>However, it does look like AGPIOC_ALLOCATE is broken. It only returns >the ->physical field of the resulting agp_memory structure. It doesn't >even look like this field is set for any chipsets other than the i810 >and i830.
You don't need anything other than the "key". This isn't a general purpose memory allocator. The agp memory doesn't have a usable address until you bind it into the gart, and at that point the client tells agpgart where in the gart to put it. It is only accessed through the gart. The "physical" pointer is returned only for a special type of memory where you need to know both the physical address of the page and the location in the gart. Intel's cursor uses this because the cursor takes a full 32 bit address so it can be used in non-gart modes too. >Is /dev/agpgart even supposed to work on Linux? I don't think any >programs use it, and the rest of the kernel used the backend interfaces >directly. It's possible that the frontend has fallen out of sync with >the back. Can anyone here confirm or refute this? The X server uses it when you are not running DRM. -Matt _______________________________________________ Dri-devel mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/dri-devel