foxgrrl | VMWare

You're viewing

foxgrrl's journal
Create a Dreamwidth Account Learn More

Reload page in style: site light

So, am I just on crack, or are the last three releases of VMWare Workstation totally broken with respect to the hardware debugging registers? If I set a hardware breakpoint [like on memory write], in a debugger, in the guest OS; It will break on every single instruction, even if it doesn't touch memory at all (like XOR EAX,EAX (in a completely different segment than DR0 or anything))

For reference, this is VMWare Workstation 5.5.2 on a Host OS: Linux x86-64, Guest OS Win2K Adv Server. Same thing with VMWare WS 5.5 and 5.0 too.

I can't find anything on Google about this; I can't be the only person on Earth who has noticed this.

Current Mood: sleepy
Current Location: In Bed

Flat | Top-Level Comments Only

From:

foxgrrl

I know at least three or four furries who work at VMWare. And maybe a few more people. I had a recruiter there last year, who was courting me for some kind of kernel hacking or low level bit-banging position, but I went with my current employer instead.

I just wanted to make sure that I wasn't mistaken before I filed a bug with VMWare. I wouldn't be surprised if VMWare itself was using the debugging registers as part of its hardware virtualization, and hence, the weird problems with them inside the Guest OS. (There's a trick for doing virtualization on Sparc, by setting the memory pages of a process to non-readable, not-writable, and non-executable, patching up the executable's problematic instructions with emulated equivalents, and then setting the page back to executable, and catching any faults if the code tries to read itself or modify itself, VMM does it's thing to make things look normal to the process, and then lets it run some more, blah blah blah... anyway, on x86 this doesn't quite work because the non-exec bit on a page's descriptor is not honored. But, I speculate, that you can work around this with some clever usage of the hardware debug registers. (I know I should really look into what Solar Designer was doing with PaX one of these days...) AMD fixed this bug with x86 finally with x86-64, except that Intel failed to copy this on the early EM64Ts.)