1/* 2 * Freescale DMA ALSA SoC PCM driver 3 * 4 * Author: Timur Tabi <timur@freescale.com> 5 * 6 * Copyright 2007-2010 Freescale Semiconductor, Inc. 7 * 8 * This file is licensed under the terms of the GNU General Public License 9 * version 2. This program is licensed "as is" without any warranty of any 10 * kind, whether express or implied. 11 * 12 * This driver implements ASoC support for the Elo DMA controller, which is 13 * the DMA controller on Freescale 83xx, 85xx, and 86xx SOCs. In ALSA terms, 14 * the PCM driver is what handles the DMA buffer. 15 */ 16 17#include <linux/module.h> 18#include <linux/init.h> 19#include <linux/platform_device.h> 20#include <linux/dma-mapping.h> 21#include <linux/interrupt.h> 22#include <linux/delay.h> 23#include <linux/gfp.h> 24#include <linux/of_platform.h> 25#include <linux/list.h> 26#include <linux/slab.h> 27 28#include <sound/core.h> 29#include <sound/pcm.h> 30#include <sound/pcm_params.h> 31#include <sound/soc.h> 32 33#include <asm/io.h> 34 35#include "fsl_dma.h" 36#include "fsl_ssi.h" /* For the offset of stx0 and srx0 */ 37 38/* 39 * The formats that the DMA controller supports, which is anything 40 * that is 8, 16, or 32 bits. 41 */ 42#define FSLDMA_PCM_FORMATS (SNDRV_PCM_FMTBIT_S8 | \ 43 SNDRV_PCM_FMTBIT_U8 | \ 44 SNDRV_PCM_FMTBIT_S16_LE | \ 45 SNDRV_PCM_FMTBIT_S16_BE | \ 46 SNDRV_PCM_FMTBIT_U16_LE | \ 47 SNDRV_PCM_FMTBIT_U16_BE | \ 48 SNDRV_PCM_FMTBIT_S24_LE | \ 49 SNDRV_PCM_FMTBIT_S24_BE | \ 50 SNDRV_PCM_FMTBIT_U24_LE | \ 51 SNDRV_PCM_FMTBIT_U24_BE | \ 52 SNDRV_PCM_FMTBIT_S32_LE | \ 53 SNDRV_PCM_FMTBIT_S32_BE | \ 54 SNDRV_PCM_FMTBIT_U32_LE | \ 55 SNDRV_PCM_FMTBIT_U32_BE) 56 57#define FSLDMA_PCM_RATES (SNDRV_PCM_RATE_5512 | SNDRV_PCM_RATE_8000_192000 | \ 58 SNDRV_PCM_RATE_CONTINUOUS) 59 60struct dma_object { 61 struct snd_soc_platform_driver dai; 62 dma_addr_t ssi_stx_phys; 63 dma_addr_t ssi_srx_phys; 64 unsigned int ssi_fifo_depth; 65 struct ccsr_dma_channel __iomem *channel; 66 unsigned int irq; 67 bool assigned; 68 char path[1]; 69}; 70 71/* 72 * The number of DMA links to use. Two is the bare minimum, but if you 73 * have really small links you might need more. 74 */ 75#define NUM_DMA_LINKS 2 76 77/** fsl_dma_private: p-substream DMA data 78 * 79 * Each substream has a 1-to-1 association with a DMA channel. 80 * 81 * The link[] array is first because it needs to be aligned on a 32-byte 82 * boundary, so putting it first will ensure alignment without padding the 83 * structure. 84 * 85 * @link[]: array of link descriptors 86 * @dma_channel: pointer to the DMA channel's registers 87 * @irq: IRQ for this DMA channel 88 * @substream: pointer to the substream object, needed by the ISR 89 * @ssi_sxx_phys: bus address of the STX or SRX register to use 90 * @ld_buf_phys: physical address of the LD buffer 91 * @current_link: index into link[] of the link currently being processed 92 * @dma_buf_phys: physical address of the DMA buffer 93 * @dma_buf_next: physical address of the next period to process 94 * @dma_buf_end: physical address of the byte after the end of the DMA 95 * @buffer period_size: the size of a single period 96 * @num_periods: the number of periods in the DMA buffer 97 */ 98struct fsl_dma_private { 99 struct fsl_dma_link_descriptor link[NUM_DMA_LINKS]; 100 struct ccsr_dma_channel __iomem *dma_channel; 101 unsigned int irq; 102 struct snd_pcm_substream *substream; 103 dma_addr_t ssi_sxx_phys; 104 unsigned int ssi_fifo_depth; 105 dma_addr_t ld_buf_phys; 106 unsigned int current_link; 107 dma_addr_t dma_buf_phys; 108 dma_addr_t dma_buf_next; 109 dma_addr_t dma_buf_end; 110 size_t period_size; 111 unsigned int num_periods; 112}; 113 114/** 115 * fsl_dma_hardare: define characteristics of the PCM hardware. 116 * 117 * The PCM hardware is the Freescale DMA controller. This structure defines 118 * the capabilities of that hardware. 119 * 120 * Since the sampling rate and data format are not controlled by the DMA 121 * controller, we specify no limits for those values. The only exception is 122 * period_bytes_min, which is set to a reasonably low value to prevent the 123 * DMA controller from generating too many interrupts per second. 124 * 125 * Since each link descriptor has a 32-bit byte count field, we set 126 * period_bytes_max to the largest 32-bit number. We also have no maximum 127 * number of periods. 