On 12/12/18 9:43 AM, Takashi Iwai wrote:
On Wed, 12 Dec 2018 16:29:23 +0100, Pierre-Louis Bossart wrote:
- snd_sof_pcm_platform_trigger(sdev, substream, cmd);
- /* send IPC to the DSP */
- ret = sof_ipc_tx_message(sdev->ipc, stream.hdr.cmd, &stream,
sizeof(stream), &reply, sizeof(reply));- return ret;
... so the whole trigger action is non-atomic PCM only?
Not sure if I fully understand your point here. The trigger does indeed need an IPC to proceed, but the front-ends are marked as such with the .nonatomic field set to true. Not sure how different this is from existing atom/sst or Skylake drivers.
Hm, so this relies purely on FE, and it seems that we have no check whether FE and BE match wrt nonatomic PCM ops. It has to be addressed, but it's a different topic.
At most, you can put some notes in the code that all these are non-atomic ops. This makes it clearer, for example, that spinlock with irqsave is often redundant.
Yes, noted, we'll check the spinlock use and add comments as need when operations are non-atomic. Thanks for the suggestions, much appreciated.