MTRRs not working in 2.6 linux kernels?

[Felix Kühling]

On Sun, 3 Oct 2004 22:33:28 -0400
Jonathan Thambidurai <jonthambi at comcast.net> wrote:

> When I try to use the application tvtime under Xorg 6.8.1 with Linux
> kernel 2.6.7, I get very choppy framerates; tvtime indicates that it is
> only getting about 50 MB/s for blitting to the screen.  With the same
> version of Xorg and Linux kernel 2.4.22, I get smooth play, and tvtime
> indicates that it gets about 150 MB/s for blitting to screen.  I made
> sure DRI wasn't playing a role in this by rmmod'ing the agpgart module
> before starting X.  
> 
> Of note is the difference in /proc/mtrr between 2.6 and 2.4
> kernels:
> 
> in 2.4, it contains only
> 
> reg00: base=0x00000000 (   0MB), size= 256MB: write-back, count=1
> reg01: base=0x0ff80000 ( 255MB), size= 512KB: uncachable, count=1
> 
> while in 2.6 it contains
> 
> reg00: base=0x00000000 (   0MB), size= 256MB: write-back, count=1
> reg01: base=0x0ff80000 ( 255MB), size= 512KB: uncachable, count=1
> reg02: base=0xe8000000 (3712MB), size=   2MB: write-combining, count=1
> 
> Lastly, in the X.org log when running a 2.6 kernel the following line is
> displayed:
> 
> (WW) RADEON(0): Failed to set up write-combining range (0xe8000000,0x1000000)
> 
> while under 2.4 the following line is displayed:
> 
> (==) RADEON(0): Write-combining range (0xe8000000,0x1000000)

This fails because someone else set up write combining for the same
start address but only 2MB long. Usually that would be a kernel frame
buffer driver that doesn't detect the correct frame buffer size
(vesafb?). I think there an option "nomtrr" you can pass to the
radeonfb/vesafb driver.

HTH,
  Felix

> 
> I am running a P4 at 1.8 Ghz Laptop with an AGP 4x Radeon 7500 and 16 MB
> of VRAM; it has 256 MB of system memory, which makes the third line of 
> /proc/mtrr in the 2.6 kernels look wrong to my untrained eyes.
> 
> I just tried this on kernel 2.6.9-rc2 with the same results
> as 2.6.7 and saw that dmesg complained about some mtrr overlapping with 
> something else; if this is important, I can report the exact message.
> 
> Perhaps this is a kernel issue and should be posted to the kernel list? 
> Any help would be appreciated.
> 
> --Jonathan Thambidurai

| Felix Kühling <fxkuehl at gmx.de>                     http://fxk.de.vu |
| PGP Fingerprint: 6A3C 9566 5B30 DDED 73C3  B152 151C 5CC1 D888 E595 |