128 * 129 * Note that we specify SNDRV_PCM_INFO_JOINT_DUPLEX here, but only because a 130 * limitation in the SSI driver requires the sample rates for playback and 131 * capture to be the same. 132 */ 133static const struct snd_pcm_hardware fsl_dma_hardware = { 134 135 .info = SNDRV_PCM_INFO_INTERLEAVED | 136 SNDRV_PCM_INFO_MMAP | 137 SNDRV_PCM_INFO_MMAP_VALID | 138 SNDRV_PCM_INFO_JOINT_DUPLEX | 139 SNDRV_PCM_INFO_PAUSE, 140 .formats = FSLDMA_PCM_FORMATS, 141 .rates = FSLDMA_PCM_RATES, 142 .rate_min = 5512, 143 .rate_max = 192000, 144 .period_bytes_min = 512, /* A reasonable limit */ 145 .period_bytes_max = (u32) -1, 146 .periods_min = NUM_DMA_LINKS, 147 .periods_max = (unsigned int) -1, 148 .buffer_bytes_max = 128 * 1024, /* A reasonable limit */ 149}; 150 151/** 152 * fsl_dma_abort_stream: tell ALSA that the DMA transfer has aborted 153 * 154 * This function should be called by the ISR whenever the DMA controller 155 * halts data transfer. 156 */ 157static void fsl_dma_abort_stream(struct snd_pcm_substream *substream) 158{ 159 unsigned long flags; 160 161 snd_pcm_stream_lock_irqsave(substream, flags); 162 163 if (snd_pcm_running(substream)) 164 snd_pcm_stop(substream, SNDRV_PCM_STATE_XRUN); 165 166 snd_pcm_stream_unlock_irqrestore(substream, flags); 167} 168 169/** 170 * fsl_dma_update_pointers - update LD pointers to point to the next period 171 * 172 * As each period is completed, this function changes the the link 173 * descriptor pointers for that period to point to the next period. 174 */ 175static void fsl_dma_update_pointers(struct fsl_dma_private *dma_private) 176{ 177 struct fsl_dma_link_descriptor *link = 178 &dma_private->link[dma_private->current_link]; 179 180 /* Update our link descriptors to point to the next period. On a 36-bit 181 * system, we also need to update the ESAD bits. We also set (keep) the 182 * snoop bits. See the comments in fsl_dma_hw_params() about snooping. 183 */ 184 if (dma_private->substream->stream == SNDRV_PCM_STREAM_PLAYBACK) { 185 link->source_addr = cpu_to_be32(dma_private->dma_buf_next); 186#ifdef CONFIG_PHYS_64BIT 187 link->source_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP | 188 upper_32_bits(dma_private->dma_buf_next)); 189#endif 190 } else { 191 link->dest_addr = cpu_to_be32(dma_private->dma_buf_next); 192#ifdef CONFIG_PHYS_64BIT 193 link->dest_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP | 194 upper_32_bits(dma_private->dma_buf_next)); 195#endif 196 } 197 198 /* Update our variables for next time */ 199 dma_private->dma_buf_next += dma_private->period_size; 200 201 if (dma_private->dma_buf_next >= dma_private->dma_buf_end) 202 dma_private->dma_buf_next = dma_private->dma_buf_phys; 203 204 if (++dma_private->current_link >= NUM_DMA_LINKS) 205 dma_private->current_link = 0; 206} 207 208/** 209 * fsl_dma_isr: interrupt handler for the DMA controller 210 * 211 * @irq: IRQ of the DMA channel 212 * @dev_id: pointer to the dma_private structure for this DMA channel 213 */ 214static irqreturn_t fsl_dma_isr(int irq, void *dev_id) 215{ 216 struct fsl_dma_private *dma_private = dev_id; 217 struct snd_pcm_substream *substream = dma_private->substream; 218 struct snd_soc_pcm_runtime *rtd = substream->private_data; 219 struct device *dev = rtd->platform->dev; 220 struct ccsr_dma_channel __iomem *dma_channel = dma_private->dma_channel; 221 irqreturn_t ret = IRQ_NONE; 222 u32 sr, sr2 = 0; 223 224 /* We got an interrupt, so read the status register to see what we 225 were interrupted for. 226 */ 227 sr = in_be32(&dma_channel->sr); 228 229 if (sr & CCSR_DMA_SR_TE) { 230 dev_err(dev, "dma transmit error\n"); 231 fsl_dma_abort_stream(substream); 232 sr2 |= CCSR_DMA_SR_TE; 233 ret = IRQ_HANDLED; 234 } 235 236 if (sr & CCSR_DMA_SR_CH) 237 ret = IRQ_HANDLED; 238 239 if (sr & CCSR_DMA_SR_PE) { 240 dev_err(dev, "dma programming error\n"); 241 fsl_dma_abort_stream(substream); 242 sr2 |= CCSR_DMA_SR_PE; 243 ret = IRQ_HANDLED; 244 } 245 246 if (sr & CCSR_DMA_SR_EOLNI) { 247 sr2 |= CCSR_DMA_SR_EOLNI; 248 ret = IRQ_HANDLED; 249 } 250 251 if (sr & CCSR_DMA_SR_CB) 252 ret = IRQ_HANDLED; 253 254 if (sr & CCSR_DMA_SR_EOSI) { 255 /* Tell ALSA we completed a period. */ 256 snd_pcm_period_elapsed(substream); 257 258 /* 259 * Update our link descriptors to point to the next period. We 260 * only need to do this if the number of periods is not equal to 261 * the number of links. 