Re: [PATCH V1 1/6] rpmsg: glink: fix destroy channel endpoint logic

[Bjorn Andersson <bjorn.andersson@linaro.org>]

On Thu 30 Jul 00:18 CDT 2020, Deepak Kumar Singh wrote:

> From: Konstantin Dorfman <kdorfman@codeaurora.org>
> 
> When rpmsg client driver destroys last channel endpoint, remove rpmsg
> device is triggered. In both cases (destroy endpoint and remove device)
> a glink close command sent to the remote peer.
> 
> This change, when for removing rpmsg device endpoint already destroyed
> will avoid sending second glink close command.
> 

Should it really be considered valid to rpmsg_destroy_ept() the
rpmsg_device's primary endpoint?

Do you have a use case where this makes sense?


Also, I think this has a potential to hide a problems of clients doing a
"double free" on the ept.

Regards,
Bjorn

> Signed-off-by: Konstantin Dorfman <kdorfman@codeaurora.org>
> Signed-off-by: Deepak Kumar Singh <deesin@codeaurora.org>
> ---
>  drivers/rpmsg/qcom_glink_native.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/drivers/rpmsg/qcom_glink_native.c b/drivers/rpmsg/qcom_glink_native.c
> index 1995f5b..2668c66 100644
> --- a/drivers/rpmsg/qcom_glink_native.c
> +++ b/drivers/rpmsg/qcom_glink_native.c
> @@ -1210,6 +1210,10 @@ static void qcom_glink_destroy_ept(struct rpmsg_endpoint *ept)
>  	unsigned long flags;
>  
>  	spin_lock_irqsave(&channel->recv_lock, flags);
> +	if (!channel->ept.cb) {
> +		spin_unlock_irqrestore(&channel->recv_lock, flags);
> +		return;
> +	}
>  	channel->ept.cb = NULL;
>  	spin_unlock_irqrestore(&channel->recv_lock, flags);
>  
> -- 
> The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
> a Linux Foundation Collaborative Project
>