[alsa-devel] [PATCH] ALSA: ASoC V2: optimize init sequence of Freescale MPC8610 sound drivers
Timur Tabi
timur at freescale.com
Fri Sep 5 22:51:28 CEST 2008
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.
Some comment blocks were not near the code they reference, so they were moved.
Signed-off-by: Timur Tabi <timur at freescale.com>
---
This patch is for ASoC V2 only.
sound/soc/fsl/fsl_dma.c | 283 ++++++++++++++++++++++-------------------------
sound/soc/fsl/fsl_ssi.c | 20 ++--
2 files changed, 141 insertions(+), 162 deletions(-)
diff --git a/sound/soc/fsl/fsl_dma.c b/sound/soc/fsl/fsl_dma.c
index 5da0069..8821289 100644
--- a/sound/soc/fsl/fsl_dma.c
+++ b/sound/soc/fsl/fsl_dma.c
@@ -319,9 +319,66 @@ error:
}
/**
- * fsl_dma_open: open a new substream.
+ * fsl_dma_open: open a new substream and allocate DMA resources for it
*
* Each substream has its own DMA buffer.
+ *
+ * ALSA divides the DMA buffer into N periods. We create NUM_DMA_LINKS link
+ * descriptors that cycle from one period to the next. For example, if
+ * there are six periods and two link descriptors, this is how they look
+ * before playback starts:
+ *
+ * The last link descriptor
+ * ____________ points back to the first
+ * | |
+ * V |
+ * ___ ___ |
+ * | |->| |->|
+ * |___| |___|
+ * | |
+ * | |
+ * V V
+ * _________________________________________
+ * | | | | | | | The DMA buffer is
+ * | | | | | | | divided into 6 parts
+ * |______|______|______|______|______|______|
+ *
+ * and here's how they look after the first period is finished playing:
+ *
+ * ____________
+ * | |
+ * V |
+ * ___ ___ |
+ * | |->| |->|
+ * |___| |___|
+ * | |
+ * |______________
+ * | |
+ * V V
+ * _________________________________________
+ * | | | | | | |
+ * | | | | | | |
+ * |______|______|______|______|______|______|
+ *
+ * The first link descriptor now points to the third period. The DMA
+ * controller is currently playing the second period. When it finishes, it
+ * will jump back to the first descriptor and play the third period.
+ *
+ * There are four reasons we do this:
+ *
+ * 1. The only way to get the DMA controller to automatically restart the
+ * transfer when it gets to the end of the buffer is to use chaining
+ * mode. Basic direct mode doesn't offer that feature.
+ * 2. We need to receive an interrupt at the end of every period. The DMA
+ * controller can generate an interrupt at the end of every link transfer
+ * (aka segment). Making each period into a DMA segment will give us the
+ * interrupts we need.
+ * 3. By creating only two link descriptors, regardless of the number of
+ * periods, we do not need to reallocate the link descriptors if the
+ * number of periods changes.
+ * 4. All of the audio data is still stored in a single, contiguous DMA
+ * buffer, which is what ALSA expects. We're just dividing it into
+ * contiguous parts, and creating a link descriptor for each one.
*/
static int fsl_dma_open(struct snd_pcm_substream *substream)
{
@@ -370,8 +427,8 @@ static int fsl_dma_open(struct snd_pcm_substream *substream)
dma_private->ld_buf_phys = ld_buf_phys;
dma_private->dma_buf_phys = substream->dma_buffer.addr;
- ret = request_irq(dma_private->dma_info->irq,
- fsl_dma_isr, 0, "DMA", dma_private);
+ ret = request_irq(dma_private->dma_info->irq, fsl_dma_isr, 0,
+ "fsl-audio-dma", dma_private);
if (ret) {
dev_err(substream->pcm->card->dev,
"can't register ISR for IRQ %u (ret=%i)\n",
@@ -393,17 +450,14 @@ static int fsl_dma_open(struct snd_pcm_substream *substream)
temp_link = dma_private->ld_buf_phys +
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);
+ for (i = 0; i < NUM_DMA_LINKS - 1; i++) {
+ dma_private->link[i].next = cpu_to_be64(temp_link);
temp_link += sizeof(struct fsl_dma_link_descriptor);
}
/* The last link descriptor points to the first */
- dma_private->link[i - 1].next = cpu_to_be64(dma_private->ld_buf_phys);
+ dma_private->link[NUM_DMA_LINKS - 1].next =
+ cpu_to_be64(dma_private->ld_buf_phys);
/* Tell the DMA controller where the first link descriptor is */
out_be32(&dma_channel->clndar,
@@ -417,11 +471,7 @@ static int fsl_dma_open(struct snd_pcm_substream *substream)
/*
* Program the mode register for interrupts, external master control,
* and source/destination hold. Also clear the Channel Abort bit.
