Re: PCI?

[shawn@iis10.iis.sinica.edu.tw (Pai-Hsiang Hsiao) writes:]

>I ported the code of PCI configuration from NetBSD, and implemented as
>a user level process on VSTa. Everything went fine, and I am trying to
>write more serious codes.

Great!

>Then, a question comes in mind. How should this routines be organized?
>I think I'd best implement it as a server, just like cam and wd do.
>The PCI peripherals need information on configuration then connect to
>and ask the server for those they need.

This would be my vote, definitely.

>Is it a good way? Yes, I can not see any problem for integrating the
>routines into the kernel, but I think it's not a good way, and disobeies
>the spirit of VSTa IMO.

No other driver has code in the kernel. I'd be very surprised if PCI
required it!

>So, the first question is, how I implement this part, as a server or a
>part of kernel?

Server.

>If the suggestion is the former one, then, how the information be
>served?

I'm afraid my knowledge of PCI is quite limited. If it's even vaguely like
SCSI, then the server would look like a directory containing the devices you
discovered on the bus. This has the disadvantage that you can't add new PCI
device types without starting the monolithic PCI server, but is fairly
efficient, and has worked pretty well in practice for SCSI.

                                                Regards,
                                                Andy Valencia
Received on Thu Nov 23 08:51:32 1995