262 */ 263 if (dma_private->num_periods != NUM_DMA_LINKS) 264 fsl_dma_update_pointers(dma_private); 265 266 sr2 |= CCSR_DMA_SR_EOSI; 267 ret = IRQ_HANDLED; 268 } 269 270 if (sr & CCSR_DMA_SR_EOLSI) { 271 sr2 |= CCSR_DMA_SR_EOLSI; 272 ret = IRQ_HANDLED; 273 } 274 275 /* Clear the bits that we set */ 276 if (sr2) 277 out_be32(&dma_channel->sr, sr2); 278 279 return ret; 280} 281 282/** 283 * fsl_dma_new: initialize this PCM driver. 284 * 285 * This function is called when the codec driver calls snd_soc_new_pcms(), 286 * once for each .dai_link in the machine driver's snd_soc_card 287 * structure. 288 * 289 * snd_dma_alloc_pages() is just a front-end to dma_alloc_coherent(), which 290 * (currently) always allocates the DMA buffer in lowmem, even if GFP_HIGHMEM 291 * is specified. Therefore, any DMA buffers we allocate will always be in low 292 * memory, but we support for 36-bit physical addresses anyway. 293 * 294 * Regardless of where the memory is actually allocated, since the device can 295 * technically DMA to any 36-bit address, we do need to set the DMA mask to 36. 296 */ 297static int fsl_dma_new(struct snd_card *card, struct snd_soc_dai *dai, 298 struct snd_pcm *pcm) 299{ 300 static u64 fsl_dma_dmamask = DMA_BIT_MASK(36); 301 int ret; 302 303 if (!card->dev->dma_mask) 304 card->dev->dma_mask = &fsl_dma_dmamask; 305 306 if (!card->dev->coherent_dma_mask) 307 card->dev->coherent_dma_mask = fsl_dma_dmamask; 308 309 /* Some codecs have separate DAIs for playback and capture, so we 310 * should allocate a DMA buffer only for the streams that are valid. 311 */ 312 313 if (pcm->streams[0].substream) { 314 ret = snd_dma_alloc_pages(SNDRV_DMA_TYPE_DEV, card->dev, 315 fsl_dma_hardware.buffer_bytes_max, 316 &pcm->streams[0].substream->dma_buffer); 317 if (ret) { 318 dev_err(card->dev, "can't alloc playback dma buffer\n"); 319 return ret; 320 } 321 } 322 323 if (pcm->streams[1].substream) { 324 ret = snd_dma_alloc_pages(SNDRV_DMA_TYPE_DEV, card->dev, 325 fsl_dma_hardware.buffer_bytes_max, 326 &pcm->streams[1].substream->dma_buffer); 327 if (ret) { 328 dev_err(card->dev, "can't alloc capture dma buffer\n"); 329 snd_dma_free_pages(&pcm->streams[0].substream->dma_buffer); 330 return ret; 331 } 332 } 333 334 return 0; 335} 336 337/** 338 * fsl_dma_open: open a new substream. 339 * 340 * Each substream has its own DMA buffer. 341 * 342 * ALSA divides the DMA buffer into N periods. We create NUM_DMA_LINKS link 343 * descriptors that ping-pong from one period to the next. For example, if 344 * there are six periods and two link descriptors, this is how they look 345 * before playback starts: 346 * 347 * The last link descriptor 348 * ____________ points back to the first 349 * | | 350 * V | 351 * ___ ___ | 352 * | |->| |->| 353 * |___| |___| 354 * | | 355 * | | 356 * V V 357 * _________________________________________ 358 * | | | | | | | The DMA buffer is 359 * | | | | | | | divided into 6 parts 360 * |______|______|______|______|______|______| 361 * 362 * and here's how they look after the first period is finished playing: 363 * 364 * ____________ 365 * | | 366 * V | 367 * ___ ___ | 368 * | |->| |->| 369 * |___| |___| 370 * | | 371 * |______________ 372 * | | 373 * V V 374 * _________________________________________ 375 * | | | | | | | 376 * | | | | | | | 377 * |______|______|______|______|______|______| 378 * 379 * The first link descriptor now points to the third period. The DMA 380 * controller is currently playing the second period. When it finishes, it 381 * will jump back to the first descriptor and play the third period. 382 * 383 * There are four reasons we do this: 384 * 385 * 1. The only way to get the DMA controller to automatically restart the 386 * transfer when it gets to the end of the buffer is to use chaining 387 * mode. Basic direct mode doesn't offer that feature. 388 * 2. We need to receive an interrupt at the end of every period. The DMA 389 * controller can generate an interrupt at the end of every link transfer 390 * (aka segment). Making each period into a DMA segment will give us the 391 * interrupts we need. 392 * 3. By creating only two link descriptors, regardless of the number of 393 * periods, we do not need to reallocate the link descriptors if the 394 * number of periods changes. 