From 85ef2375ef2ebbb2bf660ad3a27c644d0ebf1b1a Mon Sep 17 00:00:00 2001 From: Timur Tabi Date: Thu, 5 Feb 2009 17:56:02 -0600 Subject: [PATCH] ASoC: optimize init sequence of Freescale MPC8610 sound drivers In the Freescale MPC8610 sound drivers, relocate all code from the _prepare functions into the corresponding _hw_params functions. These drivers assumed that the sample size is known in the _prepare function and not in the _hw_params function, but this is not true. Move the code in fsl_dma_prepare() into fsl_dma_hw_param(). Create fsl_ssi_hw_params() and move the code from fsl_ssi_prepare() into it. Turn off snooping for DMA operations to/from I/O registers, since that's not necessary. Signed-off-by: Timur Tabi Signed-off-by: Mark Brown --- sound/soc/fsl/fsl_dma.c | 178 +++++++++++++++++++++++------------------------- sound/soc/fsl/fsl_ssi.c | 19 +++--- 2 files changed, 94 insertions(+), 103 deletions(-) diff --git a/sound/soc/fsl/fsl_dma.c b/sound/soc/fsl/fsl_dma.c index 64993ed..58a3fa4 100644 --- a/sound/soc/fsl/fsl_dma.c +++ b/sound/soc/fsl/fsl_dma.c @@ -464,11 +464,7 @@ static int fsl_dma_open(struct snd_pcm_substream *substream) sizeof(struct fsl_dma_link_descriptor); for (i = 0; i < NUM_DMA_LINKS; i++) { - struct fsl_dma_link_descriptor *link = &dma_private->link[i]; - - link->source_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP); - link->dest_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP); - link->next = cpu_to_be64(temp_link); + dma_private->link[i].next = cpu_to_be64(temp_link); temp_link += sizeof(struct fsl_dma_link_descriptor); } @@ -525,79 +521,9 @@ static int fsl_dma_open(struct snd_pcm_substream *substream) * This function obtains hardware parameters about the opened stream and * programs the DMA controller accordingly. * - * Note that due to a quirk of the SSI's STX register, the target address - * for the DMA operations depends on the sample size. So we don't program - * the dest_addr (for playback -- source_addr for capture) fields in the - * link descriptors here. We do that in fsl_dma_prepare() - */ -static int fsl_dma_hw_params(struct snd_pcm_substream *substream, - struct snd_pcm_hw_params *hw_params) -{ - struct snd_pcm_runtime *runtime = substream->runtime; - struct fsl_dma_private *dma_private = runtime->private_data; - - dma_addr_t temp_addr; /* Pointer to next period */ - - unsigned int i; - - /* Get all the parameters we need */ - size_t buffer_size = params_buffer_bytes(hw_params); - size_t period_size = params_period_bytes(hw_params); - - /* Initialize our DMA tracking variables */ - dma_private->period_size = period_size; - dma_private->num_periods = params_periods(hw_params); - dma_private->dma_buf_end = dma_private->dma_buf_phys + buffer_size; - dma_private->dma_buf_next = dma_private->dma_buf_phys + - (NUM_DMA_LINKS * period_size); - if (dma_private->dma_buf_next >= dma_private->dma_buf_end) - dma_private->dma_buf_next = dma_private->dma_buf_phys; - - /* - * The actual address in STX0 (destination for playback, source for - * capture) is based on the sample size, but we don't know the sample - * size in this function, so we'll have to adjust that later. See - * comments in fsl_dma_prepare(). - * - * The DMA controller does not have a cache, so the CPU does not - * need to tell it to flush its cache. However, the DMA - * controller does need to tell the CPU to flush its cache. - * That's what the SNOOP bit does. - * - * Also, even though the DMA controller supports 36-bit addressing, for - * simplicity we currently support only 32-bit addresses for the audio - * buffer itself. - */ - temp_addr = substream->dma_buffer.addr; - - for (i = 0; i < NUM_DMA_LINKS; i++) { - struct fsl_dma_link_descriptor *link = &dma_private->link[i]; - - link->count = cpu_to_be32(period_size); - - if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) - link->source_addr = cpu_to_be32(temp_addr); - else - link->dest_addr = cpu_to_be32(temp_addr); - - temp_addr += period_size; - } - - return 0; -} - -/** - * fsl_dma_prepare - prepare the DMA registers for playback. - * - * This function