Re: [Qemu-devel] [PATCH v2 5/8] kvmvapic: Introduce TPR access optimization for Windows guests

[Gleb Natapov <gleb@redhat.com>]

On Mon, Feb 13, 2012 at 06:50:08PM +0000, Blue Swirl wrote:
> On Mon, Feb 13, 2012 at 10:16, Jan Kiszka <jan.kiszka@siemens.com> wrote:
> > On 2012-02-11 16:25, Blue Swirl wrote:
> >> On Fri, Feb 10, 2012 at 18:31, Jan Kiszka <jan.kiszka@siemens.com> wrote:
> >>> This enables acceleration for MMIO-based TPR registers accesses of
> >>> 32-bit Windows guest systems. It is mostly useful with KVM enabled,
> >>> either on older Intel CPUs (without flexpriority feature, can also be
> >>> manually disabled for testing) or any current AMD processor.
> >>>
> >>> The approach introduced here is derived from the original version of
> >>> qemu-kvm. It was refactored, documented, and extended by support for
> >>> user space APIC emulation, both with and without KVM acceleration. The
> >>> VMState format was kept compatible, so was the ABI to the option ROM
> >>> that implements the guest-side para-virtualized driver service. This
> >>> enables seamless migration from qemu-kvm to upstream or, one day,
> >>> between KVM and TCG mode.
> >>>
> >>> The basic concept goes like this:
> >>>  - VAPIC PV interface consisting of I/O port 0x7e and (for KVM in-kernel
> >>>   irqchip) a vmcall hypercall is registered
> >>>  - VAPIC option ROM is loaded into guest
> >>>  - option ROM activates TPR MMIO access reporting via port 0x7e
> >>>  - TPR accesses are trapped and patched in the guest to call into option
> >>>   ROM instead, VAPIC support is enabled
> >>>  - option ROM TPR helpers track state in memory and invoke hypercall to
> >>>   poll for pending IRQs if required
> >>>
> >>> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
> >>
> >> I must say that I find the approach horrible, patching guests and ROMs
> >> and looking up Windows internals. Taking the same approach to extreme,
> >> we could for example patch Xen guest to become a KVM guest. Not that I
> >> object merging.
> >
> > Yes, this is horrible. But there is no real better way in the absence of
> > hardware assisted virtualization of the TPR. I think MS is recommending
> > this patching approach as well.
> 
> Maybe instead of routing via ROM and the hypercall, the TPR accesses
> could be handled directly with guest invisible breakpoints (like GDB
> breakpoints, but for QEMU internal use), much like other
> instrumentation could be handled.
> 
Hypercall is rarely called. The idea behind patching is to not
have exit on each TPR update. Breakpoint will cause exit making the
whole exercise pointless.

--
			Gleb.