[alsa-devel] ALSA Control Questions (atomicity, error handling)

Tobias Schneider tobsnyder at gmx.de
Fri Dec 4 10:31:06 CET 2009


Thanks for the reply.

Actually I know where the problem in my code is: inside the put callback 
(see below) I am calling a function (dsp_setoutput) that is sending the 
signal to the DSP, there I wanted to wait for the reply of the DSP to 
get a feedback whether the request was successful or not. This was done 
by a semaphore (going down and waiting for IRQ from DSP that is setting 
sema up), and thats the point where the posted bug occured.

So I was just wondering that it seems that there is no possibility to 
wait for a reply of the DSP, to be sure that a given value has been set?


(for the interested ones, here is the code..)

static int snd_mychip_output_set_put(struct snd_kcontrol *kcontrol,
                                     struct snd_ctl_elem_value *ucontrol)
{
  struct snd_mychip *mychip = snd_kcontrol_chip(kcontrol);
  int addr = kcontrol->private_value;
  short change = 0; 

  spin_lock_irq(&mychip->mixer_lock);
  if (ucontrol->value.integer.value[0] != mychip->output_set[addr]) {
    if (dsp_setoutput(addr+1, ucontrol->value.integer.value[0]) >= 0)
    {
      mychip->output_set[addr] = ucontrol->value.integer.value[0];
      change = 1;
    }
    else  // error sending signal
    {
      change = -EBUSY;  // device or resource busy
    }
  }
  spin_unlock_irq(&mychip->mixer_lock);
  return change;
}

later on in the called function...
 // wait for reply via semaphore
  while(!end)
  {
    if (down_interruptible(&dsp_reply_sema)==-EINTR)   // HERE WE GET 
BUG: scheduling while atomic: amixer/0x00000001/430 ...
      // waked by signal (terminate), leave thread
      printk(INFO "reply: catched signal\n");
      end=1;
      break;
    }   
    printk(INFO "dsp_reply signaled (saved status = 
%i)\n",stDspReply.status);
    end=1;
 ...
}



Another point: I figured out that the "ALSA middlelayer" seems to filter 
the given values of a control. So if the user sets value=1000 where 
maximum is 10, I will get 10 instead of 1000 in the put callback...so 
it's not necessary to check values in those callbacks, but unfortunately 
that's not always good - because sometimes it doesn't make sense to use 
the max or min values if the user just entered a wrong number...

thanks so far


More information about the Alsa-devel mailing list