Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Mon, 25 Mar 2019 13:46:32 +0100 (CET)
From:   Thomas Gleixner <tglx@linutronix.de>
To:     "Chang S. Bae" <chang.seok.bae@intel.com>
cc:     Ingo Molnar <mingo@kernel.org>, Andy Lutomirski <luto@kernel.org>,
        "H . Peter Anvin" <hpa@zytor.com>, Andi Kleen <ak@linux.intel.com>,
        Ravi Shankar <ravi.v.shankar@intel.com>,
        LKML <linux-kernel@vger.kernel.org>,
        Andrew Cooper <andrew.cooper3@citrix.com>, x86@kernel.org
Subject: Re: [RESEND PATCH v6 04/12] x86/fsgsbase/64: Enable FSGSBASE
 instructions in the helper functions
In-Reply-To: <alpine.DEB.2.21.1903251107300.1798@nanos.tec.linutronix.de>
Message-ID: <alpine.DEB.2.21.1903251345180.1656@nanos.tec.linutronix.de>
References: <1552680405-5265-1-git-send-email-chang.seok.bae@intel.com> <1552680405-5265-5-git-send-email-chang.seok.bae@intel.com> <alpine.DEB.2.21.1903251107300.1798@nanos.tec.linutronix.de>
User-Agent: Alpine 2.21 (DEB 202 2017-01-01)
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, 25 Mar 2019, Thomas Gleixner wrote:
> The whole point of FSGSBASE support is performance, right?
> 
> So can please someone explain why having the following in the context
> switch path when it can be completely avoided is enhancing performance:
> 
>   - 4 x SWAPGS
>   - 1 x RDMSR
>   - 1 x WRMSR

Corrrecting myself. That should be:

     RDGSBASE
     WRGSBASE

obviously. Still the point remains.

>   - 2 x local_irq_save()
>   - 2 x local_irq_restore()
> 
> Of course the local_irq_save/restore() pairs are utterly pointless because
> switch_to() runs with interrupts disabled already.
> 
> SWAPGS instead needs:
> 
>   1 x WRMSR
> 
> and nothing else.
> 
> So trading the single WRMSR against the above in the context switch path is
> gaining performance, right?
> 
> The only thing which gains performance is user space switching GS. And this
> user space performance gain is achieved by:
> 
>   - Inconsistent and fragile code with a guarantee for subtle and hard to
>     diagnose bugs
> 
>   - Pointless overhead in the context switch code
> 
> Sorry, not going to happen ever.
> 
> Get your act together and make this consistent. Either SWAPGS or FSGSBASE,
> but not a mix of it.
> 
> Thanks,
> 
> 	tglx
>