Re: [PATCH 2/2] Revert "KVM: x86: zero kvmclock_offset when vcpu0 initializes kvmclock system MSR"

From: Marcelo Tosatti <mtosatti@redhat.com>
To: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>,
	linux-kernel@vger.kernel.org, kvm@vger.kernel.org,
	"Luiz Capitulino" <lcapitulino@redhat.com>
Subject: Re: [PATCH 2/2] Revert "KVM: x86: zero kvmclock_offset when vcpu0 initializes kvmclock system MSR"
Date: Tue, 22 Sep 2015 17:23:47 -0300	[thread overview]
Message-ID: <20150922202345.GA29417@amt.cnet> (raw)
In-Reply-To: <5601B191.4040809@redhat.com>

On Tue, Sep 22, 2015 at 09:52:49PM +0200, Paolo Bonzini wrote:
> 
> 
> On 22/09/2015 21:01, Marcelo Tosatti wrote:
> > On Fri, Sep 18, 2015 at 05:54:30PM +0200, Radim Krčmář wrote:
> >> Shifting pvclock_vcpu_time_info.system_time on write to KVM system time
> >> MSR is a change of ABI.  Probably only 2.6.16 based SLES 10 breaks due
> >> to its custom enhancements to kvmclock, but KVM never declared the MSR
> >> only for one-shot initialization.  (Doc says that only one write is
> >> needed.)
> >>
> >> This reverts commit b7e60c5aedd2b63f16ef06fde4f81ca032211bc5.
> >> And adds a note to the definition of PVCLOCK_COUNTS_FROM_ZERO.
> >>
> >> Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
> >> ---
> >>  arch/x86/include/asm/pvclock-abi.h | 1 +
> >>  arch/x86/kvm/x86.c                 | 4 ----
> >>  2 files changed, 1 insertion(+), 4 deletions(-)
> >>
> >> diff --git a/arch/x86/include/asm/pvclock-abi.h b/arch/x86/include/asm/pvclock-abi.h
> >> index 655e07a48f6c..67f08230103a 100644
> >> --- a/arch/x86/include/asm/pvclock-abi.h
> >> +++ b/arch/x86/include/asm/pvclock-abi.h
> >> @@ -41,6 +41,7 @@ struct pvclock_wall_clock {
> >>  
> >>  #define PVCLOCK_TSC_STABLE_BIT	(1 << 0)
> >>  #define PVCLOCK_GUEST_STOPPED	(1 << 1)
> >> +/* PVCLOCK_COUNTS_FROM_ZERO broke ABI and can't be used anymore. */
> >>  #define PVCLOCK_COUNTS_FROM_ZERO (1 << 2)
> >>  #endif /* __ASSEMBLY__ */
> >>  #endif /* _ASM_X86_PVCLOCK_ABI_H */
> >> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> >> index 18d59b584dee..34d33f4757d2 100644
> >> --- a/arch/x86/kvm/x86.c
> >> +++ b/arch/x86/kvm/x86.c
> >> @@ -1707,8 +1707,6 @@ static int kvm_guest_time_update(struct kvm_vcpu *v)
> >>  		vcpu->pvclock_set_guest_stopped_request = false;
> >>  	}
> >>  
> >> -	pvclock_flags |= PVCLOCK_COUNTS_FROM_ZERO;
> >> -
> >>  	/* If the host uses TSC clocksource, then it is stable */
> >>  	if (use_master_clock)
> >>  		pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
> >> @@ -2006,8 +2004,6 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
> >>  					&vcpu->requests);
> >>  
> >>  			ka->boot_vcpu_runs_old_kvmclock = tmp;
> >> -
> >> -			ka->kvmclock_offset = -get_kernel_ns();
> >>  		}
> >>  
> >>  		vcpu->arch.time = data;
> >> -- 
> >> 2.5.2
> > 
> > ACK
> 
> So I suppose you changed your mind :) but can you explain the reasoning?
> 
> Paolo

The patch is correct. Overflow (issue raised) is only an issue 
without PVCLOCK_TSC_STABLE_BIT.