[v1,1/2] PCI: Probe for device reset support before driver claim

From: Bjorn Helgaas <bhelgaas@google.com>

Previously we called pci_probe_reset_function() in this path:

  pci_sysfs_init                              # late_initcall
    for_each_pci_dev(dev)
      pci_create_sysfs_dev_files(dev)
	pci_create_capabilities_sysfs(dev)
	  pci_probe_reset_function
	    pci_dev_specific_reset
	    pcie_has_flr
              pcie_capability_read_dword

pci_sysfs_init() is a late_initcall, and a driver may have already claimed
one of these devices and enabled runtime power management for it, so the
device could already be in D3hot by the time we get to pci_sysfs_init().

The device itself should respond to the config read even while it's in
D3hot, but if an upstream bridge is also in D3hot, the read won't even
reach the device because the bridge won't forward it downstream to the
device.  If the bridge is a PCIe port, it should complete the read as an
Unsupported Request, which may be reported to the CPU as an exception or as
invalid data.

Avoid this case by probing for reset support from pci_init_capabilities(),
before a driver can claim the device.  The device should be in D0 and fully
accessible at that point.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
---
 drivers/pci/pci-sysfs.c |    3 +--
 drivers/pci/probe.c     |    3 +++
 2 files changed, 4 insertions(+), 2 deletions(-)

On Friday, February 16, 2018 11:50:11 PM CET Bjorn Helgaas wrote:
> From: Bjorn Helgaas <bhelgaas@google.com>
> 
> Previously we called pci_probe_reset_function() in this path:
> 
>   pci_sysfs_init                              # late_initcall
>     for_each_pci_dev(dev)
>       pci_create_sysfs_dev_files(dev)
> 	pci_create_capabilities_sysfs(dev)
> 	  pci_probe_reset_function
> 	    pci_dev_specific_reset
> 	    pcie_has_flr
>               pcie_capability_read_dword
> 
> pci_sysfs_init() is a late_initcall, and a driver may have already claimed
> one of these devices and enabled runtime power management for it, so the
> device could already be in D3hot by the time we get to pci_sysfs_init().

It also may be in D3cold if, say, AML can cut off power from it or ACPI
power resources can be used for that.

> The device itself should respond to the config read even while it's in
> D3hot, but if an upstream bridge is also in D3hot, the read won't even
> reach the device because the bridge won't forward it downstream to the
> device.  If the bridge is a PCIe port, it should complete the read as an
> Unsupported Request, which may be reported to the CPU as an exception or as
> invalid data.
> 
> Avoid this case by probing for reset support from pci_init_capabilities(),
> before a driver can claim the device.  The device should be in D0 and fully
> accessible at that point.
> 
> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>

Reviewed-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

> ---
>  drivers/pci/pci-sysfs.c |    3 +--
>  drivers/pci/probe.c     |    3 +++
>  2 files changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
> index eb6bee8724cc..4933f0270471 100644
> --- a/drivers/pci/pci-sysfs.c
> +++ b/drivers/pci/pci-sysfs.c
> @@ -1542,11 +1542,10 @@ static int pci_create_capabilities_sysfs(struct pci_dev *dev)
>  	/* Active State Power Management */
>  	pcie_aspm_create_sysfs_dev_files(dev);
>  
> -	if (!pci_probe_reset_function(dev)) {
> +	if (dev->reset_fn) {
>  		retval = device_create_file(&dev->dev, &reset_attr);
>  		if (retval)
>  			goto error;
> -		dev->reset_fn = 1;
>  	}
>  	return 0;
>  
> diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
> index ef5377438a1e..489660d0d384 100644
> --- a/drivers/pci/probe.c
> +++ b/drivers/pci/probe.c
> @@ -2121,6 +2121,9 @@ static void pci_init_capabilities(struct pci_dev *dev)
>  
>  	/* Advanced Error Reporting */
>  	pci_aer_init(dev);
> +
> +	if (pci_probe_reset_function(dev) == 0)
> +		dev->reset_fn = 1;
>  }
>  
>  /*
> 
>