From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-kernel-owner@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
	id S1754616AbbBTNcT (ORCPT <rfc822;w@1wt.eu>);
	Fri, 20 Feb 2015 08:32:19 -0500
Received: from mx1.redhat.com ([209.132.183.28]:55034 "EHLO mx1.redhat.com"
	rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
	id S1754232AbbBTNcC (ORCPT <rfc822;linux-kernel@vger.kernel.org>);
	Fri, 20 Feb 2015 08:32:02 -0500
Date: Fri, 20 Feb 2015 14:30:02 +0100
From: Oleg Nesterov <oleg@redhat.com>
To: Borislav Petkov <bp@alien8.de>
Cc: Rik van Riel <riel@redhat.com>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        Suresh Siddha <sbsiddha@gmail.com>, linux-kernel@vger.kernel.org,
        mingo@redhat.com, hpa@zytor.com, matt.fleming@intel.com, bp@suse.de,
        pbonzini@redhat.com, tglx@linutronix.de, luto@amacapital.net
Subject: Re: [PATCH 0/3] x86, fpu: more eagerfpu cleanups
Message-ID: <20150220133002.GA32275@redhat.com>
References: <1421012793-30106-1-git-send-email-riel@redhat.com> <20150115191918.GA27332@redhat.com> <20150119185109.GA16427@redhat.com> <20150220121009.GA19589@pd.tnic>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20150220121009.GA19589@pd.tnic>
User-Agent: Mutt/1.5.18 (2008-05-17)
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org

On 02/20, Borislav Petkov wrote:
>
> On Mon, Jan 19, 2015 at 07:51:09PM +0100, Oleg Nesterov wrote:
> > math_state_restore() and its usage doesn't look nice. init_fpu() too,
>
> Yeah, about that:
>
> I see:
>
> math_state_restore
> ...
> 	 if (!tsk_used_math(tsk))
> 		init_fpu()
>
> and init_fpu() then does:
>
> 	if (tsk_used_math(tsk))
>

Yes, and more. math_state_restore() assumes that it is called with irqs
disabled. At least if !tsk_used_math. That is why 3/3 calls init_fpu()
first. Not only this doesn't look clean, this is simply not true in
general.

The comment above init_fpu() is simply wrong. And unlazy_fpu() from
there doesn't look nice. This mixes 2 completely differents things.

> Could use a cleanup and so on... Perhaps it is in the works already :)

Yes, I'll try to make the cleanups on top of these changes. And let me
repeat that there is another reason for 1/3 and 2/3 at least (3/3 makes
sense too), if we add TIF_LOAD_FPU we need to avoid the performance
regression (irq_fpu_usable() should not fail if !__thread_has_fpu()).

Oleg.