395 * 4. All of the audio data is still stored in a single, contiguous DMA 396 * buffer, which is what ALSA expects. We're just dividing it into 397 * contiguous parts, and creating a link descriptor for each one. 398 */ 399static int fsl_dma_open(struct snd_pcm_substream *substream) 400{ 401 struct snd_pcm_runtime *runtime = substream->runtime; 402 struct snd_soc_pcm_runtime *rtd = substream->private_data; 403 struct device *dev = rtd->platform->dev; 404 struct dma_object *dma = 405 container_of(rtd->platform->driver, struct dma_object, dai); 406 struct fsl_dma_private *dma_private; 407 struct ccsr_dma_channel __iomem *dma_channel; 408 dma_addr_t ld_buf_phys; 409 u64 temp_link; /* Pointer to next link descriptor */ 410 u32 mr; 411 unsigned int channel; 412 int ret = 0; 413 unsigned int i; 414 415 /* 416 * Reject any DMA buffer whose size is not a multiple of the period 417 * size. We need to make sure that the DMA buffer can be evenly divided 418 * into periods. 419 */ 420 ret = snd_pcm_hw_constraint_integer(runtime, 421 SNDRV_PCM_HW_PARAM_PERIODS); 422 if (ret < 0) { 423 dev_err(dev, "invalid buffer size\n"); 424 return ret; 425 } 426 427 channel = substream->stream == SNDRV_PCM_STREAM_PLAYBACK ? 0 : 1; 428 429 if (dma->assigned) { 430 dev_err(dev, "dma channel already assigned\n"); 431 return -EBUSY; 432 } 433 434 dma_private = dma_alloc_coherent(dev, sizeof(struct fsl_dma_private), 435 &ld_buf_phys, GFP_KERNEL); 436 if (!dma_private) { 437 dev_err(dev, "can't allocate dma private data\n"); 438 return -ENOMEM; 439 } 440 if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) 441 dma_private->ssi_sxx_phys = dma->ssi_stx_phys; 442 else 443 dma_private->ssi_sxx_phys = dma->ssi_srx_phys; 444 445 dma_private->ssi_fifo_depth = dma->ssi_fifo_depth; 446 dma_private->dma_channel = dma->channel; 447 dma_private->irq = dma->irq; 448 dma_private->substream = substream; 449 dma_private->ld_buf_phys = ld_buf_phys; 450 dma_private->dma_buf_phys = substream->dma_buffer.addr; 451 452 ret = request_irq(dma_private->irq, fsl_dma_isr, 0, "fsldma-audio", 453 dma_private); 454 if (ret) { 455 dev_err(dev, "can't register ISR for IRQ %u (ret=%i)\n", 456 dma_private->irq, ret); 457 dma_free_coherent(dev, sizeof(struct fsl_dma_private), 458 dma_private, dma_private->ld_buf_phys); 459 return ret; 460 } 461 462 dma->assigned = 1; 463 464 snd_pcm_set_runtime_buffer(substream, &substream->dma_buffer); 465 snd_soc_set_runtime_hwparams(substream, &fsl_dma_hardware); 466 runtime->private_data = dma_private; 467 468 /* Program the fixed DMA controller parameters */ 469 470 dma_channel = dma_private->dma_channel; 471 472 temp_link = dma_private->ld_buf_phys + 473 sizeof(struct fsl_dma_link_descriptor); 474 475 for (i = 0; i < NUM_DMA_LINKS; i++) { 476 dma_private->link[i].next = cpu_to_be64(temp_link); 477 478 temp_link += sizeof(struct fsl_dma_link_descriptor); 479 } 480 /* The last link descriptor points to the first */ 481 dma_private->link[i - 1].next = cpu_to_be64(dma_private->ld_buf_phys); 482 483 /* Tell the DMA controller where the first link descriptor is */ 484 out_be32(&dma_channel->clndar, 485 CCSR_DMA_CLNDAR_ADDR(dma_private->ld_buf_phys)); 486 out_be32(&dma_channel->eclndar, 487 CCSR_DMA_ECLNDAR_ADDR(dma_private->ld_buf_phys)); 488 489 /* The manual says the BCR must be clear before enabling EMP */ 490 out_be32(&dma_channel->bcr, 0); 491 492 /* 493 * Program the mode register for interrupts, external master control, 494 * and source/destination hold. Also clear the Channel Abort bit. 495 */ 496 mr = in_be32(&dma_channel->mr) & 497 ~(CCSR_DMA_MR_CA | CCSR_DMA_MR_DAHE | CCSR_DMA_MR_SAHE); 498 499 /* 500 * We want External Master Start and External Master Pause enabled, 501 * because the SSI is controlling the DMA controller. We want the DMA 502 * controller to be set up in advance, and then we signal only the SSI 503 * to start transferring. 504 * 505 * We want End-Of-Segment Interrupts enabled, because this will generate 506 * an interrupt at the end of each segment (each link descriptor 507 * represents one segment). Each DMA segment is the same thing as an 508 * ALSA period, so this is how we get an interrupt at the end of every 509 * period. 