[CML] ASoC: Both platform name/of_node are set #515

Activity

1. zhuyingjiang commented on Jan 9, 2019

   

   zhuyingjiang

   on Jan 9, 2019

   Author

   More actions

   log
   why_not_load_mod_2.txt
   why_not_load_mod.txt
2. plbossart commented on Jan 9, 2019

   

   plbossart

   on Jan 9, 2019 via email

   Member

   More actions

   On 1/9/19 1:19 AM, Zhu, Yingjiang wrote: When SOF start with load tplg, there will be a error with : " cml_rt5682 cml_rt5682: ASoC: Both platform name/of_node are set for SSP1-Codec",

   This is odd. I've seen this message before and it's been discussed this week on the alsa mailing list, look for "[alsa-devel] [PATCH] ASoC: core: Fix deferral of machine drivers"

   but after the kernel start, manually modprobe the module, the module can be load.

   Loading the module driver manually will not execute the probe and create a card, so that's not a solution.

   another way to make it auto load, is change the devm_kzalloc to kzalloc in snd_soc_init_platform.

   why would that help?
3. plbossart commented on Jan 9, 2019

   

   plbossart

   on Jan 9, 2019

   Last edited by plbossart

   Member

   More actions

   @zhuyingjiang when you provide logs, it's good if their names are self-explanatory. "why_not_load_mod.txt" and "why_not_load_mod_2.txt" don't tell me what you are trying to do, and how they relate to your bug description

   Also you may want to check why the codec is not registered in the first place:

   [ 3.366721] cml_rt5682 cml_rt5682: ASoC: CODEC DAI rt5682-aif1 not registered
   [ 3.381978] cml_rt5682 cml_rt5682: ASoC: CODEC DAI rt5682-aif1 not registered

   Looks to me like you didn't look at the root cause: the machine driver was selected based on the presence of an ACPI ID but the codec driver didn't probe successfully, so things won't work at this point.
4. bardliao commented on Jan 10, 2019

   

   bardliao

   on Jan 10, 2019

   Last edited by bardliao

   Collaborator

   More actions

   Thanks @plbossart for the information. I read ASoC: core: Fix deferral of machine drivers. I think that is exactly what we meet now. I applied the patch manually and it solved the issue.

   another way to make it auto load, is change the devm_kzalloc to
   kzalloc in snd_soc_init_platform.

   why would that help?

   The root cause is the "platform" is dynamic allocated by devm_kzalloc() and it will be freed automatically on error. And It will not be freed automatically If we use kzalloc(). But it will not be freed at all if we don't add a appropriate kfree() for it.
5. zhuyingjiang commented on Jan 10, 2019

   

   zhuyingjiang

   on Jan 10, 2019

   Author

   More actions

   @plbossart I am now report the issues first and I will go deep into the bugs, when the issues are reported.
   And OK, I will change the filename and the descriptions.
   Bard said he still find this problem before. @bardliao Could you provide some comments?
6. zhuyingjiang commented on Jan 11, 2019

   

   zhuyingjiang

   on Jan 11, 2019

   Author

   More actions

   @plbossart @bardliao Thanks, this fixed the issue, will wait when it merged , and verify again.
7. mengdonglin

   added

   bugSomething isn't workingSomething isn't working

   CMLApplies to Comet Lake platformApplies to Comet Lake platform

   on Jan 15, 2019
8. mengdonglin commented on Jan 15, 2019

   

   mengdonglin

   on Jan 15, 2019

   More actions

   @plbossart could you help to merge patch ASoC: core: Fix deferral of machine drivers to close this issue?
9. bardliao commented on Jan 15, 2019

   

   bardliao

   on Jan 15, 2019

   Collaborator

   More actions

   @mengdonglin It seems the patch has not been applied to upstream yet.
10. plbossart commented on Jan 15, 2019

    

    plbossart

    on Jan 15, 2019 via email

    Member

    More actions

    On 1/15/19 5:35 AM, Bard Liao wrote: @mengdonglin <https://github.com/mengdonglin> It seems the patch has not been applied to upstream yet.

