Intel driver vs MTRRs conflict

Intel driver vs MTRRs conflict

[J.A. Magallón]

Hi all...

I have a little problem (near religious, not real, as averything works fine)
with the intel driver crying about mtrrs:

[drm] Initialized drm 1.1.0 20060810
pci 0000:00:02.0: PCI INT A -> GSI 16 (level, low) -> IRQ 16
pci 0000:00:02.0: setting latency timer to 64
[drm] Initialized i915 1.6.0 20080730 on minor 0
X:3004 conflicting memory types 20000000-30000000 write-combining<->uncached-minus
reserve_memtype failed 0x20000000-0x30000000, track write-combining, req write-combining
X:3004 conflicting memory types 20000000-30000000 write-combining<->uncached-minus
reserve_memtype failed 0x20000000-0x30000000, track write-combining, req write-combining
X:3004 conflicting memory types 20000000-30000000 write-combining<->uncached-minus
reserve_memtype failed 0x20000000-0x30000000, track write-combining, req write-combining
X:3031 freeing invalid memtype 20000000-30000000
X:3004 conflicting memory types 20000000-30000000 write-combining<->uncached-minus
reserve_memtype failed 0x20000000-0x30000000, track write-combining, req write-combining
X:3032 freeing invalid memtype 20000000-30000000
X:3004 conflicting memory types 20000000-30000000 write-combining<->uncached-minus
reserve_memtype failed 0x20000000-0x30000000, track write-combining, req write-combining
X:3033 freeing invalid memtype 20000000-30000000
X:3004 conflicting memory types 20000000-30000000 write-combining<->uncached-minus
reserve_memtype failed 0x20000000-0x30000000, track write-combining, req write-combining
X:3034 freeing invalid memtype 20000000-30000000

But that mtrr is there:

one:~# cat /proc/mtrr
reg00: base=0x000000000 (    0MB), size=  512MB, count=1: write-back
reg01: base=0x01f500000 (  501MB), size=    1MB, count=1: uncachable
reg02: base=0x01f600000 (  502MB), size=    2MB, count=1: uncachable
reg03: base=0x01f800000 (  504MB), size=    8MB, count=1: uncachable
reg04: base=0x020000000 (  512MB), size=  256MB, count=1: write-combining

Where is the problem ? Kernel or driver side ?
This logs are with kernel 2.6.29-rc5, and intel driver 2.6.1.

As I said, everything works. And even fast for the kind of machine, it's an
Aspire One and gears gives 740 fps, and video playback is pretty smooth.
But perhaps it would be faster if X could get that mtrr it claims ?

TIA

-- 
J.A. Magallon <jamagallon()ono!com>     \               Software is like sex:
                                         \         It's better when it's free
Mandriva Linux release 2009.1 (Cooker) for x86_64
Linux 2.6.28.2-desktop-1mnb (gcc 4.3.2 (GCC) #1 Wed Jan

Re: Intel driver vs MTRRs conflict

[H. Peter Anvin]

J.A. Magallón wrote:
> 
> As I said, everything works. And even fast for the kind of machine, it's an
> Aspire One and gears gives 740 fps, and video playback is pretty smooth.
> But perhaps it would be faster if X could get that mtrr it claims ?
> 

Which CPU?

	-hpa

Re: Intel driver vs MTRRs conflict

[J.A. Magallón]

On Tue, 17 Feb 2009 17:31:08 -0800, "H. Peter Anvin" <hpa@zytor.com> wrote:

> J.A. Magallón wrote:
> > 
> > As I said, everything works. And even fast for the kind of machine, it's an
> > Aspire One and gears gives 740 fps, and video playback is pretty smooth.
> > But perhaps it would be faster if X could get that mtrr it claims ?
> > 
>

Sorry for the delay...
 
> Which CPU?
> 

Atom N270, 1.60GHz, hyperthreaded. 512Kb cache

I have realized that I boot with video=intelfb:mtrr,ywrap.
Could this be interferring ?

Things in syslog I think are relevant:

Feb 17 23:45:47 one klogd: x86 PAT enabled: cpu 0, old 0x7040600070406, new 0x7010600070106
Feb 17 23:45:47 one klogd: original variable MTRRs
Feb 17 23:45:47 one klogd: reg 0, base: 4194176KB, range: 128KB, type WP
Feb 17 23:45:47 one klogd: reg 1, base: 4194048KB, range: 128KB, type UC
Feb 17 23:45:47 one klogd: reg 2, base: 0GB, range: 256MB, type WB
Feb 17 23:45:47 one klogd: reg 3, base: 256MB, range: 256MB, type WB
Feb 17 23:45:47 one klogd: reg 4, base: 504MB, range: 8MB, type UC
Feb 17 23:45:47 one klogd: reg 5, base: 502MB, range: 2MB, type UC
Feb 17 23:45:47 one klogd: reg 6, base: 501MB, range: 1MB, type UC
Feb 17 23:45:47 one klogd: reg 7, base: 0GB, range: 128KB, type UC
Feb 17 23:45:47 one klogd: total RAM coverred: 501M
Feb 17 23:45:47 one klogd: Found optimal setting for mtrr clean up
Feb 17 23:45:47 one klogd:  gran_size: 64K ^Ichunk_size: 16M ^Inum_reg: 4  ^Ilose cover RAM: 0G
Feb 17 23:45:47 one klogd: New variable MTRRs
Feb 17 23:45:47 one klogd: reg 0, base: 0GB, range: 512MB, type WB
Feb 17 23:45:47 one klogd: reg 1, base: 501MB, range: 1MB, type UC
Feb 17 23:45:47 one klogd: reg 2, base: 502MB, range: 2MB, type UC
Feb 17 23:45:47 one klogd: reg 3, base: 504MB, range: 8MB, type UC
Feb 17 23:45:47 one klogd: x86 PAT enabled: cpu 0, old 0x7040600070406, new 0x7010600070106
Feb 17 23:45:47 one klogd: kernel direct mapping tables up to 1f500000 @ 7000-c000
...
Feb 17 23:45:47 one klogd: vesafb: framebuffer at 0x20000000, mapped to 0xdfd80000, using 1875k, total 7872k
Feb 17 23:45:47 one klogd: vesafb: mode is 800x600x16, linelength=1600, pages=7
Feb 17 23:45:47 one klogd: vesafb: scrolling: redraw
Feb 17 23:45:47 one klogd: vesafb: Truecolor: size=0:5:6:5, shift=0:11:5:0
Feb 17 23:45:47 one klogd: Console: switching to colour frame buffer device 100x37
Feb 17 23:45:47 one klogd: fb0: VESA VGA frame buffer device


-- 
J.A. Magallon <jamagallon()ono!com>     \               Software is like sex:
                                         \         It's better when it's free
Mandriva Linux release 2009.1 (Cooker) for x86_64
Linux 2.6.27.5-desktop-1mnb (gcc 4.3.2 (GCC) #1 Sun Nov