[alsa-devel] Query regarding snd_pcm_lib_preallocate_pages_for_all()

[Takashi Iwai]

At Wed, 14 Nov 2007 17:43:21 +0530,
Pharaoh . wrote:
> 
> On Nov 14, 2007 1:44 PM, Takashi Iwai <tiwai at suse.de> wrote:
> 
> > At Wed, 14 Nov 2007 00:12:07 +0530,
> > Pharaoh . wrote:
> > >
> > > Hi,
> > >
> > > I am trying to use snd_pcm_lib_preallocate_pages_for_all() on an arm
> > based
> > > platform for preallocating the DMA buffer.
> > > I am using it as
> > >
> > > snd_pcm_lib_preallocate_pages_for_all(pcm,
> > >             SNDRV_DMA_TYPE_DEV,
> > >             NULL,
> > >             64 * 1024, 64 * 1024);
> > >
> > > since I dont know what should be passed as the 3'rd param. I tried
> > passing
> > > my device pointer i.e. card->dev after
> > > correctly populating it but I get an error as 'DMA coherent mask in not
> > set'
> > > and buffer is not allocated.
> >
> > For ARM, avoid the present buffer-preallocation and mmap functions in
> > ALSA core.  Instead you can allocate the buffers directly via normal
> > dma_alloc_coherent() in your driver.  And, mmap it via
> > dma_mmap_coherent() by defining the own mmap callback for PCM ops.
> >
> > See other arm/* drivers for rerecence.
> >
> >
> > Takashi
> >
> 
> Yes, I am using my own buffer preallocation scheme now. I allocate a 128K
> buffer using dma_alloc_coherent in probe and free it only in remove i.e. at
> rmmod time.
> 
> I dont think keeping a huge buffer like this till driver removal is a good
> idea but when I allocate/deallocate it in hw_params/hw_free respectively, it
> is causing fragmentation and sometimes the allocation fails. Do you think
> anything better can
> be done?

Unfortuantely, no other way, so far.  It's the only way that a driver
can do -- keeping the large chunk as long as it needs.


Takashi