On Mon, 2002-09-30 at 19:04, Jeff Hartmann wrote: > I know we have talked about this issue before, but I want to rehash. The > statement that only what the bios programs is not entirely correct, but this > does hold true for some chipsets where we don't know all the details of agp > mode switching.
OK. So why do we honour AGPMode for the others ? > correctly. According to the agp spec whatever capability bits are available > are the modes that are valid for use. The agp kernel module will look at > the capability bits and only allow bits that are set in the capability > register to be used. It was designed this way to protect the user against > setting agp modes that are not supported by the hardware, thus a user can't > set agp mode to 4x if 4x isn't in the capability register. According to the > agp spec, thats how mode setting is supposed to work. Ok > memory. This is why the Xserver defaults to agp 1x in almost all cases, > since its the most reliable setting. This variable actually can be tuned, > but unless you are agp bandwidth limited (which is not the common case, I get reports from people quite often (because agp is clearly the kernel) where setting AGP to the mode currently active according to the cap registers (typically 2x) works and 1x hangs randomly. > If someone will give me a list of chipsets (pci vendor/device pairs) that > we know require the agp mode to be programmed by the bios I will write an > overrided function for their agp drivers so they will not set the graphics > mode to anything but what is required. This is the simple solution to the > problems people are having, and I can get a patch ready very shortly. I will try and get PCI idents with future bug reports. Actually I seem to remember at least one recent report like this to dri-devel as well ? > I hope this clears up the issue, cause I know it has been a source of pain > for Alan and many other kernel developers. I wish this was a simple issue, > but it is not unfortunately. Thanks. As ever the truth and the specification tend to be different things 8( ------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ Dri-devel mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/dri-devel
