Re: [PATCH] xen-pciback: provide a "reset" sysfs file to try harder at an SBR

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Wed, Jul 09, 2014 at 05:03:17PM +0100, David Vrabel wrote:
> On 09/07/14 15:56, Konrad Rzeszutek Wilk wrote:
> > 
> > I think for your patch to work it has to de-register what the 
> > generic PCI code does in pci_create_capabilities_sysfs:
> > 
> > 1339         if (!pci_probe_reset_function(dev)) {
> > 1340                 retval = device_create_file(&dev->dev, &reset_attr);
> > 1341                 if (retval)
> > 1342                         goto error;
> > 1343                 dev->reset_fn = 1;
> > 1344         }
> 
> It just needs to have the inverse test.
> 
> --- a/drivers/xen/xen-pciback/pci_stub.c
> +++ b/drivers/xen/xen-pciback/pci_stub.c
> @@ -151,14 +151,11 @@ static int pcistub_try_create_reset_file(struct
> pci_dev *pci)
>  {
>  	struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(pci);
>  	struct device *dev = &pci->dev;
> -	struct kernfs_node *reset_dirent;
>  	int ret;
> 
> -	reset_dirent = sysfs_get_dirent(dev->kobj.sd, "reset");
> -	if (reset_dirent) {
> -		sysfs_put(reset_dirent);
> +	/* Already have a per-function reset? */
> +	if (pci_dev_reset(dev, 1) == 0)

OK, however the'reset' in SysFS does not do bus/slot reset.

It only calls 'pci_reset_function' - which is inadequate
in certain cases (GPUs, device does not really go in D3 when
asked too, etc) and that is where bus/slot reset functionality
is needed.

>  		return 0;
> -	}
> 
>  	ret = device_create_file(dev, &dev_attr_reset);
>  	if (ret < 0)
> 
> David