Received-SPF: pass (google.com: domain of linux-kernel+bounces-133706-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Date: Fri, 5 Apr 2024 16:20:41 -0700
In-Reply-To: <20240323080541.10047-2-pmenzel@molgen.mpg.de>
Precedence: bulk
Mime-Version: 1.0
References: <20240323080541.10047-2-pmenzel@molgen.mpg.de>
Message-ID: <ZhCHSYwS5_o-OKs0@google.com>
Subject: Re: [PATCH] KVM: VMX: make vmx_init a late init call to get to init
 process faster
From: Sean Christopherson <seanjc@google.com>
To: Paul Menzel <pmenzel@molgen.mpg.de>
Cc: Paolo Bonzini <pbonzini@redhat.com>, Thomas Gleixner <tglx@linutronix.de>, 
	Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>, 
	Dave Hansen <dave.hansen@linux.intel.com>, x86@kernel.org, 
	"H. Peter Anvin" <hpa@zytor.com>, Colin Ian King <colin.i.king@intel.com>, kvm@vger.kernel.org, 
	linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="us-ascii"

On Sat, Mar 23, 2024, Paul Menzel wrote:
> From: Colin Ian King <colin.i.king@intel.com>
> 
> Making vmx_init a late initcall improves QEMU kernel boot times to
> get to the init process. Average of 100 boots, QEMU boot average
> reduced from 0.776 seconds to 0.622 seconds (~19.8% faster) on
> Alder Lake i9-12900 and ~0.5% faster for non-QEMU UEFI boots.

The changelog needs to better explain what "QEMU kernel boot times" means.  I
assume the test is a QEMU VM running a kernel KVM_INTEL built-in?  This should
also call out that late_initcall is #defined to module_init() when KVM is built
as a module.

> Signed-off-by: Colin Ian King <colin.i.king@intel.com>
> [Take patch
> https://github.com/clearlinux-pkgs/linux/commit/797db35496031b19ba37b1639ac5fa5db9159a06
> and fix spelling of Alder Lake.]
> Signed-off-by: Paul Menzel <pmenzel@molgen.mpg.de>
> ---
>  arch/x86/kvm/vmx/vmx.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> index c37a89eda90f..0a9f4b20fbda 100644
> --- a/arch/x86/kvm/vmx/vmx.c
> +++ b/arch/x86/kvm/vmx/vmx.c
> @@ -8783,4 +8783,4 @@ static int __init vmx_init(void)
>  	kvm_x86_vendor_exit();
>  	return r;
>  }
> -module_init(vmx_init);
> +late_initcall(vmx_init);

_If_ we do this, then we should also give svm_init() and kvm_x86_init() the same
treatment.  I see no reason for vmx_init() to be special.

I'm not opposed to this, but I also have zero idea if this could have a negative
impact userspace.  E.g. what happens if some setup's init process expects /dev/kvm
to exist?  Will this break that?