http://bugzilla.novell.com/show_bug.cgi?id=515352
User libv@skynet.be added comment
http://bugzilla.novell.com/show_bug.cgi?id=515352#c3
Luc Verhaegen changed:
What |Removed |Added
----------------------------------------------------------------------------
Summary|xf86-video-unichrome: |xf86-video-unichrome:
|VT3108 support broken on |mapping FB above known ram
|K8M800 Socket AM2 |broken.
--- Comment #3 from Luc Verhaegen 2009-08-28 06:07:18 MDT ---
This problem exists on all chipsets with hypertransport based AMD cpus.
Why: The FB sits in main RAM which sits off the CPUs memory controller. Any
normal access to the unichrome's announced RAM means that accesses go from the
cpu, over HT, to the northbridge, hits the memory controller, which sends the
access straight back over the same HT link to the CPU, which then sends the
memory access out the other side of the CPU on to its ram controller into the
RAM.
We cut short this path and send access straight to the ram controller. This
makes the driver a lot faster.
But for that we have to map a part of main RAM that neither the kernel nor the
PCI device think is there. Due to improved checking in newer kernels, this
mapping is no longer allowed.
Will hopefully have a solution in the next few hours that will retain this
clean, fast path to memory.
--
Configure bugmail: http://bugzilla.novell.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.