is called after the specifics of the audio data are known, - * i.e. snd_pcm_runtime is initialized. - * - * In this function, we finish programming the registers of the DMA - * controller that are dependent on the sample size. - * - * One of the drawbacks with big-endian is that when copying integers of - * different sizes to a fixed-sized register, the address to which the - * integer must be copied is dependent on the size of the integer. + * One drawback of big-endian is that when copying integers of different + * sizes to a fixed-sized register, the address to which the integer must be + * copied is dependent on the size of the integer. * * For example, if P is the address of a 32-bit register, and X is a 32-bit * integer, then X should be copied to address P. However, if X is a 16-bit @@ -613,22 +539,58 @@ static int fsl_dma_hw_params(struct snd_pcm_substream *substream, * and 8 bytes at a time). So we do not support packed 24-bit samples. * 24-bit data must be padded to 32 bits. */ -static int fsl_dma_prepare(struct snd_pcm_substream *substream) +static int fsl_dma_hw_params(struct snd_pcm_substream *substream, + struct snd_pcm_hw_params *hw_params) { struct snd_pcm_runtime *runtime = substream->runtime; struct fsl_dma_private *dma_private = runtime->private_data; + + /* Number of bits per sample */ + unsigned int sample_size = + snd_pcm_format_physical_width(params_format(hw_params)); + + /* Number of bytes per frame */ + unsigned int frame_size = 2 * (sample_size / 8); + + /* Bus address of SSI STX register */ + dma_addr_t ssi_sxx_phys = dma_private->ssi_sxx_phys; + + /* Size of the DMA buffer, in bytes */ + size_t buffer_size = params_buffer_bytes(hw_params); + + /* Number of bytes per period */ + size_t period_size = params_period_bytes(hw_params); + + /* Pointer to next period */ + dma_addr_t temp_addr = substream->dma_buffer.addr; + + /* Pointer to DMA controller */ struct ccsr_dma_channel __iomem *dma_channel = dma_private->dma_channel; - u32 mr; + + u32 mr; /* DMA Mode Register */ + unsigned int i; - dma_addr_t ssi_sxx_phys; /* Bus address of SSI STX register */ - unsigned int frame_size; /* Number of bytes per frame */ - ssi_sxx_phys = dma_private->ssi_sxx_phys; + /* Initialize our DMA tracking variables */ + dma_private->period_size = period_size; + dma_private->num_periods = params_periods(hw_params); + dma_private->dma_buf_end = dma_private->dma_buf_phys + buffer_size; + dma_private->dma_buf_next = dma_private->dma_buf_phys + + (NUM_DMA_LINKS * period_size); + + if (dma_private->dma_buf_next >= dma_private->dma_buf_end) + /* This happens if the number of periods == NUM_DMA_LINKS */ + dma_private->dma_buf_next = dma_private->dma_buf_phys; mr = in_be32(&dma_channel->mr) & ~(CCSR_DMA_MR_BWC_MASK | CCSR_DMA_MR_SAHTS_MASK | CCSR_DMA_MR_DAHTS_MASK); - switch (runtime->sample_bits) { + /* Due to a quirk of the SSI's STX register, the target address + * for the DMA operations depends on the sample size. So we calculate + * that offset here. While we're at it, also tell the DMA controller + * how much data to transfer per sample. + */ + switch (sample_size) { case 8: mr |= CCSR_DMA_MR_DAHTS_1 | CCSR_DMA_MR_SAHTS_1; ssi_sxx_phys += 3; @@ -641,12 +603,12 @@ static int fsl_dma_prepare(struct snd_pcm_substream *substream) mr |= CCSR_DMA_MR_DAHTS_4 | CCSR_DMA_MR_SAHTS_4; break; default: + /* We should never get here */ dev_err(substream->pcm->card->dev, - "unsupported sample size %u\n", runtime->sample_bits); + "unsupported sample size %u\n", sample_size); return -EINVAL; } - frame_size = runtime->frame_bits / 8; /* * BWC should always be a multiple of the frame size. BWC determines * how many bytes are sent/received before the DMA controller checks the @@ -655,7 +617,6 @@ static int fsl_dma_prepare(struct snd_pcm_substream *substream) * capture, the receive FIFO is triggered when it contains one frame, so * we want to receive one frame at a time. */ - if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) mr |= CCSR_DMA_MR_BWC(2 * frame_size); else @@ -663,16 +624,48 @@ static int fsl_dma_prepare(struct