For MTRR error, I think the kernel is not happy with the uncachable entry for the memory hole.
Try to install only 2G to see if the problem will go away.
Then we can
- modify the kernel to make it like mtrr layout for setting.
- or change mtrr to increase style..., if the kernel is too ..
Yep, that's the ticket. All remaining is the message: PCI: No IRQ known for interrupt pin C of device 0000:40:01.0. Probably buggy MP table. which seems plausibly ignorable since it seems to actually be getting interrupts.
sh-3.1# cat /proc/mtrr reg00: base=0x00000000 ( 0MB), size=2048MB: write-back, count=1 reg01: base=0xfd500000 (4053MB), size= 1MB: write-combining, count=1
fd400000-fd5fffff : 0000:40:01.0 fd400000-fd5fffff : ib_ipath
Given that, how do you want to go forward? I've got compute nodes with nothing but the HTX installed and storage nodes with HTX and a couple PCI-X 3ware cards, so only two cases for me admits hardcoding hacks. Of course, it would be nice to solve the problem in general and call the DK8-HTX 'working' and I'm happy to spend effort to get there rather than the hack. :)
Best,
-mcq