Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
To:     Marcelo Tosatti <mtosatti@redhat.com>
Cc:     linux-kernel@vger.kernel.org, kvm@vger.kernel.org,
        vkuznets@redhat.com, dwmw@amazon.co.uk
References: <20210330165958.3094759-1-pbonzini@redhat.com>
 <20210330165958.3094759-2-pbonzini@redhat.com>
 <20210407174021.GA30046@fuller.cnet>
 <51cae826-8973-5113-7e12-8163eab36cb7@redhat.com>
 <20210408120021.GA65315@fuller.cnet>
From:   Paolo Bonzini <pbonzini@redhat.com>
Subject: Re: [PATCH 1/2] KVM: x86: reduce pvclock_gtod_sync_lock critical
 sections
Message-ID: <2abe4b19-e41e-34f9-0a3c-30812c7b719e@redhat.com>
Date:   Thu, 8 Apr 2021 14:25:41 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.7.0
MIME-Version: 1.0
In-Reply-To: <20210408120021.GA65315@fuller.cnet>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 08/04/21 14:00, Marcelo Tosatti wrote:
>>
>> KVM_REQ_MCLOCK_INPROGRESS is only needed to kick running vCPUs out of the
>> execution loop;
> We do not want vcpus with different system_timestamp/tsc_timestamp
> pair:
> 
>   * To avoid that problem, do not allow visibility of distinct
>   * system_timestamp/tsc_timestamp values simultaneously: use a master
>   * copy of host monotonic time values. Update that master copy
>   * in lockstep.
> 
> So KVM_REQ_MCLOCK_INPROGRESS also ensures that no vcpu enters
> guest mode (via vcpu->requests check before VM-entry) with a
> different system_timestamp/tsc_timestamp pair.

Yes this is what KVM_REQ_MCLOCK_INPROGRESS does, but it does not have to 
be done that way.  All you really need is the IPI with KVM_REQUEST_WAIT, 
which ensures that updates happen after the vCPUs have exited guest 
mode.  You don't need to loop on vcpu->requests for example, because 
kvm_guest_time_update could just spin on pvclock_gtod_sync_lock until 
pvclock_update_vm_gtod_copy is done.

So this morning I tried protecting the kvm->arch fields for kvmclock 
using a seqcount, which is nice also because get_kvmclock_ns() does not 
have to bounce the cacheline of pvclock_gtod_sync_lock anymore.  I'll 
post it tomorrow or next week.

Paolo