- */
- mr = in_be32(&dma_channel->mr) &
- ~(CCSR_DMA_MR_CA | CCSR_DMA_MR_DAHE | CCSR_DMA_MR_SAHE);
-
- /*
+ *
* We want External Master Start and External Master Pause enabled,
* because the SSI is controlling the DMA controller. We want the DMA
* controller to be set up in advance, and then we signal only the SSI
@@ -436,7 +486,7 @@ static int fsl_dma_open(struct snd_pcm_substream *substream)
* We want Error Interrupt enabled, so that we can get an error if
* the DMA controller is mis-programmed somehow.
*/
- mr |= CCSR_DMA_MR_EOSIE | CCSR_DMA_MR_EIE | CCSR_DMA_MR_EMP_EN |
+ mr = CCSR_DMA_MR_EOSIE | CCSR_DMA_MR_EIE | CCSR_DMA_MR_EMP_EN |
CCSR_DMA_MR_EMS_EN;
/* For playback, we want the destination address to be held. For
@@ -450,78 +500,47 @@ static int fsl_dma_open(struct snd_pcm_substream *substream)
}
/**
- * fsl_dma_hw_params: allocate the DMA buffer and the DMA link descriptors.
- *
- * ALSA divides the DMA buffer into N periods. We create NUM_DMA_LINKS link
- * descriptors that ping-pong from one period to the next. For example, if
- * there are six periods and two link descriptors, this is how they look
- * before playback starts:
- *
- * The last link descriptor
- * ____________ points back to the first
- * | |
- * V |
- * ___ ___ |
- * | |->| |->|
- * |___| |___|
- * | |
- * | |
- * V V
- * _________________________________________
- * | | | | | | | The DMA buffer is
- * | | | | | | | divided into 6 parts
- * |______|______|______|______|______|______|
+ * fsl_dma_hw_params: program the audio paramters into the DMA controller
*
- * and here's how they look after the first period is finished playing:
+ * This function takes the audio parameters (period size, sample size, etc)
+ * and finishes programming the DMA controller accordingly.
*
- * ____________
- * | |
- * V |
- * ___ ___ |
- * | |->| |->|
- * |___| |___|
- * | |
- * |______________
- * | |
- * V V
- * _________________________________________
- * | | | | | | |
- * | | | | | | |
- * |______|______|______|______|______|______|
+ * The actual address in STX0 (destination for playback, source for capture)
+ * is based on the sample size.
*
- * The first link descriptor now points to the third period. The DMA
- * controller is currently playing the second period. When it finishes, it
- * will jump back to the first descriptor and play the third period.
+ * 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.
*
- * There are four reasons we do this:
+ * 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
+ * integer, then it should be copied to P+2. If X is an 8-bit register,
+ * then it should be copied to P+3.
*
- * 1. The only way to get the DMA controller to automatically restart the
- * transfer when it gets to the end of the buffer is to use chaining
- * mode. Basic direct mode doesn't offer that feature.
- * 2. We need to receive an interrupt at the end of every period. The DMA
- * controller can generate an interrupt at the end of every link transfer
- * (aka segment). Making each period into a DMA segment will give us the
- * interrupts we need.
- * 3. By creating only two link descriptors, regardless of the number of
- * periods, we do not need to reallocate the link descriptors if the
- * number of periods changes.
- * 4. All of the audio data is still stored in a single, contiguous DMA
- * buffer, which is what ALSA expects. We're just dividing it into
- * contiguous parts, and creating a link descriptor for each one.
