[Fixed] Windows 7 64 bit guest memory corruption

[Ramshankar]

Here is a Windows test build which contains changes which hopefully fix your problem. Could you verify? Thank you!

Make sure you test with large pages on (just a gently reminder, as I often tend to forget reverting internal VM settings)

[mhanor]

It seems fine with the new build, on Windows and Ubuntu as host OS. Were you able to reproduce the issue?

[Ramshankar]

It seems fine with the new build, on Windows and Ubuntu as host OS. Were you able to reproduce the issue?

Good to hear.

No, I was not able to reproduce it on a release build, but on a debug build it triggers an assertion in the PGM code path while freeing large pages used by Hyper-V while mapping the overlay MMIO2 TSC/hypercall page. I've ran into this assertion before but wasn't aware of the exact implications of ignoring it (as happens on release builds).

I fixed/worked-around it to directly rewrite guest memory directly rather than perform the complicated MMIO2 overlay that the Hyper-V spec wants (as PGM can't handle that yet). The guest shouldn't be trying funny tricks with the page as the Hyper-V spec states will lead to hypervisor specific behavior anyway, and as far as I checked, no guests do. They just use a free page and don't really rely on the overlay mapping to work when unmapped/moved.

[mhanor]

Just to let you know, I have discovered that my PC was running with a bad RAM DIMM (1 bit always set in a byte near the top of the 8GB of RAM). I'm not sure how it affected the Windows 7 guest or the host. I hope that the changes you made (through changeset 56672) do have a good reasoning, so it wasn't influenced by something that could have been caused by the hardware problem. I'm only puzzled that this was the only guest which had a bad behaviour. I will try to reproduce the issue with the old builds once I get the replacement RAM.

[mhanor]

With the new RAM modules and the older beta builds, I can reproduce the guru meditation. No BSODs or file integrity errors, though.