Re: [PATCH v2] drivers/perf: xgene_pmu: Fix uninitialized resource struct

[Mark Salter]

On Mon, 2020-09-14 at 12:28 +0100, Will Deacon wrote:
> On Sun, Sep 13, 2020 at 01:45:36PM -0400, Mark Salter wrote:
> > @@ -1483,11 +1473,23 @@ xgene_pmu_dev_ctx *acpi_get_pmu_hw_inf(struct xgene_pmu *xgene_pmu,
> >  		return NULL;
> >  
> >  	INIT_LIST_HEAD(&resource_list);
> > -	rc = acpi_dev_get_resources(adev, &resource_list,
> > -				    acpi_pmu_dev_add_resource, &res);
> > +	rc = acpi_dev_get_resources(adev, &resource_list, NULL, NULL);
> > +	if (rc <= 0) {
> > +		dev_err(dev, "PMU type %d: No resources found\n", type);
> > +		return NULL;
> > +	}
> > +
> > +	list_for_each_entry(rentry, &resource_list, node) {
> > +		if (resource_type(rentry->res) == IORESOURCE_MEM) {
> > +			res = *rentry->res;
> > +			rentry = NULL;
> > +			break;
> > +		}
> > +	}
> >  	acpi_dev_free_resource_list(&resource_list);
> > -	if (rc < 0) {
> > -		dev_err(dev, "PMU type %d: No resource address found\n", type);
> > +
> > +	if (rentry) {
> 
> I'm curious as to why you've had to change the failure logic here, setting
> rentry to NULL instead of checking 'rentry->res' like the TX2 driver (which
> I don't immediately understand at first glance).
> 
> Will
> 

I don't fully understand the tx2 logic. I do see there's a memory leak if
that (!rentry->res) is true. I was going to dig deeper and follow up with
a patch for tx2. I suspect that rentry->res should never be true. And tx2
won't detect if no memory resource is in the table.

Mark