On 11/16/22 10:33, AngeloGioacchino Del Regno wrote:
> Il 16/11/22 16:04, Pierre-Louis Bossart ha scritto:
>>
>>>> diff --git a/include/sound/sof/dai.h b/include/sound/sof/dai.h
>>>> index 21d98f31a9ca..83fd81c82e4c 100644
>>>> --- a/include/sound/sof/dai.h
>>>> +++ b/include/sound/sof/dai.h
>>>> @@ -84,6 +84,7 @@ enum sof_ipc_dai_type {
>>>> SOF_DAI_AMD_BT, /**< AMD ACP BT*/
>>>> SOF_DAI_AMD_SP, /**< AMD ACP SP */
>>>> SOF_DAI_AMD_DMIC, /**< AMD ACP DMIC */
>>>> + SOF_DAI_AMD_HS, /**< Amd HS */
>>>> SOF_DAI_MEDIATEK_AFE, /**< Mediatek AFE */
>>>
>>> Adding SOF_DAI_AMD_HS before SOF_DAI_MEDIATEK_AFE desynced this
>>> enumeration
>>> so the DAI type is now 11 and not 10 anymore, leading to a failure in
>>> firmware
>>> at IPC3 helper function `dai_get()`, as when `dai_find_type()` is
>>> called, the
>>> DAI type that the firmware expects doesn't match with the one that gets
>>> sent
>>> in the request message from the kernel.
>>>
>>> As a local test, I tried moving SOF_DAI_AMD_HS after
>>> SOF_DAI_MEDIATEK_AFE and
>>> this has restored full functionality on my MT8195 platform (Tomato
>>> Chromebook).
>>>
>>> If SOF is supposed to guarantee backwards compatibility (and I believe
>>> it is),
>>> this commit breaks that.
>>>
>>> I would be tempted to send a commit that moves SOF_DAI_AMD_HS to the
>>> end, but
>>> that would break the already compiled firmware for AMD platforms, so I
>>> am not
>>> sure how to proceed.
>>
>> D'oh. Yes this breaks backwards-compatibility and this is a clear
>> mistake. I think your suggestion to add the AMD_HS at the end is the
>> only practical solution indeed - this would need to be done for both
>> kernel and SOF version of dai.h.
>>
>
> Okay, I will send a commit tomorrow :-)
I sent those two GitHub pull requests already:
https://github.com/thesofproject/linux/pull/4017
https://github.com/thesofproject/sof/pull/6616
