cvs commit: src/sys/i386/pci pci_cfgreg.c

[Scott Long]

Peter Wemm wrote:
> On Monday 13 December 2004 05:33 pm, David O'Brien wrote:
> 
>>On Mon, Dec 13, 2004 at 02:16:53PM -0800, Peter Wemm wrote:
>>
>>>On Monday 13 December 2004 01:47 pm, Dag-Erling Sm?rgrav wrote:
>>>
>>>>"David O'Brien" <obrien at FreeBSD.org> writes:
>>>>
>>>>>At the moment I don't know of anyone with an AMD64 PCI Express
>>>>>motherboard.
>>>>
>>>>The Celestica A8440 has dual onboard PCI-X NICs and two PCI-X
>>>>slots. Surely someone in the project should be able to get their
>>>>hands on one?  I for one would love to have one for the
>>>>tinderbox...
>>>>
>>>>DES
>>>
>>>Just for clarity, we were talking about PCI-Express, not PCI-X.
>>>
>>>BTW, I think it is amusing that the only machines readily available
>>>that have PCI-Express to run FreeBSD/amd64 on are Intel systems...
>>
>>Why?
>>Today there is zero benefit to PCI-Express vs. AGP 8x or PCI-X
>>133mhz.
> 
> 
> Just like there was zero benefit to having a prototype amd64 machine to 
> do porting work on before they started coming out in retail?  At the 
> time, there was zero benefit compared to a 32 bit Athlon-XP system.
> 
> It's going to suck when they do reach the stores and our answer to 
> support questions is "no", while Linux and windows do.
> 
> On that note, we still barely support anything more from AMD's platform 
> than the basic instruction set due to lack of documentation..  That's 
> why FreeBSD/amd64 ran so quickly on the Intel hardware.   We only used 
> the most basic features of the generic instruction set in an otherwise 
> PC platform.  We don't take advantage of the IOMMU, the timers, etc.  
> (Still no documentation).  I'd have expected AMD to push as hard as 
> they can to get every advantage while Intel is still trying to recover 
> their balance.

Having IOMMU docs could have a very positive benefit for things like ATA
on amd64.  However, I can't even begin to evaluate what that benefit
might be until docs become available.


Scott