Bug 39845

Please file all bugs regarding the binary-only NVidia driver to NVidia

This is *not* a nvidia related bug as it turns out it still exists when I shut
down X and unload the NVdriver module. I can empty the mtrr register:

echo "disable=reg00" > /proc/mtrr

/proc/mtrr is empty now and the system is *slow*.

I enter this to mtrr:

echo "base=0x00000000 size=0x10000000 type=write-back" > /proc/mtrr

/proc/mtrr now has this to say:

reg00: base=0x00000000 (   0MB), size=16773376MB: write-back, count=1

This is without any nvidia related binaries or anything else loaded.

Please retry without the nvidia modules EVER being loaded since the reboot.

I have now performed a reboot and without ever having started the nvidia module
the problem still persists. This is what I did just after the reboot:

[root@overmind /root]# cat /proc/mtrr
reg00: base=0x00000000 (   0MB), size=16711936MB: write-back, count=1
[root@overmind /root]# echo "disable=reg00" > /proc/mtrr
[root@overmind /root]# echo "base=0x00000000 size=0x10000000 type=write-back" >
/proc/mtrr
[root@overmind /root]# cat /proc/mtrr
reg00: base=0x00000000 (   0MB), size=16773376MB: write-back, count=1
[root@overmind /root]# lsmod
Module                  Size  Used by
nfsd                   69696   8  (autoclean)
lockd                  52464   1  (autoclean) [nfsd]
sunrpc                 61328   1  (autoclean) [nfsd lockd]
libafs-2.4.2-2        381264   2
autofs                 11264   2  (autoclean)
8139too                16480   1  (autoclean)
ipchains               38976   0  (unused)
ide-scsi                8352   0
scsi_mod               95104   1  [ide-scsi]
ide-cd                 26848   0
cdrom                  27232   0  [ide-cd]
nls_iso8859-1           2880   2  (autoclean)
nls_cp437               4384   2  (autoclean)
vfat                    9392   2  (autoclean)
fat                    32672   0  (autoclean) [vfat]

Any ideas?

Is it correct that you have 256Mb of ram ?

Yes, that is correct.

Sometimes, e.g. starting X, I see the same "mtrr: type mismatch" messages on 
Athlon MS-6195 motherboard (AMD-751/756) with Rage 128 pro AGP card. Do you need
more information?

I just tried on my home athlon machine, and it works there. Although I have
768Mb of ram... This is a strange problem.

I have just installed the rawhide kernel, 2.4.3-2.14.14, rebooted and now MTRR
works as expected. Now alsasound cannot be recompiled
(/usr/src/linux-2.4.3/include/asm/pgalloc-2level.h is missing). kernel-headers
are also too old.

"pgalloc-2level.h" is no longer present in the kernel. 
It is really surprising ALSA needs this header as it is a very very architecture
specific file and normal drivers shouldn't need to include it.
As this is a generic kernel change, I would not be surprised if a newer ALSA
driver is released that fixes this.

I forgot a make clean in alsasound and that caused the problem. alsasound works
just fine with kernel-2.4.3-2.14.14.

This problem seems to have been a symptom of changes to the MTRR support in 2.4

The fix is mtrr.c version 1.39 which contains the comment

  v1.39
    20010312   Dave Jones <davej>
               Ugh, I broke AMD support.
               Reworked fix by Troels Walsted Hansen <troels>

This is included in 2.4.3 kernels from RawHide so all AMD users seeing errors
like this should probably consider upgrading.

This was an initialisation problem in the mtrr coding.
For current Athlon CPUs it was pretty likely that the
and/or mask values didnt get initialized and therfore
gigantic size values got composed from CPU registers.

The bug is fixed in official Linux Kernel 2.4.4 and later.