Vincent ETIENNE <[EMAIL PROTECTED]> posted
[EMAIL PROTECTED], excerpted below, on  Thu, 30 Aug 2007
20:49:40 +0200:

>> This probably won't help you directly, but here, running 8 gig memory
>> on a dual Opteron Tyan (s2885) mobo, original AMD 8000 chipset, with a
>> Radeon 9200SE, with the native kernel and xorg drivers, DRM works fine,
>> even tho it /does/ protest about running out of MTRRs.
> 
> Very interesting indeed, i don't know that it was possible to have DRM
> without MTRR : will have to found some documentation

Well, that's the thing.  I'm not sure it's running without /any/.  It may 
have some, but not what it would like.  I don't even know if that makes 
sense, but that's where I am in knowledge, or more precisely lack thereof.

> Yes don't seem easy to found a more or less complete and reliable
> documentation source. For the momemnt i have only found some sparse
> information in a forum about rage 3D card. But none up to date. For
> reference here is the link
>       http://www.rage3d.com/board/showthread.php?threadid=33736241
> and http://www.rage3d.com/board/showthread.php?t=33821469

Thanks.
 
> I have try  with only 2G ram. Everything seems to work ( DRM is
> activated, X11 start,  open gl crash as soon as an application use it...
>  but it may be unrelated )

Something to note.  Again it may differ on your hardware, but here, I had 
to disable "fast writes" in the BIOS to get DRM working.  Also, with < 4 
gig RAM, SATA access functioned, with > 4 gig, it did with early 2.6 
kernels but stopped around (IIRC) 2.6.18, until I figured out that "fast 
writes" was the problem there as well.  That's actually what got me to 
turn it off.  Before that, DRM simply didn't work, I didn't know why, but 
when I couldn't boot, I /had/ to figure out why, and it was the fast 
writes.

The problem is related to the 32-bit PCI memory hole at the top of 4 gig 
memory.  On at least AMD hardware, there's a hardware IOMMU that allows 
PCI DMA to memory beyond that.  Something about fast writes disabled that 
IOMMU, and before the kernel that broke, it was using a software bounce-
buffer instead.  Eventually, the bounce-buffer was disabled where they 
hardware IOMMU was expected, causing SATA disk access to break.  Since 
that's what I boot off of... .  Anyway, once I disabled fast writes, the 
hardware IOMMU started working and I got the disks back.  Something in 
the research done to figure out what was going on with that hinted to me 
that DRM might work now as well (not unreasonable, since the hardware 
IOMMU uses the AGP memory portal, what do they call it?), and sure enough.

Intel doesn't have that IOMMU so must use the bouncebuffer anyway, so I'm 
not sure how much of this relates to it, but try disabling fast writes 
anyway, and see if it makes a difference.

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman

-- 
[EMAIL PROTECTED] mailing list

Reply via email to