Dne 13. 01. 22 v 16:08 Jaroslav Kysela napsal(a):
On 13. 01. 22 14:45, Takashi Iwai wrote:
Also ideally, there may be a check in hw_params, if parameters (buffers) are changed, but the implementation is not so easy. Maybe we can allow OPEN -> PREPARE transition for this case, so the applications may just restart the streaming in the most light way.
Hmm. Reading more about those restrictions and requirements, I feel that this might be better implemented in the gadget driver side locally at first. Basically we can handle similarly: add a new local flag, set it at the stream stop, and return an error at prepare until hw_params gets reconfigured. This might be even smaller changes?
Pavel reported that stop to SETUP is not enough for sox, but it's true that the driver may fail in the prepare() callback until the standard stream operation is not recovered. I think that sox is trying to recover and it succeeds - then the I/O timeout occurs.
Ref: https://lore.kernel.org/alsa-devel/9635d70f-dc12-f9ed-29f5-ce34a1d4b112@ivit...
The gadget driver (drivers/usb/gadget/function/u_audio.c) has empty prepare callback at the moment.
Pavel, could you try to add the no-stream flag management to the gadget driver and return an error in the prepare callback in this case?
My apology for the delay. I can make the changes to the gadget to be prepared for the alsa change. Let me recap to see if I understand your plan correctly:
* Currently the stopped stream is indicated by unsetting prm->active https://kernel.googlesource.com/pub/scm/linux/kernel/git/gregkh/usb/+/554237... .
* A hw_params_set flag indicating that hw params have been configured should be set at https://kernel.googlesource.com/pub/scm/linux/kernel/git/gregkh/usb/+/554237... . Where should it be unset? At snd_pcm_ops.stop (currently empty), at https://kernel.googlesource.com/pub/scm/linux/kernel/git/gregkh/usb/+/554237... for active = false, or somewhere else?
* A check should be added to snd_pcm_ops.prepare which fails if the hw_params_set flag is unset. What error type should the fail return?
Thanks a lot for help.
Best regards,
Pavel.