
709 {
710         void __iomem *addr = IO_ADDRESS(TEGRA_APB_DMA_BASE);
711         u32 *ctx = apb_dma;
712         int i;
713
714         *ctx++ = readl(addr + APB_DMA_GEN);
715         *ctx++ = readl(addr + APB_DMA_CNTRL);
716         *ctx++ = readl(addr + APB_DMA_IRQ_MASK);
717
718         for (i = 0; i < TEGRA_SYSTEM_DMA_CH_NR; i++) {
719                 addr = IO_ADDRESS(TEGRA_APB_DMA_CH0_BASE +
720                                   TEGRA_APB_DMA_CH0_SIZE * i);
721
722                 *ctx++ = readl(addr + APB_DMA_CHAN_CSR);
723                 *ctx++ = readl(addr + APB_DMA_CHAN_AHB_PTR);
724                 *ctx++ = readl(addr + APB_DMA_CHAN_AHB_SEQ);
725                 *ctx++ = readl(addr + APB_DMA_CHAN_APB_PTR);
726                 *ctx++ = readl(addr + APB_DMA_CHAN_APB_SEQ);
727         }
728 }
729
730 void tegra_dma_resume(void)
731 {
732         void __iomem *addr = IO_ADDRESS(TEGRA_APB_DMA_BASE);
733         u32 *ctx = apb_dma;
734         int i;
735
736         writel(*ctx++, addr + APB_DMA_GEN);
737         writel(*ctx++, addr + APB_DMA_CNTRL);
738         writel(*ctx++, addr + APB_DMA_IRQ_MASK);
739
740         for (i = 0; i < TEGRA_SYSTEM_DMA_CH_NR; i++) {
741                 addr = IO_ADDRESS(TEGRA_APB_DMA_CH0_BASE +
742                                   TEGRA_APB_DMA_CH0_SIZE * i);
743
744                 writel(*ctx++, addr + APB_DMA_CHAN_CSR);
745                 writel(*ctx++, addr + APB_DMA_CHAN_AHB_PTR);
746                 writel(*ctx++, addr + APB_DMA_CHAN_AHB_SEQ);
747                 writel(*ctx++, addr + APB_DMA_CHAN_APB_PTR);
748                 writel(*ctx++, addr + APB_DMA_CHAN_APB_SEQ);
749         }
750 }
751
752 #endif