Re: Please revert / review 077fc1c04d70ef1748ac2daa6622b3320a1a004c

["Zhang, Yang Z" <yang.z.zhang@intel.com>]

Matthias wrote on 2014-06-20:
> Did some testing with your patches:
> 
> 
> Test 1:
> 
> 
> Xen-Version: Compiled a xen with the 077 commit, your first patch and your 3.
> (latest) patch for testing.
> 
> Passed-through devices: only vga
> 
> Result: At some late point in the boot of the domU, the dom0 (the 
> server itself) restarted itself. This is repeatable. I added the 
> qemu-log and what I was able to capture from the xl dmesg (I only hit 
> 'xl dmesg' numerous times until the server restarted and might not 
> have gotten the end of it, but since there is no xl dmesg log i'm 
> aware off, I couldn't think of another way. If you have any, please 
> tell me). From the qemu log  I would guess the restart was caused by some problem in the vram mapping?

I am a little confused: Previously, you said the failure is caused by the IOMMU fault:

(XEN) AMD-Vi: IO_PAGE_FAULT: domain = 1, device id = 0x500, fault address = 0x1d27d0000, flags = 0

But now, it seems you are encountering another problem. What's the difference between the two testings? The patch 1 and patch 3 only add some debug codes. They would not affect your testing. 

Also, I would suggest you to enable the serial console and get the log from console.

> 
> 
> Test 2:
> 
> 
> Xen-Version: Compiled a normal 077 commit version without any of your 
> patches
> 
> Xen cmd line: added immo=no-sharept as suggested
> 
> Passed-through devices: vga, 2 usb controller (ohci+ehci), sound card
> Result: works fine. domU boots, no BSOD, no other problems, server 
> seems to be stable. So not sharing the VT-d page tables seems to do 
> the trick / the problem is in the page table sharing.
>

Yes, this is what I am expecting.

Best regards,
Yang