510 * 511 * We want Error Interrupt enabled, so that we can get an error if 512 * the DMA controller is mis-programmed somehow. 513 */ 514 mr |= CCSR_DMA_MR_EOSIE | CCSR_DMA_MR_EIE | CCSR_DMA_MR_EMP_EN | 515 CCSR_DMA_MR_EMS_EN; 516 517 /* For playback, we want the destination address to be held. For 518 capture, set the source address to be held. */ 519 mr |= (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) ? 520 CCSR_DMA_MR_DAHE : CCSR_DMA_MR_SAHE; 521 522 out_be32(&dma_channel->mr, mr); 523 524 return 0; 525} 526 527/** 528 * fsl_dma_hw_params: continue initializing the DMA links 529 * 530 * This function obtains hardware parameters about the opened stream and 531 * programs the DMA controller accordingly. 532 * 533 * One drawback of big-endian is that when copying integers of different 534 * sizes to a fixed-sized register, the address to which the integer must be 535 * copied is dependent on the size of the integer. 536 * 537 * For example, if P is the address of a 32-bit register, and X is a 32-bit 538 * integer, then X should be copied to address P. However, if X is a 16-bit 539 * integer, then it should be copied to P+2. If X is an 8-bit register, 540 * then it should be copied to P+3. 541 * 542 * So for playback of 8-bit samples, the DMA controller must transfer single 543 * bytes from the DMA buffer to the last byte of the STX0 register, i.e. 544 * offset by 3 bytes. For 16-bit samples, the offset is two bytes. 545 * 546 * For 24-bit samples, the offset is 1 byte. However, the DMA controller 547 * does not support 3-byte copies (the DAHTS register supports only 1, 2, 4, 548 * and 8 bytes at a time). So we do not support packed 24-bit samples. 549 * 24-bit data must be padded to 32 bits. 550 */ 551static int fsl_dma_hw_params(struct snd_pcm_substream *substream, 552 struct snd_pcm_hw_params *hw_params) 553{ 554 struct snd_pcm_runtime *runtime = substream->runtime; 555 struct fsl_dma_private *dma_private = runtime->private_data; 556 struct snd_soc_pcm_runtime *rtd = substream->private_data; 557 struct device *dev = rtd->platform->dev; 558 559 /* Number of bits per sample */ 560 unsigned int sample_bits = 561 snd_pcm_format_physical_width(params_format(hw_params)); 562 563 /* Number of bytes per frame */ 564 unsigned int sample_bytes = sample_bits / 8; 565 566 /* Bus address of SSI STX register */ 567 dma_addr_t ssi_sxx_phys = dma_private->ssi_sxx_phys; 568 569 /* Size of the DMA buffer, in bytes */ 570 size_t buffer_size = params_buffer_bytes(hw_params); 571 572 /* Number of bytes per period */ 573 size_t period_size = params_period_bytes(hw_params); 574 575 /* Pointer to next period */ 576 dma_addr_t temp_addr = substream->dma_buffer.addr; 577 578 /* Pointer to DMA controller */ 579 struct ccsr_dma_channel __iomem *dma_channel = dma_private->dma_channel; 580 581 u32 mr; /* DMA Mode Register */ 582 583 unsigned int i; 584 585 /* Initialize our DMA tracking variables */ 586 dma_private->period_size = period_size; 587 dma_private->num_periods = params_periods(hw_params); 588 dma_private->dma_buf_end = dma_private->dma_buf_phys + buffer_size; 589 dma_private->dma_buf_next = dma_private->dma_buf_phys + 590 (NUM_DMA_LINKS * period_size); 591 592 if (dma_private->dma_buf_next >= dma_private->dma_buf_end) 593 /* This happens if the number of periods == NUM_DMA_LINKS */ 594 dma_private->dma_buf_next = dma_private->dma_buf_phys; 595 596 mr = in_be32(&dma_channel->mr) & ~(CCSR_DMA_MR_BWC_MASK | 597 CCSR_DMA_MR_SAHTS_MASK | CCSR_DMA_MR_DAHTS_MASK); 598 599 /* Due to a quirk of the SSI's STX register, the target address 600 * for the DMA operations depends on the sample size. So we calculate 601 * that offset here. While we're at it, also tell the DMA controller 602 * how much data to transfer per sample. 