Update: IOMMU on P151HM for VT-d PCI passthrough

A couple weeks ago I was asking about BIOS updates that allowed VT-d options, to be told Intel's website is correct that this chipset cannot support these functions, despite having seen a report from someone who ran Qubes on one of these computers.

That is incorrect. No settings changes are required, and DMAR was configured by Sager. Simply put in a 2820QM or better and enable IOMMU in your Linux kernel's boot parameters.

The extra full-length mPCIe slot is all alone, and looks like it's two channels (but my eGPU setup is only wired for one)

Code:

[    1.519409] DMAR: Hardware identity mapping for device 0000:06:00.0
[    1.519411] DMAR: Hardware identity mapping for device 0000:06:00.1
[    1.519412] DMAR: Setting RMRR:
[    1.519414] DMAR: Ignoring identity map for HW passthrough device 0000:00:1a.0 [0xcf543000 - 0xcf551fff]
[    1.519415] DMAR: Ignoring identity map for HW passthrough device 0000:00:1d.0 [0xcf543000 - 0xcf551fff]
[    1.519417] DMAR: Prepare 0-16MiB unity mapping for LPC
[    1.519419] DMAR: Ignoring identity map for HW passthrough device 0000:00:1f.0 [0x0 - 0xffffff]
[    1.519420] DMAR: Intel(R) Virtualization Technology for Directed I/O

My dual GTX770 shows up at device 06:00.0 and its audio at 06:00.1.

All I have remaining to solve for the passthrough is the Windows Code 43 error, which is just a configuration issue.

Onto my new question: Does anyone have a working DSDT with a 56-bit address space for a P151HM? The decompilers I've tried don't work quite right, so I can recompile, but I get buffer underflows when I go to use it. I'd like to try running the card on a non-virtualized Windows install.

 

That's because Clevo's DSDT is so full of errors that you won't get around fixing a bunch to get it to re-compile.

 

I decided the DSDT editing was a waste of my time, and this "impossible" setup was more likely to work. It does!


First run, QEMU inside Manjaro:
3DMark11 GPU score: 8300
CPU score: 5300 (No optimizations besides pinning yet)
-4 cores, no hyperthreading, pinned to 4, 5, 6, 7 (Physical cores)
-8GB memory, no static hugepages
-VirtIO SCSI raw .img drive
-Geforce 331.40 mobile drivers (edited the .inf for a 770M)- I think I'm configured to move to better drivers without NVidia detecting the VM, but we'll see.

 

Looks good. I've found KVM + VFIO works perfectly well. My HTPC operates exactly like that.

Be careful with CPU-pinning, especially if you don't have enough physical cores to 100% dedicate to the VM. You can end up with some crazy DPC issues because the host no longer load shares properly.

As far as drivers under KVM, assuming you have the Nvidia workaround present, you can use all the way up to the latest drivers. On a related note, GPU overclocking should work in the VM as well.

 

Amazing !!!

Could you post a tutorial of what you did?, it seems you are the first one to make work a pci passthrough in a laptop