On Wed, Dec 26, 2001 at 02:26:13PM -0800, Philip Brown wrote: >I now have my solaris agp driver to where it can be mmap-able from the user >perspective. I have not tried to recompile xfree with it yet, because : > > The "testgart" program I have, does an UNBIND and re-BIND as part of its > testing. My driver fails the rebind. > >Is this rebinding common? >The only time I can see this as being useful, is if the user-level program >allocates more AGP keyed memory than the capacity of the aperture, and then >uses UNBIND/reBIND as essentially fast-copy operations. > >However, if I'm reading the BSD agp driver, it does not allow >AGPIOC_ALLOCATE of more memory than the size of the aperture.
I don't know what precisely testgart is testing here, but I do know of a case where there can be an unbind followed by a rebind. The i810 driver does unbind/rebind. It unbinds (and releases) when the X server gives up it VT, and it (acquires and) re-binds when it gets it back. It does this so that something else can bind when it's not using it. It doesn't unalloc/realloc because there's no guarantee that the realloc would succeed (and maybe other reasons?). If you're using the BSD driver as a reference, use a recent version. Until recently it didn't allow the unbind/release/acquire/rebind sequence (the release freed all allocations), and this broke VT switching with the i810 driver. David -- David Dawes Email: [EMAIL PROTECTED] Tungsten Graphics, Inc http://www.tungstengraphics.com Founder/President, Release Engineer Phone: +1 570 764 0288 The XFree86 Project, Inc http://www.xfree86.org/~dawes _______________________________________________ Dri-devel mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/dri-devel
