The recent upstream changes break the way SOF used to override the BE information to reuse machine drivers.
Test code: https://github.com/plbossart/sound/tree/fix/soc-core-probe-defer
With the upstream changes, the tests on the dailink platform names are done before we even have a chance to override them. Moving the call of soc_check_tplg_fes() from soc_instantiate_card() to soc_register_card() partly solves the problem. This can be shown with the PLB_FIX1 define. It is not clear if this solution is acceptable since there soc_register_card() is very minimal and plays with mutexes, we may need to do something less invasive
There is however a second problem where we seem to have a memory leak in snd_soc_init_platform(), the use of the legacy_plaform field makes no sense. this can be shown with the PLB_FIX2 define.
This can be tested with the Up2 pcm512 machine driver.
dmesg_fail.log
dmesg_plbfix1.log
dmesg_plbfix2.log
The recent upstream changes break the way SOF used to override the BE information to reuse machine drivers.
Test code: https://github.com/plbossart/sound/tree/fix/soc-core-probe-defer
With the upstream changes, the tests on the dailink platform names are done before we even have a chance to override them. Moving the call of soc_check_tplg_fes() from soc_instantiate_card() to soc_register_card() partly solves the problem. This can be shown with the PLB_FIX1 define. It is not clear if this solution is acceptable since there soc_register_card() is very minimal and plays with mutexes, we may need to do something less invasive
There is however a second problem where we seem to have a memory leak in snd_soc_init_platform(), the use of the legacy_plaform field makes no sense. this can be shown with the PLB_FIX2 define.
This can be tested with the Up2 pcm512 machine driver.
dmesg_fail.log
dmesg_plbfix1.log
dmesg_plbfix2.log