[PATCH] ASoC: da7219: Fix general protection fault in da7219_register_dai_clks

[Adam Thomson]

On 30 July 2020 15:04, Yong Zhi wrote:

> clkdev_drop(cl) does not null the removed cl, if da7219_register_dai_clks()
> entered again after card removal, devm_clk_register() will return -EEXIST,
> the goto err to clkdev_drop() will trigger board reboot.
> 
> Test commands:
>    modprobe -r snd_sof_pci
>    modprobe snd_sof_pci
> 
> The oops looks like:
> 
> da7219 i2c-DLGS7219:00: Using default DAI clk names: da7219-dai-wclk, da7219-
> dai-bclk
> da7219 i2c-DLGS7219:00: Failed to register da7219-dai-wclk: -17
> general protection fault: 0000 [#1] PREEMPT SMP NOPTI
> RIP: 0010:clkdev_drop+0x20/0x52
> Call Trace:
>  da7219_probe+0x52e/0x6dc [snd_soc_da7219]
>  soc_probe_component+0x206/0x3a1
>  snd_soc_bind_card+0x4ee/0x9a6
>  devm_snd_soc_register_card+0x48/0x7b
>  audio_probe+0x1f0/0x221 [snd_soc_sof_da7219_max98373]
>  platform_drv_probe+0x89/0xa2
>  really_probe+0x129/0x30d
>  driver_probe_device+0x59/0xec
>  ? driver_sysfs_remove+0x55/0x55
>  bus_for_each_drv+0xa1/0xdc
>  __device_attach+0xc2/0x146
>  bus_probe_device+0x32/0x97
>  device_add+0x311/0x3b4
>  platform_device_add+0x184/0x1eb
> 
> Fix by marking (nullifying) the da7219->dai_clks_lookup[i]
> after clkdev_drop().
> 
> Signed-off-by: Yong Zhi <yong.zhi at intel.com>
> ---
>  sound/soc/codecs/da7219.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/sound/soc/codecs/da7219.c b/sound/soc/codecs/da7219.c
> index 153ea30b5a8f..54da7cfbb5f4 100644
> --- a/sound/soc/codecs/da7219.c
> +++ b/sound/soc/codecs/da7219.c
> @@ -2369,8 +2369,10 @@ static void da7219_remove(struct
> snd_soc_component *component)
> 
>  #ifdef CONFIG_COMMON_CLK
>  	for (i = DA7219_DAI_NUM_CLKS - 1; i >= 0; --i) {
> -		if (da7219->dai_clks_lookup[i])
> +		if (da7219->dai_clks_lookup[i]) {
>  			clkdev_drop(da7219->dai_clks_lookup[i]);
> +			da7219->dai_clks_lookup[i] = NULL;
> +		}

It seems to me that devm_* functions should have freed up everything when the
codec module was removed. I can only assume the codec isn't being removed in
your test hence devm is never freeing the clock resource and is why you're
getting -EEXIST. Is this the case and is your use-case expected behaviour? It's
not something that has been reported previously so am keen to understand exactly
what's happening here.

>  	}
>  #endif
> 
> --
> 2.7.4