- spin_lock_irq(&chip->substream_lock);
- /* Takashi Iwai says regarding this extra lock:
-
- If the irq handler handles some data on the DMA buffer, it should
- do snd_pcm_stream_lock().
- That protects basically against all races among PCM callbacks, yes.
- However, there are two remaining issues:
- 1. The substream pointer you try to lock isn't protected _before_
- this lock yet.
- 2. snd_pcm_period_elapsed() itself acquires the lock.
- The requirement of another lock is because of 1. When you get
- chip->playback_substream, it's not protected.
- Keeping this lock while snd_pcm_period_elapsed() assures the substream
- is still protected (at least, not released). And the other status is
- handled properly inside snd_pcm_stream_lock() in
- snd_pcm_period_elapsed().
-
- */
- if (!chip->playback_substream)
- goto exit_nr_unlock1;
- substream = chip->playback_substream;
- snd_pcm_stream_lock(substream);