From mboxrd@z Thu Jan 1 00:00:00 1970 From: Paolo Bonzini Subject: Re: [PATCH] arm64: KVM: export current vcpu->pause state via pseudo regs Date: Thu, 31 Jul 2014 19:21:44 +0200 Message-ID: <53DA7B28.701@redhat.com> References: <1404914112-7298-1-git-send-email-alex.bennee@linaro.org> <53DA7583.2030504@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: =?UTF-8?B?QWxleCBCZW5uw6ll?= , "kvmarm@lists.cs.columbia.edu" , arm-mail-list , kvm-devel , Catalin Marinas , Will Deacon , open list , Gleb Natapov To: Peter Maydell Return-path: In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org List-Id: kvm.vger.kernel.org Il 31/07/2014 19:04, Peter Maydell ha scritto: > On 31 July 2014 17:57, Paolo Bonzini wrote: >> Il 09/07/2014 15:55, Alex Benn=C3=A9e ha scritto: >>> To cleanly restore an SMP VM we need to ensure that the current pau= se >>> state of each vcpu is correctly recorded. Things could get confused= if >>> the CPU starts running after migration restore completes when it wa= s >>> paused before it state was captured. >>> >>> I've done this by exposing a register (currently only 1 bit used) v= ia >>> the GET/SET_ONE_REG logic to pass the state between KVM and the VM >>> controller (e.g. QEMU). >>> >>> Signed-off-by: Alex Benn=C3=A9e >>> --- >>> arch/arm64/include/uapi/asm/kvm.h | 8 +++++ >>> arch/arm64/kvm/guest.c | 61 +++++++++++++++++++++++++++= +++++++++++- >>> 2 files changed, 68 insertions(+), 1 deletion(-) >> >> Since it's a pseudo register anyway, would it make sense to use the >> existing KVM_GET/SET_MP_STATE ioctl interface? >=20 > That appears to be an x86-specific thing relating to > IRQ chips. No, it's not. It's just the state of the CPU, s390 will be using it to= o. On x86 the states are uninitialized (UNINITIALIZED), stopped (INIT_RECEIVED), running (RUNNABLE), halted (HALTED). CPU 0 starts in RUNNABLE state, other CPUs start in UNINITIALIZED state. There are x86-specific cases (uninitialized) and x86-isms (the INIT_RECEIVED name), but the idea is widely applicable. >> Also, how is KVM/ARM >> representing (and passing to QEMU) the halted state of the >> VCPU? >=20 > We don't. In ARM the equivalent of x86 HLT (which is > WFI, wait-for-interrupt) is allowed to resume at any time. > So we don't need to care about saving and restoring > whether we were sat in a WFI at point of migration. What does ARM do if you have a WFI while interrupts are disabled? On x86 after "cli;hlt" only an NMI will wake you up. With spurious wakeups, it's pretty much guaranteed that you will break such "cli;hlt" sequences. Paolo