snd_pcm_substream *substream) out_be32(&dma_channel->mr, mr); - /* - * Program the address of the DMA transfer to/from the SSI. - */ for (i = 0; i < NUM_DMA_LINKS; i++) { struct fsl_dma_link_descriptor *link = &dma_private->link[i]; - if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) + link->count = cpu_to_be32(period_size); + + /* Even though the DMA controller supports 36-bit addressing, + * for simplicity we allow only 32-bit addresses for the audio + * buffer itself. This was enforced in fsl_dma_new() with the + * DMA mask. + * + * The snoop bit tells the DMA controller whether it should tell + * the ECM to snoop during a read or write to an address. For + * audio, we use DMA to transfer data between memory and an I/O + * device (the SSI's STX0 or SRX0 register). Snooping is only + * needed if there is a cache, so we need to snoop memory + * addresses only. For playback, that means we snoop the source + * but not the destination. For capture, we snoop the + * destination but not the source. + * + * Note that failing to snoop properly is unlikely to cause + * cache incoherency if the period size is larger than the + * size of L1 cache. This is because filling in one period will + * flush out the data for the previous period. So if you + * increased period_bytes_min to a large enough size, you might + * get more performance by not snooping, and you'll still be + * okay. + */ + if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK) { + link->source_addr = cpu_to_be32(temp_addr); + link->source_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP); + link->dest_addr = cpu_to_be32(ssi_sxx_phys); - else + link->dest_attr = cpu_to_be32(CCSR_DMA_ATR_NOSNOOP); + } else { link->source_addr = cpu_to_be32(ssi_sxx_phys); + link->source_attr = cpu_to_be32(CCSR_DMA_ATR_NOSNOOP); + + link->dest_addr = cpu_to_be32(temp_addr); + link->dest_attr = cpu_to_be32(CCSR_DMA_ATR_SNOOP); + } + + temp_addr += period_size; } return 0; @@ -808,7 +801,6 @@ static struct snd_pcm_ops fsl_dma_ops = { .ioctl = snd_pcm_lib_ioctl, .hw_params = fsl_dma_hw_params, .hw_free = fsl_dma_hw_free, - .prepare = fsl_dma_prepare, .pointer = fsl_dma_pointer, }; diff --git a/sound/soc/fsl/fsl_ssi.c b/sound/soc/fsl/fsl_ssi.c index c6d6eb7..6844009 100644 --- a/sound/soc/fsl/fsl_ssi.c +++ b/sound/soc/fsl/fsl_ssi.c @@ -400,7 +400,7 @@ static int fsl_ssi_startup(struct snd_pcm_substream *substream, } /** - * fsl_ssi_prepare: prepare the SSI. + * fsl_ssi_hw_params - program the sample size * * Most of the SSI registers have been programmed in the startup function, * but the word length must be programmed here. Unfortunately, programming @@ -412,20 +412,19 @@ static int fsl_ssi_startup(struct snd_pcm_substream *substream, * Note: The SxCCR.DC and SxCCR.PM bits are only used if the SSI is the * clock master. */ -static int fsl_ssi_prepare(struct snd_pcm_substream *substream, - struct snd_soc_dai *dai) +static int fsl_ssi_hw_params(struct snd_pcm_substream *substream, + struct snd_pcm_hw_params *hw_params, struct snd_soc_dai *cpu_dai) { - struct snd_pcm_runtime *runtime = substream->runtime; - struct snd_soc_pcm_runtime *rtd = substream->private_data; - struct fsl_ssi_private *ssi_private = rtd->dai->cpu_dai->private_data; - - struct ccsr_ssi __iomem *ssi = ssi_private->ssi; + struct fsl_ssi_private *ssi_private = cpu_dai->private_data; if (substream == ssi_private->first_stream) { + struct ccsr_ssi __iomem *ssi = ssi_private->ssi; + unsigned int sample_size = + snd_pcm_format_width(params_format(hw_params)); u32 wl; /* The SSI should always be disabled at this points (SSIEN=0) */ - wl = CCSR_SSI_SxCCR_WL(snd_pcm_format_width(runtime->format)); + wl = CCSR_SSI_SxCCR_WL(sample_size); /* In synchronous mode, the SSI uses STCCR for capture */ clrsetbits_be32(&ssi->stccr, CCSR_SSI_SxCCR_WL_MASK, wl); @@ -579,7 +578,7 @@ static struct snd_soc_dai fsl_ssi_dai_template = { }, .ops = { .startup = fsl_ssi_startup, - .prepare = fsl_ssi_prepare, + .hw_params = fsl_ssi_hw_params, .shutdown = fsl_ssi_shutdown, .trigger = fsl_ssi_trigger, .set_sysclk = fsl_ssi_set_sysclk, -- 1.8.2.3