Index | Thread | Search

From:
Miguel Landaeta <miguel@miguel.cc>
Subject:
Re: vmm(4): return error for VMCALL instead of injecting #UD
To:
tech@openbsd.org
Date:
Mon, 13 Apr 2026 21:44:14 +0000

Download raw body.

Thread
  • Miguel Landaeta:

    vmm(4): return error for VMCALL instead of injecting #UD

On Tue, Apr 07, 2026 at 12:45:01PM +0000, hshoexer wrote:
> Hi,
> 
> here is reworked version based on the original diff:  I've implemented
> a function vcpu_get_insnlen() to get the instruction length for
> both VMX and SVM based virtualization.  It could be used in several
> places where we retrieve the instruction length.  For now, I only
> use it for handling vmcall/vmmcall.
> 
> ok?

Hi,

FWIW, I tested this diff on Intel hardware, and it addresses the issue
I reported here:

https://marc.info/?l=openbsd-tech&m=177366472302263&w=2

VMCALL instructions no longer trigger panics or crashes in Linux guests
such as Talos.

I can't comment on AMD CPUs yet, as I don't currently have one available
for testing, but I'll try to borrow one soon and test it there as well.

Thanks for addressing this issue.
Miguel