Re: [PATCH 1/3] ASoC: codecs: tas675x: use READ_ONCE for params to be used concurrently

[Wang, Sen]

On 7/1/2026 1:55 PM, Mark Brown wrote:
> On Tue, Jun 30, 2026 at 01:31:20PM -0500, Sen Wang wrote:
> > active_playback_dais and active_capture_dais are written atomically via
> > set_bit()/clear_bit() and can be read concurrently from the
> > fault_check_work delayed work handler.
> >
> > fault_check_work already uses READ_ONCE; extend the same guard to all other
> > reads in tas675x_hw_params() and tas675x_mute_stream().
>
> I'm not convinced that any of this READ_ONCE stuff is safe, that's a
> very unusual pattern for an I2C device like these, but given that I
> missed it earlier I'll take the patch for now.

You're right mark, READ_ONCE only makes the variable volatile. I had a 
false assumption that it also adds memory barrier as well. This patch 
should be dropped as it provides no inherent benefit.

Tracing back, the reasons for needing to keeping a separate record of 
active DAIs was the unreliable active count when calling 
snd_soc_component_active in mute_stream when I first developed the 
driver, I should look deeper into that instead...