    It's even worse, it creates issues with the way we use .platform_name in all machine drivers. At any rate I don't get why we have this issue only on CML so I would err on the side of a mistake in the CML support.
11. bardliao commented on Jan 16, 2019

    

    bardliao

    on Jan 16, 2019

    Collaborator

    More actions

    @plbossart I can see this issue on ICL. APL, and BYT. An easy way to reproduce it is to use an invalid .codec_dai_name on any dai link. So bind dai link will get error, and it is easy to see the issue when the soc core retry to bind dai link.
12. plbossart commented on Jan 16, 2019

    

    plbossart

    on Jan 16, 2019

    Member

    More actions

    @bardliao I have never seen this issue on APL/BYT so you will have to provide me with a configuration and log for me to believe it. My money is on a missing configuration of your machine driver. There are actually no usable logs for this issue, we'd need to progress on what the root cause before speculating further.
13. bardliao commented on Jan 16, 2019

    

    bardliao

    on Jan 16, 2019

    Collaborator

    More actions

    @plbossart attached the .config and dmesg. the specific config can reproduce the issue on Up2.
    Which is made by scripts/kconfig/merge_config.sh .config <PATH>/kconfig/base-defconfig <PATH>/kconfig/sof-defconfig <PATH>/kconfig/hdaudio-codecs-defconfig
    with the following change.

    # CONFIG_SND_HDA_INTEL is not set
    +CONFIG_DEBUG_KMEMLEAK=y
    +CONFIG_DEBUG_KMEMLEAK_EARLY_LOG_SIZE=4000
    

    However I can't find a working for CNL. I am checking if there is anything wrong on machine driver.

    config.txt
    dmesg.txt
14. bardliao commented on Jan 16, 2019

    

    bardliao

    on Jan 16, 2019

    Collaborator

    More actions

    @plbossart I think I get the reason why the issue is always seen with rt5682 codec. There are so many sleep in rt5682_i2c_probe(), and snd_soc_register_component() will not called before dai link binding. So dai link binding won't success in the first try, and we will meet the issue in the second try. We don't see this issue on other platforms since we can bind dai link successfully at the first try and the issue can only be seen after second try of dai link binding.
15. zhuyingjiang commented on Jan 16, 2019

    

    zhuyingjiang

    on Jan 16, 2019

    Author

    More actions

    There are so many sleep in rt5682_i2c_probe(), and snd_soc_register_component() will not called before dai link binding.

    @bardliao So this is a issue/bug of rt5682?
16. bardliao commented on Jan 16, 2019

    

    bardliao

    on Jan 16, 2019

    Collaborator

    More actions

    @zhuyingjiang No the sleep are necessary for rt5682. ASoC allows retry on dal link binding. The issue is as ASoC: core: Fix deferral of machine drivers says the "platform" is dynamic allocated by devm_kzalloc() and it will be freed automatically on error. So now we will see the "ASoC: Both platform name/of_node are set" error after second try in dai link binding.
17. zhuyingjiang commented on Jan 16, 2019

    

    zhuyingjiang

    on Jan 16, 2019

    Author

    More actions

    @plbossart Yes, I have saw a illegal name of platform, why the name is set should be root cause
    ====plat name: ;ݱ\xffffffff\xffffffff\xffffffff\xffffffff0�2\xffffffb2\xffffffff\xffffffff\xffffffff\xffffffff====of_node (null)===
18. bardliao

    closed this as completedon Feb 1, 2019
19. plbossart

    added a commit that references this issue on Apr 17, 2020

    kbuild: link lib-y objects to vmlinux forcibly when CONFIG_MODULES=y

    ...

    7273ad2
20. naveen-manohar

    added a commit that references this issue on May 14, 2020

    UPSTREAM: kbuild: link lib-y objects to vmlinux forcibly when CONFIG_…

    ...

    3972eda
21. ujfalusi

    added a commit that references this issue on Apr 30, 2026

    smb: smbdirect: introduce smbdirect_public.h with prototypes

    ...

    b2261ce