603 */ 604 switch (sample_bits) { 605 case 8: 606 mr |= CCSR_DMA_MR_DAHTS_1 | CCSR_DMA_MR_SAHTS_1; 607 ssi_sxx_phys += 3; 608 break; 609 case 16: 610 mr |= CCSR_DMA_MR_DAHTS_2 | CCSR_DMA_MR_SAHTS_2; 611 ssi_sxx_phys += 2; 612 break; 613 case 32: 614 mr |= CCSR_DMA_MR_DAHTS_4 | CCSR_DMA_MR_SAHTS_4; 615 break; 616 default: 617 /* We should never get here */ 618 dev_err(dev, "unsupported sample size %u\n", sample_bits); 619 return -EINVAL; 620 } 621 622 /* 623 * BWC determines how many bytes are sent/received before the DMA 624 * controller checks the SSI to see if it needs to stop. BWC should 625 * always be a multiple of the frame size, so that we always transmit 626 * whole frames. Each frame occupies two slots in the FIFO. The 627 * parameter for CCSR_DMA_MR_BWC() is rounded down the next power of two 628 * (MR[BWC] can only represent even powers of two). 629 * 630 * To simplify the process, we set BWC to the largest value that is 631 * less than or equal to the FIFO watermark. For playback, this ensures 632 * that we transfer the maximum amount without overrunning the FIFO. 633 * For capture, this ensures that we transfer the maximum amount without 634 * underrunning the FIFO. 635 * 636 * f = SSI FIFO depth 637 * w = SSI watermark value (which equals f - 2) 638 * b = DMA bandwidth count (in bytes) 639 * s = sample size (in bytes, which equals frame_size * 2) 640 * 641 * For playback, we never transmit more than the transmit FIFO 642 * watermark, otherwise we might write more data than the FIFO can hold. 643 * The watermark is equal to the FIFO depth minus two. 644 * 645 * For capture, two equations must hold: 646 * w > f - (b / s) 647 * w >= b / s 648 * 649 * So, b > 2 * s, but b must also be <= s * w. To simplify, we set 650 * b = s * w, which is equal to 651 * (dma_private->ssi_fifo_depth - 2) * sample_bytes. 652 */ 653 mr |= CCSR_DMA_MR_BWC((dma_private->ssi_fifo_depth - 2) * sample_bytes); 654 655 out_be32(&dma_channel->mr, mr); 656 657 for (i = 0; i < NUM_DMA_LINKS; i++) { 658 struct fsl_dma_link_descriptor *link = &dma_private->link[i]; 659 660 link->count = cpu_to_be32(period_size); 661 662 /* The snoop bit tells the DMA controller whether it should tell 663 * the ECM to snoop during a read or write to an address. For 664 * audio, we use DMA to transfer data between memory and an I/O 665 * device (the SSI's STX0 or SRX0 register). Snooping is only 666 * needed if there is a cache, so we need to snoop memory 667 * addresses only. For playback, that means we snoop the source 668 * but not the destination. For capture, we snoop the 669 * destination but not the source. 670 * 671 * Note that failing to snoop properly is unlikely to cause 672 * cache incoherency if the period size is larger than the 673 * size of L1 cache. This is because filling in one period will 674 * flush out the data for the previous period. So if you 675 * increased period_bytes_min to a large enough size, you might 676 * get more performance by not snooping, and you'll still be 677 * okay. You'll need to update fsl_dma_update_pointers() also. 678 */ 679 if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) { 680 link->source_addr = cpu_to_be32(temp_addr); 681 link->source_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP | 682 upper_32_bits(temp_addr)); 683 684 link->dest_addr = cpu_to_be32(ssi_sxx_phys); 685 link->dest_attr = cpu_to_be32(CCSR_DMA_ATR_NOSNOOP | 686 upper_32_bits(ssi_sxx_phys)); 687 } else { 688 link->source_addr = cpu_to_be32(ssi_sxx_phys); 689 link->source_attr = cpu_to_be32(CCSR_DMA_ATR_NOSNOOP | 690 upper_32_bits(ssi_sxx_phys)); 691 692 link->dest_addr = cpu_to_be32(temp_addr); 693 link->dest_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP | 694 upper_32_bits(temp_addr)); 695 } 696 697 temp_addr += period_size; 698 } 699 700 return 0; 701} 702 703/** 704 * fsl_dma_pointer: determine the current position of the DMA transfer 705 * 706 * This function is called by ALSA when ALSA wants to know where in the 707 * stream buffer the hardware currently is. 708 * 709 * For playback, the SAR register contains the physical address of the most 710 * recent DMA transfer. For capture, the value is in the DAR register. 711 * 712 * The base address of the buffer is stored in the source_addr field of the 713 * first link descriptor. 