[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <20220421051244.187733-1-mlevitsk@redhat.com>
Date: Thu, 21 Apr 2022 08:12:34 +0300
From: Maxim Levitsky <mlevitsk@...hat.com>
To: kvm@...r.kernel.org
Cc: Rodrigo Vivi <rodrigo.vivi@...el.com>,
Paolo Bonzini <pbonzini@...hat.com>,
intel-gfx@...ts.freedesktop.org,
Joonas Lahtinen <joonas.lahtinen@...ux.intel.com>,
Jani Nikula <jani.nikula@...ux.intel.com>,
Thomas Gleixner <tglx@...utronix.de>,
linux-kernel@...r.kernel.org, Wanpeng Li <wanpengli@...cent.com>,
Jim Mattson <jmattson@...gle.com>,
Tvrtko Ursulin <tvrtko.ursulin@...ux.intel.com>,
"H. Peter Anvin" <hpa@...or.com>,
Vitaly Kuznetsov <vkuznets@...hat.com>,
Zhi Wang <zhi.a.wang@...el.com>,
Daniel Vetter <daniel@...ll.ch>,
intel-gvt-dev@...ts.freedesktop.org,
dri-devel@...ts.freedesktop.org, x86@...nel.org,
David Airlie <airlied@...ux.ie>,
Sean Christopherson <seanjc@...gle.com>,
Ingo Molnar <mingo@...hat.com>, Joerg Roedel <joro@...tes.org>,
Dave Hansen <dave.hansen@...ux.intel.com>,
Borislav Petkov <bp@...en8.de>,
Zhenyu Wang <zhenyuw@...ux.intel.com>,
Maxim Levitsky <mlevitsk@...hat.com>
Subject: [RFC PATCH v2 00/10] RFCv2: nested AVIC
This patch series implement everything that is needed to
use AMD's AVIC while a nested guest is running including
ability of the nested guest to use it, and brings feature
parity vs APICv.
Compared to v1 of the series, there are lot of fixes,
and refactoring.
This version still use unconditional read-only apic id,
which will be addressed in the next version.
For the last patch, which allows to avoid cleaning is_running
bit in physid pages as long as it is possible, I measured
what would happen in a worst case:
- A malicious guest runs with 2 vCPUs pinned,
its first vCPU pounds on ICR sending IPIs to the 2nd vCPU
and 2nd vCPU scheduled out forever and not halted
(something that should not happen with Qemu though)
- A normal guest with single vCPU is pinned to run
on the same CPU as the 2nd vCPU of the first guest.
The normal guest continued to run, but was observed to run
about 40% slower.
Therefore AVIC doorbel is strict by default but if the admin
policy is to pin guests and not allow them to share a physical
cpu, then strict doorbel can be set to false and that does
improves the nested (and not nested as well) AVIC perf futher.
Suggestions, comments are welcome.
Best regards,
Maxim Levitsky
Maxim Levitsky (10):
KVM: x86: mmu: allow to enable write tracking externally
x86: KVMGT: use kvm_page_track_write_tracking_enable
KVM: x86: mmu: add gfn_in_memslot helper
KVM: x86: mmu: tweak fast path for emulation of access to nested NPT
pages
KVM: x86: lapic: don't allow to change APIC ID when apic acceleration
is enabled
KVM: x86: SVM: remove avic's broken code that updated APIC ID
KVM: x86: SVM: move avic state to separate struct
KVM: x86: rename .set_apic_access_page_addr to reload_apic_access_page
KVM: nSVM: implement support for nested AVIC
KVM: SVM: allow to avoid not needed updates to is_running
arch/x86/include/asm/kvm-x86-ops.h | 2 +-
arch/x86/include/asm/kvm_host.h | 5 +-
arch/x86/include/asm/kvm_page_track.h | 1 +
arch/x86/kvm/Kconfig | 3 -
arch/x86/kvm/lapic.c | 28 +-
arch/x86/kvm/mmu.h | 8 +-
arch/x86/kvm/mmu/mmu.c | 21 +-
arch/x86/kvm/mmu/page_track.c | 10 +-
arch/x86/kvm/svm/avic.c | 949 ++++++++++++++++++++++++--
arch/x86/kvm/svm/nested.c | 131 +++-
arch/x86/kvm/svm/svm.c | 31 +-
arch/x86/kvm/svm/svm.h | 165 ++++-
arch/x86/kvm/trace.h | 140 +++-
arch/x86/kvm/vmx/vmx.c | 8 +-
arch/x86/kvm/x86.c | 17 +-
drivers/gpu/drm/i915/Kconfig | 1 -
drivers/gpu/drm/i915/gvt/kvmgt.c | 5 +
include/linux/kvm_host.h | 10 +-
18 files changed, 1413 insertions(+), 122 deletions(-)
--
2.26.3
Powered by blists - more mailing lists