On Sun, 2012-10-07 at 17:06 +0200, Ingo wrote: > Interesting to see the differences between the 2 kernels. Compared to > my messages I found following lines in dmesg have disappeared with 3.5: > > mtrr: type mismatch for e0000000,10000000 old: write-back new: > write-combining > > [drm] MTRR allocation failed. Graphics performance may suffer. > [drm:intel_dsm_platform_mux_info] *ERROR* MUX INFO call failed > > i915 0000:00:02.0: PCI INT A -> GSI 16 (level, low) -> IRQ 16 > i915 0000:00:02.0: restoring config space at offset 0xf (was 0x100, > writing 0x10b) > i915 0000:00:02.0: restoring config space at offset 0x1 (was 0x900007, > writing 0x900407) > i915 0000:00:02.0: setting latency timer to 64 > > while enabling RC6 states has been added: > > [drm] Enabling RC6 states: RC6 on, RC6p on, RC6pp off > > > Probably a hint to the root cause? Don't think so. > I also compared the BIOS settings: > > Javier has these options/settings: > Integrated Graphics Share Memory [64MB] > DVMT Memory [256MB] > > I only have: > IGD DVMT Memory [256MB] > > (which I called previously "agp-aperture" - ancient name when PC's had > a dedicated AGP-Port). This is the setting which I did change from > "max" -> "256MB" and got my system stable - which still holds right now. > > However the "mtrr" setups (/proc/mtrr) are still identical with both > kernels. Seems the "write-combining area" has become obsolete - or > still not implemented in i915? The driver calls ioremap_wc() which will enable write-combining through the PAT in recent processors. An MTRR is not needed and in 3.5 the driver doesn't even bother to allocate one: commit 9e984bc1dffd405138ff22356188b6a1677c64c8 Author: Adam Jackson <ajax@redhat.com> Date: Wed Mar 14 11:22:11 2012 -0400 drm/i915: Don't do MTRR setup if PAT is enabled Ben. -- Ben Hutchings You can't have everything. Where would you put it?
Attachment:
signature.asc
Description: This is a digitally signed message part