Re: [alsa-devel] [PATCH] ALSA: snd-aloop: make preallocated buffer size configurable
At Wed, 21 Sep 2011 10:02:27 -0500, Pierre-Louis Bossart wrote:
I meant to have a kconfig option to limit the upper-bound of pre-alloc buffer sizes for all drivers. That is, a patch like below (untested)
That's a good idea, but the two patches aren't necessarily exclusive. In an embedded solution you may want a 2s PCM buffer for low-power playback and a 64k buffer for snd-aloop as it is only used with small periods. Also one would need to check the buffer/period sizes in each driver, and make sure they are smaller than this global max.
Well, if it's only about snd-aloop, another option would be to use vmalloc'ed buffer for aloop. Since it's no hardware buffer, it's more system-friendly, and you don't need a pre-allocation. Again, an untested patch is below.
thanks,
Takashi
--- diff --git a/sound/drivers/aloop.c b/sound/drivers/aloop.c index a0da775..4067f15 100644 --- a/sound/drivers/aloop.c +++ b/sound/drivers/aloop.c @@ -575,7 +575,8 @@ static void loopback_runtime_free(struct snd_pcm_runtime *runtime) static int loopback_hw_params(struct snd_pcm_substream *substream, struct snd_pcm_hw_params *params) { - return snd_pcm_lib_malloc_pages(substream, params_buffer_bytes(params)); + return snd_pcm_lib_alloc_vmalloc_buffer(substream, + params_buffer_bytes(params)); }
static int loopback_hw_free(struct snd_pcm_substream *substream) @@ -587,7 +588,7 @@ static int loopback_hw_free(struct snd_pcm_substream *substream) mutex_lock(&dpcm->loopback->cable_lock); cable->valid &= ~(1 << substream->stream); mutex_unlock(&dpcm->loopback->cable_lock); - return snd_pcm_lib_free_pages(substream); + return snd_pcm_lib_free_vmalloc_buffer(substream); }
static unsigned int get_cable_index(struct snd_pcm_substream *substream) @@ -740,6 +741,8 @@ static struct snd_pcm_ops loopback_playback_ops = { .prepare = loopback_prepare, .trigger = loopback_trigger, .pointer = loopback_pointer, + .page = snd_pcm_lib_get_vmalloc_page, + .mmap = snd_pcm_lib_mmap_vmalloc, };
static struct snd_pcm_ops loopback_capture_ops = { @@ -751,6 +754,8 @@ static struct snd_pcm_ops loopback_capture_ops = { .prepare = loopback_prepare, .trigger = loopback_trigger, .pointer = loopback_pointer, + .page = snd_pcm_lib_get_vmalloc_page, + .mmap = snd_pcm_lib_mmap_vmalloc, };
static int __devinit loopback_pcm_new(struct loopback *loopback, @@ -771,10 +776,6 @@ static int __devinit loopback_pcm_new(struct loopback *loopback, strcpy(pcm->name, "Loopback PCM");
loopback->pcm[device] = pcm; - - snd_pcm_lib_preallocate_pages_for_all(pcm, SNDRV_DMA_TYPE_CONTINUOUS, - snd_dma_continuous_data(GFP_KERNEL), - 0, 2 * 1024 * 1024); return 0; }
Well, if it's only about snd-aloop, another option would be to use vmalloc'ed buffer for aloop. Since it's no hardware buffer, it's more system-friendly, and you don't need a pre-allocation. Again, an untested patch is below.
I wasn't sure why physically-contiguous memory would be required, just reduced it, but it makes sense to remove it altogether.
I used a double loopback, one in alsa and one in pulseaudio. Works fine. $ pactl load-module module-loopback source=alsa_input.1.analog-stereo sink=alsa_output.pci-0000_00_1b.0.analog-stereo $ aplay -Dhw:1,1 viol-1mn.wav
For some reason I couldn't apply the patch as is, here's what I used Thanks! -Pierre
At Fri, 23 Sep 2011 16:03:31 -0500, Pierre-Louis Bossart wrote:
Well, if it's only about snd-aloop, another option would be to use vmalloc'ed buffer for aloop. Since it's no hardware buffer, it's more system-friendly, and you don't need a pre-allocation. Again, an untested patch is below.
I wasn't sure why physically-contiguous memory would be required, just reduced it, but it makes sense to remove it altogether.
I used a double loopback, one in alsa and one in pulseaudio. Works fine. $ pactl load-module module-loopback source=alsa_input.1.analog-stereo sink=alsa_output.pci-0000_00_1b.0.analog-stereo $ aplay -Dhw:1,1 viol-1mn.wav
For some reason I couldn't apply the patch as is, here's what I used Thanks!
OK, I applied the patch now to sound git tree.
thanks,
Takashi
participants (2)
-
Pierre-Louis Bossart
-
Takashi Iwai