> Record the error code before freeing "aon_chan" to avoid a
> use after free.
Would it become helpful to mention which selection of source code
analysis tools detected such a questionable implementation detail?
…
> +++ b/drivers/firmware/thead,th1520-aon.c
> @@ -203,6 +203,7 @@ struct th1520_aon_chan *th1520_aon_init(struct device *dev)
> {
> struct th1520_aon_chan *aon_chan;
> struct mbox_client *cl;
> + int ret;
>
> aon_chan = kzalloc(sizeof(*aon_chan), GFP_KERNEL);
> if (!aon_chan)
> @@ -217,8 +218,9 @@ struct th1520_aon_chan *th1520_aon_init(struct device *dev)
> aon_chan->ch = mbox_request_channel_byname(cl, "aon");
> if (IS_ERR(aon_chan->ch)) {
> dev_err(dev, "Failed to request aon mbox chan\n");
> + ret = PTR_ERR(aon_chan->ch);
> kfree(aon_chan);
> - return ERR_CAST(aon_chan->ch);
> + return ERR_PTR(ret);
> }
>
> mutex_init(&aon_chan->transaction_lock);
May the additional variable (for an information) be defined only for
the affected if branch?
Would a smaller scope be more appropriate here?
Regards,
Markus
