On Mon, Dec 10, 2001 at 12:05:47PM +0200, Abraham vd Merwe wrote: > ... > It is not always necessary to memory map the device. For instance with the > I810 driver, just plain 2D, there is no need to mmap the device, but the GTT > (or GATT if you wish) must be populated because the card has no onboard > memory and by default. Since the ALLOCATE ioctl() allocates physical memory > and the BIND ioctl() populates the GTT, the process looks like this: ...
Errr.... wonder how xfree86 works with i810 currently under solaris then, since there is currently no AGP driver for solaris. > > Just the "physical" address return, which could perfectly easily be done > > through an agp_allocate() call that would otherwise do nothing, I think. > > ? > > Sometimes, the X driver needs to know the _physical_ location of the page > which it allocates. For instance, there is 2 cases where the I810 series of > chips need this information: 1) hardware cursor and 2) if you want to use > the hardware status page and not do the silly spin until head == tail for > the ring buffer (which btw is the way the Linux driver does this right now). > That is why the agpgart module provides a way to get the physcal location of > a page. Right, I read that code. [otherwise, I wasnt going to return the physical addr at all :-)] So if I provided that through the agp_allocate() return, but had already allocated the memory for the entire aperture beforehand... sounds like it would work fine? _______________________________________________ Dri-devel mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/dri-devel
