Re: [PATCH] PCI: make pci_set_mwi a dummy for PCIe devices

[Bjorn Helgaas <helgaas@kernel.org>]

On Sun, Oct 21, 2018 at 05:06:58PM +0200, Heiner Kallweit wrote:
> Memory-Write-Invalidate isn't applicable for PCIe devices, therefore
> make the MWI functions a dummy and don't even write to the (dummy)
> PCI config space registers.
> 
> Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
> ---
>  drivers/pci/pci.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
> index d068f11d0..fb63b3d89 100644
> --- a/drivers/pci/pci.c
> +++ b/drivers/pci/pci.c
> @@ -4137,6 +4137,9 @@ int pci_set_mwi(struct pci_dev *dev)
>  	int rc;
>  	u16 cmd;
>  
> +	if (pci_is_pcie(dev))
> +		return 0;

I don't think this works quite right for the case of a
PCIe-to-PCI/PCI-X bridge.  That is a PCIe device (it has a PCIe
Capability, and pci_is_pcie() should return true), but the MWI bit is
writable and meaningful (it controls whether the bridge can convert a
PCIe Memory Write into a PCI Memory Write and Invalidate), so I don't
think it would be correct to just skip this.

I'm looking at the "PCI EXPRESS TO PCI/PCI-X BRIDGE" spec, r1.0, sec
5.1.1.1.

>  	rc = pci_set_cacheline_size(dev);
>  	if (rc)
>  		return rc;
> @@ -4203,6 +4206,9 @@ void pci_clear_mwi(struct pci_dev *dev)
>  #ifndef PCI_DISABLE_MWI
>  	u16 cmd;
>  
> +	if (pci_is_pcie(dev))
> +		return;
> +
>  	pci_read_config_word(dev, PCI_COMMAND, &cmd);
>  	if (cmd & PCI_COMMAND_INVALIDATE) {
>  		cmd &= ~PCI_COMMAND_INVALIDATE;
> -- 
> 2.19.1
>