Re: [PATCH -next 8/8] soc: ti: knav_qmss_queue: Simplify with scoped for each OF child loop

[Jonathan Cameron <jic23@kernel.org>]

On Fri, 30 Aug 2024 11:24:14 +0800
Jinjie Ruan <ruanjinjie@huawei.com> wrote:

> On 2024/8/29 23:58, Kousik Sanagavarapu wrote:
> > Jinjie Ruan <ruanjinjie@huawei.com> writes:  
> >> @@ -1080,17 +1080,13 @@ static int knav_queue_setup_regions(struct knav_device *kdev,
> >>  {
> >>  	struct device *dev = kdev->dev;
> >>  	struct knav_region *region;
> >> -	struct device_node *child;
> >>  	u32 temp[2];
> >>  	int ret;
> >>  
> >> -	for_each_child_of_node(regions, child) {
> >> +	for_each_child_of_node_scoped(regions, child) {  
> > 
> > Are you sure using *_scoped() is better here?  Since it seems that we
> > need the memory pointed to by "child" in cases where we don't go into an
> > error path.  
> 
> Hi, Jonathan, could you help review this code?

I don't understand the review comment.
The reference counting before and after this patch is the same, just
with the error path handled in a simpler fashion and the scope of
the child variable reduced.


> 
> >   
> >>  		region = devm_kzalloc(dev, sizeof(*region), GFP_KERNEL);
> >> -		if (!region) {
> >> -			of_node_put(child);
> >> -			dev_err(dev, "out of memory allocating region\n");
> >> -			return -ENOMEM;
> >> -		}
> >> +		if (!region)
> >> +			return dev_err_probe(dev, -ENOMEM, "out of memory allocating region\n");
> >>  
> >>  		region->name = knav_queue_find_name(child);
> >>  		of_property_read_u32(child, "id", &region->id);  
> > 
> > Similarly in most of the other cases in this series where a similar
> > change is done.
> > 
> > Also FYI, as for dev_err_probe(), I think I covered all of them in this
> > file and a patch for it is currently sitting in ti-drivers-soc-next.
> > 
> > Thanks