On Fri, 2007-05-04 at 13:13 -0700, Nate Lawson wrote: > I have a desktop with i945GMS builtin video. It appears agp_i810.c is > not built on amd64, hence no support for agp. This is on 6.x. > > none0_at_pci0:2:0: class=0x030000 card=0x31031565 chip=0x27728086 rev=0x02 > hdr=0x00 > vendor = 'Intel Corporation' > device = 'Integrated Graphics Controller' > class = display > subclass = VGA > > sys/conf/files.i386:pci/agp_i810.c optional agp > > $ ls /sys/amd64/compile/SMP/*agp* > /sys/amd64/compile/SMP/agp.o /sys/amd64/compile/SMP/agp_if.h > /sys/amd64/compile/SMP/agp_amd64.o /sys/amd64/compile/SMP/agp_if.o > /sys/amd64/compile/SMP/agp_if.c /sys/amd64/compile/SMP/agp_intel.o > > Is there any reason not to use i810 on amd64? A quick glance through > doesn't show it doing anything 32-bit specific. On -current, it appears > to be enabled for amd64. It appears this MFC never happened: > > revision 1.93 > date: 2006/09/05 16:55:12; author: anholt; state: Exp; lines: +1 -0 > Include agp_i810.c in amd64 AGP builds to get support for the Intel 915 > Express chipsets. Nope, there's no good reason for the MFC to have not happened. We've only got one amd64 issue that I know of, but it's actually not unique to amd64. We're not allocating our pages to back aperture allocations to meet the requirements of the page tables, so for example on i915-class hardware with 4GB of ram, you'll get pages sometimes that are above the 4GB range, and we'll just drop the top bit of the address when filling in the page table, resulting in pain. We need a way for AGP drivers to communicate their requirements for AGP memory to the core, likely through busdma tags. In the current agp_i810.c patch I've been working on, I at least panic when we get out-of-range pages, instead of scribbling on random memory. -- Eric Anholt anholt_at_FreeBSD.org eric_at_anholt.net eric.anholt_at_intel.com
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:09 UTC