714 */ 715static snd_pcm_uframes_t fsl_dma_pointer(struct snd_pcm_substream *substream) 716{ 717 struct snd_pcm_runtime *runtime = substream->runtime; 718 struct fsl_dma_private *dma_private = runtime->private_data; 719 struct snd_soc_pcm_runtime *rtd = substream->private_data; 720 struct device *dev = rtd->platform->dev; 721 struct ccsr_dma_channel __iomem *dma_channel = dma_private->dma_channel; 722 dma_addr_t position; 723 snd_pcm_uframes_t frames; 724 725 /* Obtain the current DMA pointer, but don't read the ESAD bits if we 726 * only have 32-bit DMA addresses. This function is typically called 727 * in interrupt context, so we need to optimize it. 728 */ 729 if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) { 730 position = in_be32(&dma_channel->sar); 731#ifdef CONFIG_PHYS_64BIT 732 position |= (u64)(in_be32(&dma_channel->satr) & 733 CCSR_DMA_ATR_ESAD_MASK) << 32; 734#endif 735 } else { 736 position = in_be32(&dma_channel->dar); 737#ifdef CONFIG_PHYS_64BIT 738 position |= (u64)(in_be32(&dma_channel->datr) & 739 CCSR_DMA_ATR_ESAD_MASK) << 32; 740#endif 741 } 742 743 /* 744 * When capture is started, the SSI immediately starts to fill its FIFO. 745 * This means that the DMA controller is not started until the FIFO is 746 * full. However, ALSA calls this function before that happens, when 747 * MR.DAR is still zero. In this case, just return zero to indicate 748 * that nothing has been received yet. 749 */ 750 if (!position) 751 return 0; 752 753 if ((position < dma_private->dma_buf_phys) || 754 (position > dma_private->dma_buf_end)) { 755 dev_err(dev, "dma pointer is out of range, halting stream\n"); 756 return SNDRV_PCM_POS_XRUN; 757 } 758 759 frames = bytes_to_frames(runtime, position - dma_private->dma_buf_phys); 760 761 /* 762 * If the current address is just past the end of the buffer, wrap it 763 * around. 764 */ 765 if (frames == runtime->buffer_size) 766 frames = 0; 767 768 return frames; 769} 770 771/** 772 * fsl_dma_hw_free: release resources allocated in fsl_dma_hw_params() 773 * 774 * Release the resources allocated in fsl_dma_hw_params() and de-program the 775 * registers. 776 * 777 * This function can be called multiple times. 778 */ 779static int fsl_dma_hw_free(struct snd_pcm_substream *substream) 780{ 781 struct snd_pcm_runtime *runtime = substream->runtime; 782 struct fsl_dma_private *dma_private = runtime->private_data; 783 784 if (dma_private) { 785 struct ccsr_dma_channel __iomem *dma_channel; 786 787 dma_channel = dma_private->dma_channel; 788 789 /* Stop the DMA */ 790 out_be32(&dma_channel->mr, CCSR_DMA_MR_CA); 791 out_be32(&dma_channel->mr, 0); 792 793 /* Reset all the other registers */ 794 out_be32(&dma_channel->sr, -1); 795 out_be32(&dma_channel->clndar, 0); 796 out_be32(&dma_channel->eclndar, 0); 797 out_be32(&dma_channel->satr, 0); 798 out_be32(&dma_channel->sar, 0); 799 out_be32(&dma_channel->datr, 0); 800 out_be32(&dma_channel->dar, 0); 801 out_be32(&dma_channel->bcr, 0); 802 out_be32(&dma_channel->nlndar, 0); 803 out_be32(&dma_channel->enlndar, 0); 804 } 805 806 return 0; 807} 808 809/** 810 * fsl_dma_close: close the stream. 811 */ 812static int fsl_dma_close(struct snd_pcm_substream *substream) 813{ 814 struct snd_pcm_runtime *runtime = substream->runtime; 815 struct fsl_dma_private *dma_private = runtime->private_data; 816 struct snd_soc_pcm_runtime *rtd = substream->private_data; 817 struct device *dev = rtd->platform->dev; 818 struct dma_object *dma = 819 container_of(rtd->platform->driver, struct dma_object, dai); 820 821 if (dma_private) { 822 if (dma_private->irq) 823 free_irq(dma_private->irq, dma_private); 824 825 if (dma_private->ld_buf_phys) { 826 dma_unmap_single(dev, dma_private->ld_buf_phys, 827 sizeof(dma_private->link), 828 DMA_TO_DEVICE); 829 } 830 831 /* Deallocate the fsl_dma_private structure */ 832 dma_free_coherent(dev, sizeof(struct fsl_dma_private), 833 dma_private, dma_private->ld_buf_phys); 834 substream->runtime->private_data = NULL; 835 } 836 837 dma->assigned = 0; 838 839 return 0; 840} 841 842/* 843 * Remove this PCM driver. 844 */ 845static void fsl_dma_free_dma_buffers(struct snd_pcm *pcm) 846{ 847 struct snd_pcm_substream *substream; 848 unsigned int i; 849 850 for (i = 0; i < ARRAY_SIZE(pcm->streams); i++) { 851 substream = pcm->streams[i].substream; 852 if (substream) { 853 snd_dma_free_pages(&substream->dma_buffer); 854 substream->dma_buffer.area = NULL; 855 substream->dma_buffer.addr = 0; 856 } 857 } 858} 859 860/** 861 * find_ssi_node -- returns the SSI node that points to his DMA channel node 862 * 863 * Although this DMA driver attempts to operate independently of the other 864 * devices, it still needs to determine some information about the SSI device 865 * that it's working with. Unfortunately, the device tree does not contain 866 * a pointer from the DMA channel node to the SSI node -- the pointer goes the 867 * other way. So we need to scan the device tree for SSI nodes until we find 868 * the one that points to the given DMA channel node. It's ugly, but at least 869 * it's contained in this one function. 