On Mon, Apr 16, 2012 at 12:20 PM, Andreas Färber <afaer...@suse.de> wrote:
> Am 16.04.2012 11:14, schrieb Stefan Hajnoczi:
>> On Sat, Apr 14, 2012 at 12:32:00PM +0000, Blue Swirl wrote:
>>> On Fri, Apr 13, 2012 at 14:27, Stefan Hajnoczi
>>> <stefa...@linux.vnet.ibm.com> wrote:
>>>> This patch adds a common PCI bus driver library which works for
>>>> i386/x86-64 targets. Tests can use the library to probe for PCI
>>>> devices, map BARs, and access configuration space.
>>>
>>> I guess we have almost identical code in SeaBIOS, OpenBIOS, various OS
>>> and maybe userland PCI tools. Would it be possible to reduce NIH
>>> somewhere?
>>
>> Probably not given how small these functions are and how they use glib
>> assert calls because they are part of tests.
>>
>>>> +void pci_config_writel(PciDevice *dev, unsigned int offset, uint32_t l)
>>>> +{
>>>> + pci_config_setup(dev, offset);
>>>> + outl(PCI_CONFIG_DATA, l);
>>>> +}
>>>
>>> All code above is specific to i440fx or similar PCI bridges, other
>>> bridges may use different config space access methods. If we want to
>>> share the rest for example with Sparc64 or PPC, the above would need
>>> to be changed. How about splitting the above to a separate file? It
>>> could be done later too.
>>
>> Yes, it's only i440fx for now. I think it makes sense to move it later
>> since we have no non-x86 qtests yet.
>
> As stated before, I'm very grateful of your work in this area. I'd be
> very interested in having PCI-based tests for the PReP devices we're
> introducing - some outline of how non-x86 targets are supposed to fit in
> here would be appreciated. For example, would we want to rename the file
> to libpci-i440fx.c and have libpci.h be a common interface for multiple
> implementations? Or are you expecting some #ifdef TARGET_FOO inside
> libpci.c?
I think it's cleanest to have libpci-i440fx.c. When support for the
next PCI controller gets added we can decide the details of how to
split it.
Stefan
