Re: [PATCH v3 2/2] PCI: Add support for Enhanced Allocation devices

[David Daney <ddaney@caviumnetworks.com>]

On 09/30/2015 03:50 PM, Sean O. Stalley wrote:
> On Tue, Sep 29, 2015 at 04:53:20PM -0700, David Daney wrote:
>> On 09/29/2015 03:47 PM, Sean O. Stalley wrote:
>> [...]
>>>>>> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
>>>>>> index 6a9a111..7c60b16 100644
>>>>>> --- a/drivers/pci/pci.c
>>>>>> +++ b/drivers/pci/pci.c
>>>>>> @@ -2148,6 +2148,284 @@ void pci_pm_init(struct pci_dev *dev)
>>>>>>   	}
>>>>>>   }
>>>>>>
>>>>>> +static unsigned long pci_ea_set_flags(struct pci_dev *dev, u8 prop)
>>>>>> +{
>>>>>> +	unsigned long flags = IORESOURCE_PCI_FIXED | IORESOURCE_SIZEALIGN;
>>>>>
>>>>> Why did you add the IORESOURCE_SIZEALIGN flag? EA allows for unaligned resources.
>>>>
>>>> pci_bus_assign_resources() fails causing the devices to be unusable
>>>> if resource_alignment() returns zero.  The easiest fix for this was
>>>> to specify IORESOURCE_SIZEALIGN.
>>>>
>>>> An alternative would be to change the code in setup-bus.c so that
>>>> for IORESOURCE_PCI_FIXED resources, it didn't barf.
>>>
>>> I would do this alternative, but with a IORESOURCE_PCI_EA flag.
>>>
>>
>> I don't think we need IORESOURCE_PCI_EA.  If a resource is tagged as
>> IORESOURCE_PCI_FIXED that means that it cannot be changed, we
>> shouldn't care why it is fixed (due to an EA capability for
>> example).  We just need to gracefully handle IORESOURCE_PCI_FIXED in
>> the places where things currently go wrong.
>>
>> David Daney
>>
>
> I agree that we need to make sure fixed things stay fixed,
> regardless of if they are fixed by EA or something else.
>
> The question is: do we want to allow all fixed resources to be non-aligned, or just EA?

The alignment should only be important if a new address is being 
calculated for a BAR or behind the bridge memory/io region.  If the 
resource is fixed, I would almost say that "alignment" is an undefined 
concept.  So, in the places were we need special treatment for 
IORESOURCE_PCI_FIXED, I would argue that we shouldn't be use the 
resource alignment information.

David Daney

>
> -Sean
>