Subject: Re: [RFC PATCH v3 0/6] KVM: x86: avoid redundant REQ_EVENT
To: Paolo Bonzini <pbonzini@redhat.com>, <linux-kernel@vger.kernel.org>,
        <kvm@vger.kernel.org>
References: <1482140838-6303-1-git-send-email-pbonzini@redhat.com>
CC: <rkrcmar@redhat.com>, <rkagan@virtuozzo.com>
From: Denis Plotnikov <dplotnikov@virtuozzo.com>
Message-ID: <e805e616-f840-8cb2-3133-4557c80afff9@virtuozzo.com>
Date: Mon, 19 Dec 2016 16:05:55 +0300
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101
 Thunderbird/45.5.1
MIME-Version: 1.0
In-Reply-To: <1482140838-6303-1-git-send-email-pbonzini@redhat.com>
Content-Type: text/plain; charset="windows-1252"; format=flowed
Content-Transfer-Encoding: 7bit
SpamDiagnosticOutput: 1:99
SpamDiagnosticMetadata: NSPM
X-MS-Exchange-CrossTenant-OriginalArrivalTime: 19 Dec 2016 13:06:44.0394 (UTC)
X-MS-Exchange-CrossTenant-FromEntityHeader: Hosted
X-MS-Exchange-Transport-CrossTenantHeadersStamped: AM4PR0802MB2212
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org
Content-Length: 1992
Lines: 58


On 19.12.2016 12:47, Paolo Bonzini wrote:
> This is the result of cleaning up the places that set REQ_EVENT unnecessarily.
> The savings on self-IPI kvm-unit-tests are:
>
>     self_ipi_sti_nop                ~300 clock cycles
>     self_ipi_sti_hlt                ~300 clock cycles
>     self_ipi_tpr                    ~400 clock cycles
>     self_ipi_tpr_sti_nop            ~700 clock cycles
>     self_ipi_tpr_sti_hlt            ~700 clock cycles
>
> which is in the 5-10% range.  Please help me comparing this series with
> "[PATCH v2] KVM: x86: avoid redundant REQ_EVENT" on the real workload!
>
> Thanks,
>
> Paolo
This new patch-set does avoid unnecessary interrupt re-injections - checked.

The test (MS Windows, sending messages between multiple windows using 
windows-specific interface),
which showed performance growth with "[PATCH v2] KVM: x86: avoid 
redundant REQ_EVENT" showed pretty much the same performance level with 
this new patch-set.
The test score difference (+2.4% to [PATCH v2]) is within the tolerance 
range(5%).

The test score on mainstream v4.9 kernel CPU Intel i5-2400, guest 
Windows Server 2012 2VCPU, 2GB:

without patch: 31510 (+/- 4%)
    with patch: 36270 (+/- 5%)

difference = (36270-31510)/31510 * 100 = +15% -- looks good!

Thanks!

>
> Paolo Bonzini (6):
>   KVM: x86: add VCPU stat for KVM_REQ_EVENT processing
>   KVM: lapic: do not set KVM_REQ_EVENT unnecessarily on EOI
>   KVM: vmx: speed up TPR below threshold vmexits
>   KVM: lapic: remove unnecessary KVM_REQ_EVENT on PPR update
>   KVM: lapic: do not set KVM_REQ_EVENT unnecessarily on PPR update
>   KVM: lapic: do not scan IRR when delivering an interrupt
>
>  arch/x86/include/asm/kvm_host.h |  1 +
>  arch/x86/kvm/lapic.c            | 58 +++++++++++++++++++++++++----------------
>  arch/x86/kvm/lapic.h            |  1 +
>  arch/x86/kvm/vmx.c              |  2 +-
>  arch/x86/kvm/x86.c              |  2 ++
>  5 files changed, 41 insertions(+), 23 deletions(-)
>

-- 
Best,
Denis