Re: [PATCH v16 14/18] PCI/CXL: Reject reset with unsafe function scope

[Srirangan Madhavan <smadhavan@nvidia.com>]

On 11/5/26 6:36 AM, Joshua Machauer wrote:
> External email: Use caution opening links or attachments
> 
> 
> On Thu, Oct 29, 2026 at 02:56:59PM +0000, Srirangan Madhavan wrote:
>> @@ -515,6 +515,9 @@ int cxl_reset_function(struct pci_dev *pdev, bool probe)
>>            !(ctrl & PCI_DVSEC_CXL_MEM_ENABLE))
>>                return -ENOTTY;
>>
>> +     if (pdev->is_virtfn || pdev->multifunction || pci_num_vf(pdev))
>> +             return -ENOTTY;
>> +
>>        if (probe)
>>                return 0;
> 
> [...]
> 
>> @@ -5026,6 +5026,9 @@ static int cxl_reset_bus_function(struct pci_dev *dev, bool probe)
>>        if (!dvsec)
>>                return -ENOTTY;
>>
>> +     if (dev->is_virtfn || dev->multifunction || pci_num_vf(dev))
>> +             return -ENOTTY;
>> +
>>        if (probe)
>>                return 0;
> 
> Could both checks go at the top of their functions, before the DVSEC
> lookups? None of the three conditions depends on what the lookups
> find.
> 
> pci_init_reset_methods() probes every reset method for every device
> that is added, and enabling SR-IOV adds the VFs one after another.
> Where the checks are now, each VF probe walks the extended
> capabilities twice before it returns -ENOTTY: cxl_reset walks the
> VF's own list for the CXL Device DVSEC, and cxl_bus walks the
> upstream bridge's list for the CXL Port DVSEC. All VFs of a PF share
> that bridge, so its walk repeats once per VF with the same result.
> 
> With the checks first, a VF returns -ENOTTY without a config read,
> and the result for every device stays the same.
> 
> Thanks,
> Joshua

Hi Joshua,

Thanks, that makes sense. In v17 both checks run at the top of
cxl_reset_function() and cxl_reset_bus_function(), before the DVSEC
lookups.

-- 
Regards,
Srirangan