Re: [PATCH v2 0/8] KVM: x86: Emulator _regs fixes and cleanups

[Robert Dinse]

     It >DID< build ok with this patch, I applied it and built last night,
I'm presently running that kernel with a KVM/Qemu Windows guest on my
workstation:

Linux nanook 5.18.0 #1 SMP PREEMPT Wed May 25 18:14:43 PDT 2022 x86_64 x86_64 
x86_64 GNU/Linux

     The Windows guest starts and runs but networking isn't working (I am
using a bridge for the guest hosts and a static IP for both the main machine
and guest), it all works including networking with 5.17.9.

-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-
 Eskimo North Linux Friendly Internet Access, Shell Accounts, and Hosting.
   Knowledgeable human assistance, not telephone trees or script readers.
 See our web site: http://www.eskimo.com/ (206) 812-0051 or (800) 246-6874.

On Thu, 26 May 2022, Sean Christopherson wrote:

> Date: Thu, 26 May 2022 21:08:09 +0000
> From: Sean Christopherson <seanjc@xxxxxxxxxx>
> To: Paolo Bonzini <pbonzini@xxxxxxxxxx>
> Cc: Sean Christopherson <seanjc@xxxxxxxxxx>,
>     Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>,
>     Wanpeng Li <wanpengli@xxxxxxxxxxx>, Jim Mattson <jmattson@xxxxxxxxxx>,
>     Joerg Roedel <joro@xxxxxxxxxx>, kvm@xxxxxxxxxxxxxxx,
>     linux-kernel@xxxxxxxxxxxxxxx, Kees Cook <keescook@xxxxxxxxxxxx>,
>     Robert Dinse <nanook@xxxxxxxxxx>
> Subject: [PATCH v2 0/8] KVM: x86: Emulator _regs fixes and cleanups
>
> Clean up and harden the use of the x86_emulate_ctxt._regs, which is
> surrounded by a fair bit of magic.  This series was prompted by bug reports
> by Kees and Robert where GCC-12 flags an out-of-bounds _regs access.  The
> warning is a false positive due to a now-known GCC bug, but it's cheap and
> easy to harden the _regs usage, and doing so minimizing the risk of more
> precisely handling 32-bit vs. 64-bit GPRs.
>
> I didn't tag patch 2 with Fixes or Cc: stable@.  It does remedy the
> GCC-12 warning, but AIUI the GCC-12 bug affects other KVM paths that
> already have explicit guardrails, i.e. fixing this one case doesn't
> guarantee happiness when building with CONFIG_KVM_WERROR=y, let alone
> CONFIG_WERROR=y.  That said, it might be worth sending to the v5.18 stable
> tree[*] as it does appear to make some configs/setups happy.
>
> [*] KVM hasn't changed, but the warning=>error was introduced in v5.18 by
>   commit e6148767825c ("Makefile: Enable -Warray-bounds").
>
> v2:
>  - Collect reviews and tests. [Vitaly, Kees, Robert]
>  - Tweak patch 1's changelog to explicitly call out that dirty_regs is a
>    4 byte field. [Vitaly]
>  - Add Reported-by for Kees and Robert since this does technically fix a
>    build breakage.
>  - Use a raw literal for NR_EMULATOR_GPRS instead of VCPU_REGS_R15+1 to
>    play nice with 32-bit builds. [kernel test robot]
>  - Reduce the number of emulated GPRs to 8 for 32-bit builds.
>  - Add and use KVM_EMULATOR_BUG_ON() to bug/kill the VM when an emulator
>    bug is detected.  [Vitaly]
>
> v1: https://lore.kernel.org/all/20220525222604.2810054-1-seanjc@xxxxxxxxxx
>
> Sean Christopherson (8):
>  KVM: x86: Grab regs_dirty in local 'unsigned long'
>  KVM: x86: Harden _regs accesses to guard against buggy input
>  KVM: x86: Omit VCPU_REGS_RIP from emulator's _regs array
>  KVM: x86: Use 16-bit fields to track dirty/valid emulator GPRs
>  KVM: x86: Reduce the number of emulator GPRs to '8' for 32-bit KVM
>  KVM: x86: Bug the VM if the emulator accesses a non-existent GPR
>  KVM: x86: Bug the VM if the emulator generates a bogus exception
>    vector
>  KVM: x86: Bug the VM on an out-of-bounds data read
>
> arch/x86/kvm/emulate.c     | 26 ++++++++++++++++++++------
> arch/x86/kvm/kvm_emulate.h | 28 +++++++++++++++++++++++++---
> arch/x86/kvm/x86.c         |  9 +++++++++
> 3 files changed, 54 insertions(+), 9 deletions(-)
>
>
> base-commit: 90bde5bea810d766e7046bf5884f2ccf76dd78e9
> --
> 2.36.1.255.ge46751e96f-goog