Re: 2.4.8-ac8, agpgart, r128, and mtrr

Re: 2.4.8-ac8, agpgart, r128, and mtrr

[Keith Owens]

On Tue, 21 Aug 2001 05:01:53 -0700 (PDT), 
Brian Dushaw <dushaw@apl.washington.edu> wrote:
>I have an ATI Technologies Inc Rage 128 PF video card.
>The module r128.o needs to have agpgart loaded first and depmod
>does not seem to set this up properly.  I've added the line:
>"pre-install r128 modprobe agpgart " to my /etc/modules.conf file.

"before r128 agpgart" is better.

>I wouldn't have thought this would require user intervention like this.

modprobe and depmod only handle direct symbol dependencies between
modules.  The only symbols exported by agpgart are agp_free_memory,
agp_allocate_memory, agp_copy_info, agp_bind_memory, agp_unbind_memory,
agp_enable, agp_backend_acquire, agp_backend_release.  None of these
symbols are referenced by r128 so modprobe is quite correct, there is
no symbol dependency between agpgart and r128.  If you want it to be
automatic, then r128 must reference an agpgart symbol.

But the drm maintainers decided not to rely on agp.  If agp is present
it will be used, if agp is not present then drm runs without it.
drivers/char/drm/drm_agpsupport.h uses inter_module_get() to see if agp
is loaded and to extract agp data, if available.  This allows drm to
run without forcing agp to be present and loaded.  So if you want agp
before drm, it must be explicitly loaded.  Design decision.

>Aug 20 13:04:37 localhost modprobe: modprobe: Can't locate module block-major-33

block-major-33 is the third IDE hard disk/CD-ROM interface, for hd[ef].

>Aug 20 13:04:40 localhost modprobe: modprobe: Can't locate module char-major-226

char-major-226 is DRI (DRM_MAJOR).  That one puzzles me, I cannot see
where it is being triggered from.

2.4.8-ac8, agpgart, r128, and mtrr

[Brian Dushaw]

I've had a few problems/issues in implementing kernel 2.4.8-ac8
(and others) and getting agpgart/r128 to work.  I've continued the insanity of
chasing the latest and greatest upgrades, and upgraded to XFree86 4.1.
But I had a similar experience with 4.0.

I have an ATI Technologies Inc Rage 128 PF video card.
The module r128.o needs to have agpgart loaded first and depmod
does not seem to set this up properly.  I've added the line:
"pre-install r128 modprobe agpgart " to my /etc/modules.conf file.
I wouldn't have thought this would require user intervention like this.

My motherboard is FIC AD11 with the AMD 760 chipset, so I had to add
"options agpgart agp_try_unsupported=1" to my /etc/modules.conf file,
which seems to work o.k.
[/proc/pci gives:
PCI devices found:
  Bus  0, device   0, function  0:
    Host bridge: Advanced Micro Devices [AMD] AMD-760 [Irongate] System Controller (rev 19).      Master Capable.  Latency=32.
      Prefetchable 32 bit memory at 0xd0000000 [0xd3ffffff].
      Prefetchable 32 bit memory at 0xdb001000 [0xdb001fff].
      I/O at 0xa000 [0xa003].
  Bus  0, device   1, function  0:
    PCI bridge: Advanced Micro Devices [AMD] AMD-760 [Irongate] AGP Bridge (rev 0).
      Master Capable.  Latency=32.  Min Gnt=14.
  Bus  0, device   7, function  0:
to be precise.]

When agpgart and r128 modules are loaded there are some mtrr complaints
of which I know nothing.  dmesg gives:

mtrr: type mismatch for d4000000,2000000 old: uncachable new: write-combining
Linux agpgart interface v0.99 (c) Jeff Hartmann
agpgart: Maximum main memory to use for agp memory: 203M
agpgart: Trying generic AMD routines for device id: 700e
agpgart: AGP aperture is 64M @ 0xd0000000
[drm] AGP 0.99 on AMD @ 0xd0000000 64MB
mtrr: type mismatch for d0000000,4000000 old: uncachable new: write-combining
[drm] Initialized r128 2.1.6 20010405 on minor 0
mtrr: type mismatch for d4000000,2000000 old: uncachable new: write-combining
mtrr: type mismatch for d4000000,2000000 old: uncachable new: write-combining
mtrr: type mismatch for d4000000,2000000 old: uncachable new: write-combining

/var/log/messages gives:
Aug 20 13:04:37 localhost modprobe: modprobe: Can't locate module block-major-33
Aug 20 13:04:40 localhost kernel: mtrr: type mismatch for d4000000,2000000 old: uncachable new: write-combining
Aug 20 13:04:40 localhost modprobe: modprobe: Can't locate module char-major-226
Aug 20 13:04:40 localhost modprobe: modprobe: Can't locate module char-major-226
Aug 20 13:04:40 localhost kernel: Linux agpgart interface v0.99 (c) Jeff Hartmann
Aug 20 13:04:40 localhost kernel: agpgart: Maximum main memory to use for agp memory: 203M
Aug 20 13:04:40 localhost kernel: agpgart: Trying generic AMD routines for device id: 700e
Aug 20 13:04:40 localhost kernel: agpgart: AGP aperture is 64M @ 0xd0000000
Aug 20 13:04:40 localhost kernel: [drm] AGP 0.99 on AMD @ 0xd0000000 64MB
Aug 20 13:04:40 localhost kernel: mtrr: type mismatch for d0000000,4000000 old: uncachable new: write-combining
Aug 20 13:04:40 localhost kernel: [drm] Initialized r128 2.1.6 20010405 on minor 0
Aug 20 13:04:40 localhost kernel: mtrr: type mismatch for d4000000,2000000 old: uncachable new: write-combining
Aug 20 13:04:42 localhost last message repeated 2 times


Lastly, for all that, agpgart doesn't seem to do much for me -
when loaded in, lsmod shows it as being unused, if r128 is not
loaded.  I have the XFree86 glx module loaded, I think.  I think I don't
understand this at all... without the dri modules agpgart seems
innocuous...so why bother?

Anyways, that's my experience with this; I thought the report would be worthwhile.
-- 
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

Brian Dushaw
Applied Physics Laboratory
University of Washington
1013 N.E. 40th Street
Seattle, WA  98105-6698
(206) 685-4198   (206) 543-1300
(206) 543-6785 (fax)
dushaw@apl.washington.edu

Web Page:  http://staff.washington.edu/dushaw/index.html

Re: 2.4.8-ac8, agpgart, r128, and mtrr

[Alan Cox]

> The module r128.o needs to have agpgart loaded first and depmod
> does not seem to set this up properly.  I've added the line:

Its not a dependancy it can deduce. Imagine your card was PCI

> "pre-install r128 modprobe agpgart " to my /etc/modules.conf file.

Correct.

> when loaded in, lsmod shows it as being unused, if r128 is not
> loaded.  I have the XFree86 glx module loaded, I think.  I think I don't
> understand this at all... without the dri modules agpgart seems
> innocuous...so why bother?

AGP for most cards is currently used for 3D acceleration only