Re: [PATCH] usb: typec: ucsi: gaokun: unwind notifier on UCSI register failure

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

On Mon, Jun 15, 2026 at 02:53:57PM +0800, Pengpeng Hou wrote:
> gaokun_ucsi_register_worker() registers the EC notifier before
> registering the UCSI device.
> 
> If ucsi_register() fails, the worker only reports the error and leaves
> the notifier registered.  Later EC events can then call back into an
> unregistered UCSI instance.  The remove path also unconditionally
> unregisters both the notifier and UCSI device even when the delayed
> worker failed part way through registration.
> 
> Track which publication steps succeeded, unregister the EC notifier when
> UCSI registration fails, and make remove only undo the steps that were
> actually published.
> 
> Signed-off-by: Pengpeng Hou <pengpeng@iscas.ac.cn>
> ---
>  drivers/usb/typec/ucsi/ucsi_huawei_gaokun.c | 16 +++++++++++++---
>  1 file changed, 13 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/usb/typec/ucsi/ucsi_huawei_gaokun.c b/drivers/usb/typec/ucsi/ucsi_huawei_gaokun.c
> index ca749fde49bd..29d2e76b2fe9 100644
> --- a/drivers/usb/typec/ucsi/ucsi_huawei_gaokun.c
> +++ b/drivers/usb/typec/ucsi/ucsi_huawei_gaokun.c
> @@ -98,6 +98,8 @@ struct gaokun_ucsi {
>  	struct device *dev;
>  	struct delayed_work work;
>  	struct notifier_block nb;
> +	bool notifier_registered;
> +	bool ucsi_registered;
>  	u16 version;
>  	u8 num_ports;
>  };
> @@ -457,10 +459,16 @@ static void gaokun_ucsi_register_worker(struct work_struct *work)
>  		dev_err_probe(ucsi->dev, ret, "notifier register failed\n");
>  		return;
>  	}
> +	uec->notifier_registered = true;
>  
>  	ret = ucsi_register(ucsi);
> -	if (ret)
> +	if (ret) {
>  		dev_err_probe(ucsi->dev, ret, "ucsi register failed\n");
> +		gaokun_ec_unregister_notify(uec->ec, &uec->nb);
> +		uec->notifier_registered = false;

Why this convulted logic of setting to true and then false?  Did you
have an AI write this?

And why are 2 flags needed?  Are you sure that's necessary?

How was this tested?

thanks,

greg k-h