[alsa-devel] ASoC: atmel-pcm-dma: Leaking memory in atmel_pcm_hw_params

Lars-Peter Clausen lars at metafoo.de
Mon Mar 18 16:36:02 CET 2013


On 03/18/2013 04:59 AM, Bo Shen wrote:
> Hi Lars,
> 
> On 3/17/2013 22:39, Lars-Peter Clausen wrote:
>> Hi,
>>
>> The dmaengine based pcm driver for atmel calls snd_dmaengine_pcm_open from
>> it's hw_params callback. There is nothing preventing an application calling
>> hw_params more than once. snd_dmaengine_pcm_open allocates a new prtd struct
>> each time it gets called. So in case hw_params is called multiple times we
>> leak memory here. 
> 
> I use alsa utils and don't meet this issue. Please point out which software
> you use meet this issue.

I found the issue by looking at the code, but I think any application using
the oss compatibility layer of ALSA will trigger this.

> 
>> Is there any specific reason why snd_dmaengine_pcm_open() needs to be called
>> from the hw_params callback and why it can't be called from the open
>> callback?
> 
> This is because dma data setting and usage.
> 
> The filter function depends on the data returned by
>> snd_soc_dai_get_dma_data(), but as far as I can see the DAI driver calls
>> snd_soc_dai_set_dma_data() from its startup() callback, so the data is
>> available in the PCM drivers open callback.
> 
> Have you test this yet?

I don't have the hardware. I just looked at the code again, and it looks
like I must have misread it the first time. snd_soc_dai_set_dma_data() is
called from the DAI drivers hwparams callback. But the data that gets passed
to snd_soc_dai_set_dma_data() doesn't really depend on anything that's only
available in the hwparams callback. So the snd_soc_dai_set_dma_data() could
easily be move to the DAI driver startup() callback.

- Lars


More information about the Alsa-devel mailing list