870 */ 871static struct device_node *find_ssi_node(struct device_node *dma_channel_np) 872{ 873 struct device_node *ssi_np, *np; 874 875 for_each_compatible_node(ssi_np, NULL, "fsl,mpc8610-ssi") { 876 /* Check each DMA phandle to see if it points to us. We 877 * assume that device_node pointers are a valid comparison. 878 */ 879 np = of_parse_phandle(ssi_np, "fsl,playback-dma", 0); 880 if (np == dma_channel_np) 881 return ssi_np; 882 883 np = of_parse_phandle(ssi_np, "fsl,capture-dma", 0); 884 if (np == dma_channel_np) 885 return ssi_np; 886 } 887 888 return NULL; 889} 890 891static struct snd_pcm_ops fsl_dma_ops = { 892 .open = fsl_dma_open, 893 .close = fsl_dma_close, 894 .ioctl = snd_pcm_lib_ioctl, 895 .hw_params = fsl_dma_hw_params, 896 .hw_free = fsl_dma_hw_free, 897 .pointer = fsl_dma_pointer, 898}; 899 900static int __devinit fsl_soc_dma_probe(struct platform_device *pdev) 901 { 902 struct dma_object *dma; 903 struct device_node *np = pdev->dev.of_node; 904 struct device_node *ssi_np; 905 struct resource res; 906 const uint32_t *iprop; 907 int ret; 908 909 /* Find the SSI node that points to us. */ 910 ssi_np = find_ssi_node(np); 911 if (!ssi_np) { 912 dev_err(&pdev->dev, "cannot find parent SSI node\n"); 913 return -ENODEV; 914 } 915 916 ret = of_address_to_resource(ssi_np, 0, &res); 917 if (ret) { 918 dev_err(&pdev->dev, "could not determine resources for %s\n", 919 ssi_np->full_name); 920 of_node_put(ssi_np); 921 return ret; 922 } 923 924 dma = kzalloc(sizeof(*dma) + strlen(np->full_name), GFP_KERNEL); 925 if (!dma) { 926 dev_err(&pdev->dev, "could not allocate dma object\n"); 927 of_node_put(ssi_np); 928 return -ENOMEM; 929 } 930 931 strcpy(dma->path, np->full_name); 932 dma->dai.ops = &fsl_dma_ops; 933 dma->dai.pcm_new = fsl_dma_new; 934 dma->dai.pcm_free = fsl_dma_free_dma_buffers; 935 936 /* Store the SSI-specific information that we need */ 937 dma->ssi_stx_phys = res.start + offsetof(struct ccsr_ssi, stx0); 938 dma->ssi_srx_phys = res.start + offsetof(struct ccsr_ssi, srx0); 939 940 iprop = of_get_property(ssi_np, "fsl,fifo-depth", NULL); 941 if (iprop) 942 dma->ssi_fifo_depth = *iprop; 943 else 944 /* Older 8610 DTs didn't have the fifo-depth property */ 945 dma->ssi_fifo_depth = 8; 946 947 of_node_put(ssi_np); 948 949 ret = snd_soc_register_platform(&pdev->dev, &dma->dai); 950 if (ret) { 951 dev_err(&pdev->dev, "could not register platform\n"); 952 kfree(dma); 953 return ret; 954 } 955 956 dma->channel = of_iomap(np, 0); 957 dma->irq = irq_of_parse_and_map(np, 0); 958 959 dev_set_drvdata(&pdev->dev, dma); 960 961 return 0; 962} 963 964static int __devexit fsl_soc_dma_remove(struct platform_device *pdev) 965{ 966 struct dma_object *dma = dev_get_drvdata(&pdev->dev); 967 968 snd_soc_unregister_platform(&pdev->dev); 969 iounmap(dma->channel); 970 irq_dispose_mapping(dma->irq); 971 kfree(dma); 972 973 return 0; 974} 975 976static const struct of_device_id fsl_soc_dma_ids[] = { 977 { .compatible = "fsl,ssi-dma-channel", }, 978 {} 979}; 980MODULE_DEVICE_TABLE(of, fsl_soc_dma_ids); 981 982static struct platform_driver fsl_soc_dma_driver = { 983 .driver = { 984 .name = "fsl-pcm-audio", 985 .owner = THIS_MODULE, 986 .of_match_table = fsl_soc_dma_ids, 987 }, 988 .probe = fsl_soc_dma_probe, 989 .remove = __devexit_p(fsl_soc_dma_remove), 990}; 991 992static int __init fsl_soc_dma_init(void) 993{ 994 pr_info("Freescale Elo DMA ASoC PCM Driver\n"); 995 996 return platform_driver_register(&fsl_soc_dma_driver); 997} 998 999static void __exit fsl_soc_dma_exit(void) 1000{ 1001 platform_driver_unregister(&fsl_soc_dma_driver); 1002} 1003 1004module_init(fsl_soc_dma_init); 1005module_exit(fsl_soc_dma_exit); 1006 1007MODULE_AUTHOR("Timur Tabi <timur@freescale.com>"); 1008MODULE_DESCRIPTION("Freescale Elo DMA ASoC PCM Driver"); 1009MODULE_LICENSE("GPL v2"); 1010