+ * So for playback of 8-bit samples, the DMA controller must transfer single
+ * bytes from the DMA buffer to the last byte of the STX0 register, i.e.
+ * offset by 3 bytes. For 16-bit samples, the offset is two bytes.
*
- * 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()
+ * For 24-bit samples, the offset is 1 byte. However, the DMA controller
+ * does not support 3-byte copies (the DAHTS register supports only 1, 2, 4,
+ * 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_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;
+ struct ccsr_dma_channel __iomem *dma_channel =
+ dma_private->dma_info->channel;
+
+ dma_addr_t ssi_sxx_phys; /* Bus address of SSI STX register */
dma_addr_t temp_addr; /* Pointer to next period */
+ uint32_t mr;
+ unsigned int sample_size; /* Number of bits per sample */
+ unsigned int frame_size; /* Number of bytes per frame */
unsigned int i;
/* Get all the parameters we need */
@@ -537,83 +556,16 @@ static int fsl_dma_hw_params(struct snd_pcm_substream *substream,
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.
- *
- * 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
- * integer, then it should be copied to P+2. If X is an 8-bit register,
- * then it should be copied to P+3.
- *
- * So for playback of 8-bit samples, the DMA controller must transfer single
- * bytes from the DMA buffer to the last byte of the STX0 register, i.e.
- * offset by 3 bytes. For 16-bit samples, the offset is two bytes.
- *
- * For 24-bit samples, the offset is 1 byte. However, the DMA controller
- * does not support 3-byte copies (the DAHTS register supports only 1, 2, 4,
- * 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)
-{
- struct snd_pcm_runtime *runtime = substream->runtime;
- struct fsl_dma_private *dma_private = runtime->private_data;
- struct ccsr_dma_channel __iomem *dma_channel =
- dma_private->dma_info->channel;
- uint32_t mr;
- unsigned int i;
- dma_addr_t ssi_sxx_phys; /* Bus address of SSI STX register */
- unsigned int frame_size; /* Number of bytes per frame */
+ /* Program the sample size and I/O address into the DMA controller */
ssi_sxx_phys = dma_private->dma_info->ssi_sxx_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) {
+ sample_size = snd_pcm_format_physical_width(params_format(hw_params));
+
+ switch (sample_size) {
case 8:
mr |= CCSR_DMA_MR_DAHTS_1 | CCSR_DMA_MR_SAHTS_1;
ssi_sxx_phys += 3;
@@ -631,7 +583,8 @@ static int fsl_dma_prepare(struct snd_pcm_substream *substream)
return -EINVAL;
}
- frame_size = runtime->frame_bits / 8;
+ frame_size = 2 * (sample_size / 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
@@ -648,16 +601,45 @@ 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.
- */
+ temp_addr = substream->dma_buffer.addr;
+
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 this is unlikely to cause cache incoherency unless
+ * the DMA buffer is smaller than the size of L1 cache.
+ */
+ 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;
@@ -800,7 +782,6 @@ 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 319cc97..fc52d30 100644
--- a/sound/soc/fsl/fsl_ssi.c
+++ b/sound/soc/fsl/fsl_ssi.c
@@ -344,30 +344,28 @@ 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
- * the SxCCR.WL bits requires the SSI to be temporarily disabled. This can
- * cause a problem with supporting simultaneous playback and capture. If
- * the SSI is already playing a stream, then that stream may be temporarily
- * stopped when you start capture.
+ * but the word length must be programmed here. Since we support
+ * synchronous mode only, only the master stream can program STCRR[WL].
*
* 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 *cpu_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 fsl_ssi_info *ssi_info = cpu_dai->private_data;
if (substream == ssi_info->master_stream) {
- struct snd_pcm_runtime *runtime = substream->runtime;
struct ccsr_ssi __iomem *ssi = ssi_info->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);
@@ -588,7 +586,7 @@ static struct snd_soc_dai_caps capture = {
static struct snd_soc_dai_ops ops = {
.startup = fsl_ssi_startup,
- .prepare = fsl_ssi_prepare,
+ .hw_params = fsl_ssi_hw_params,
.shutdown = fsl_ssi_shutdown,
.trigger = fsl_ssi_trigger,
--
1.5.5
More information about the Alsa-devel
mailing list