Myles Watson escribió:
It turns out that the problem is before the print statements I inserted. For some reason, your domain's IO resource is only giving you 0x7ff of IO space. That's not enough. I can't see anywhere in the code where that would be coming from, so I'm assuming memory corruption. At this point the limit should still be 0000ffff. Could you try increasing the stack and heap sizes? If that doesn't work we'll have to find where the limit is getting changed, since it's correct the first time that the resources are printed. PNP: 002e.b resource base 290 size 8 align 3 gran 3 limit fff
flags c0000100 index 60 PNP: 002e.b resource base 5 size 1 align 0 gran 0 limit 0 flags c0000400 index 70
PNP: 002e.3 links 0 child on link 0 NULL PNP: 002e.3 resource base 0 size 8 align 3 gran 3 limit 7ff flags
100 index 60 PNP: 002e.3 resource base 0 size 1 align 0 gran 0 limit 0 flags 400 index 70 ... PCI: 00:06.2 20 * [0x1cf0 - 0x1cff] io PNP: 002e.3 60 * [0x2000 - 0x2007] io
Never mind about the memory corruption. PNP: 002e.3 has some resources that aren't "fixed", so the allocator is trying to allocate them. In the SuperIO code where those resources are declared, set the flags to match the other resources.
Problem solved I killed it :) means I disabled it in the device tree. The H9DMR board has two serials and my board H8QME has only one.
Now, I've got till the MTRR initialization and here is my reward:
POST: 0x93 Setting up local apic... apic_id: 0x0a done. POST: 0x9b CPU model: Quad-Core AMD Opteron(tm) Processor 8350 siblings = 03, CPU #10 initialized All AP CPUs stopped PCI: 00:18.0 init PCI: 00:02.0 init Unexpected Exception: 6 @ 10:00207ff1 - Halting Code: 0 eflags: 00010046 eax: 00000000 ebx: 00226358 ecx: 0021ea74 edx: 00000001 edi: 0021ea74 esi: 00000001 ebp: 0023fff4 esp: 0023ff50 POST: 0xff
the strange thing is that it seems to work for like 11 other cores. And this log is from my last try but before it always starts rebooting with SOFT RESET after: MTRR check Fixed MTRRs : Enabled
Attached there is the whole log its getting bigger but still not till the end :(
THX, Knut Kujat.