Re: [PATCH 21/23] pci: add MemoryRegion based BAR management API

[Avi Kivity <avi@redhat.com>]

On 07/25/2011 11:20 PM, Anthony Liguori wrote:
> On 07/25/2011 09:03 AM, Avi Kivity wrote:
>> Allow registering a BAR using a MemoryRegion.  Once all users are 
>> converted,
>> pci_register_bar() and pci_register_bar_simple() will be removed.
>>
>> Signed-off-by: Avi Kivity<avi@redhat.com>
>
>> diff --git a/hw/pci.h b/hw/pci.h
>> index cfeb042..c51156d 100644
>> --- a/hw/pci.h
>> +++ b/hw/pci.h
>> @@ -94,6 +94,7 @@ typedef struct PCIIORegion {
>>       uint8_t type;
>>       PCIMapIORegionFunc *map_func;
>>       ram_addr_t ram_addr;
>> +    MemoryRegion *memory;
>>   } PCIIORegion;
>>
>>   #define PCI_ROM_SLOT 6
>> @@ -204,6 +205,8 @@ void pci_register_bar(PCIDevice *pci_dev, int 
>> region_num,
>>                               PCIMapIORegionFunc *map_func);
>>   void pci_register_bar_simple(PCIDevice *pci_dev, int region_num,
>>                                pcibus_t size, uint8_t attr, 
>> ram_addr_t ram_addr);
>> +void pci_register_bar_region(PCIDevice *pci_dev, int region_num,
>> +                             uint8_t attr, MemoryRegion *memory);
>
> This ends up being a very nice API.  I had always thought this should 
> be a PCI specific set of callbacks but I do see the benefits of having 
> the callbacks be generic.
>
> Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
>
> One thing I'm curious about, what's the symmetric view of this API?
>
> Would you see a device doing something like:
>
> memory_region_read(&dev->pci_bus->memory, addr, &data, sizeof(data))
>

I haven't really considered this, but it makes sense to consider the 
initiating point for DMA.  This allows us to do IOMMU transformations 
naturally.

-- 
error compiling committee.c: too many arguments to function