Re: [PATCH] PCI: dwc: Fix potential NULL dereference

[Bjorn Helgaas]

The subject line should be:

  PCI: al: ...

since this fix is specific to the "al" driver, not generic to "dwc".

On Thu, Mar 28, 2024 at 09:01:26PM +0300, Aleksandr Mishin wrote:
> In al_pcie_config_prepare() resource_list_first_type() may return
> NULL which is later dereferenced. Fix this bug by adding NULL check.
> 
> Found by Linux Verification Center (linuxtesting.org) with SVACE.
> 
> Fixes: 0f71c60ffd26 ("PCI: dwc: Remove storing of PCI resources")
> Signed-off-by: Aleksandr Mishin <amishin@xxxxxxxxxx>
> ---
>  drivers/pci/controller/dwc/pcie-al.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/pci/controller/dwc/pcie-al.c b/drivers/pci/controller/dwc/pcie-al.c
> index 6dfdda59f328..29bc99d48295 100644
> --- a/drivers/pci/controller/dwc/pcie-al.c
> +++ b/drivers/pci/controller/dwc/pcie-al.c
> @@ -252,7 +252,12 @@ static void al_pcie_config_prepare(struct al_pcie *pcie)
>  	u8 secondary_bus;
>  	u32 cfg_control;
>  	u32 reg;
> -	struct resource *bus = resource_list_first_type(&pp->bridge->windows, IORESOURCE_BUS)->res;
> +
> +	struct resource_entry *ft = resource_list_first_type(&pp->bridge->windows, IORESOURCE_BUS); 
> +	if (!ft)
> +		return;

I don't think this is right.  If we don't have an IORESOURCE_BUS
resource and we just silently return here, we will not write the
CFG_CONTROL register.  It looks essential that CFG_CONTROL be set, so
if we can't do that, the .probe() should fail.

But I think we are actually guaranteed that there is an IORESOURCE_BUS
resource because this path fabricates one if the "bus-range" DT
property doesn't exist:

 al_pcie_probe
    dw_pcie_host_init
      devm_pci_alloc_host_bridge
        devm_of_pci_bridge_init
          pci_parse_request_of_pci_ranges
            devm_of_pci_get_host_bridge_resources
              err = of_pci_parse_bus_range
                if (err)
                  bus_range->flags = IORESOURCE_BUS  # <--

I wouldn't necessarily object to doing something like other drivers
do:

  gen_pci_init
    bus = resource_list_first_type(&bridge->windows, IORESOURCE_BUS);
    if (!bus)
      return ERR_PTR(-ENODEV);

  xilinx_cpm_pcie_probe
    bus = resource_list_first_type(&bridge->windows, IORESOURCE_BUS);
    if (!bus)
      return -ENODEV;

But it would have to lead to .probe() failing, not just a silent
skipping of CFG_CONTROL setup.

> +	struct resource *bus = ft->res;
>  
>  	target_bus_cfg = &pcie->target_bus_cfg;
>  
> -- 
> 2.30.2
>