Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 7 Jun 2023 02:08:11 +0000
From:   Luiz Capitulino <luizcap@amazon.com>
To:     Sean Christopherson <seanjc@google.com>
CC:     Paolo Bonzini <pbonzini@redhat.com>, <kvm@vger.kernel.org>,
        <linux-kernel@vger.kernel.org>, Li RongQing <lirongqing@baidu.com>,
        Yong He <zhuangel570@gmail.com>,
        Robert Hoo <robert.hoo.linux@gmail.com>,
        Kai Huang <kai.huang@intel.com>
Subject: Re: [PATCH] KVM: x86/mmu: Add "never" option to allow sticky
 disabling of nx_huge_pages
Message-ID: <20230607020743.GA53474@dev-dsk-luizcap-1d-37beaf15.us-east-1.amazon.com>
References: <20230602005859.784190-1-seanjc@google.com>
 <20230606202557.GA71782@dev-dsk-luizcap-1d-37beaf15.us-east-1.amazon.com>
 <ZH+tOj7lRAGsua9X@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Disposition: inline
In-Reply-To: <ZH+tOj7lRAGsua9X@google.com>
User-Agent: Mutt/1.5.21 (2010-09-15)
Precedence: bulk

On Tue, Jun 06, 2023 at 03:03:38PM -0700, Sean Christopherson wrote:
> 
> 
> 
> On Tue, Jun 06, 2023, Luiz Capitulino wrote:
> > On Thu, Jun 01, 2023 at 05:58:59PM -0700, Sean Christopherson wrote:
> > However, why don't we make nx_huge_pages=never the default behavior if the
> > CPU is not vulnerable?
> 
> Mainly because the mitigation has been around for 3.5 years, and there's a non-zero
> chance that making "never" the default could cause hiccups for unsuspecting users.
> If this were brand new code, I would definitely opt for "never" as the default.

OK.

> > If there are concerns about not being able to restart the worker thread, then
> > maybe we could make this a .config option?
> 
> Eh, a Kconfig is unnecessarily complex, and wouldn't really change anything, e.g.
> for users in the know, it's just as easy to force a module param as it is to force
> a Kconfig, and to gain any benefit from the param being !never by default, the
> Kconfig would also have to be off by default.

I agree it adds some complexity. The benefit is to allow shipping a kernel with
a good default where KVM users with non-vulnerable CPUs get low latency out
of the box (vs. having to figure out if they are vulnerable or not and changing
a run-time configuration).

But the idea would be to set "never" by default as long as the CPU is not vulnerable.

> If "everyone" wants never to be the default, and Paolo doesn't object, I'd rather
> just tack on a patch to make that happen, and cross my fingers there's no fallout :-)

This would work too :)