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;
MIME-Version: 1.0
In-Reply-To: <20180728215357.3249-11-riel@surriel.com>
References: <20180728215357.3249-1-riel@surriel.com> <20180728215357.3249-11-riel@surriel.com>
From:   Andy Lutomirski <luto@kernel.org>
Date:   Sat, 28 Jul 2018 21:21:17 -0700
Message-ID: <CALCETrWS4pE5Tv3OErM3AD-E=CGG3U+2CBVecRF3N0c+g42vEg@mail.gmail.com>
Subject: Re: [PATCH 10/10] mm,sched: conditionally skip lazy TLB mm refcounting
To:     Rik van Riel <riel@surriel.com>
Cc:     LKML <linux-kernel@vger.kernel.org>,
        kernel-team <kernel-team@fb.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Andrew Lutomirski <luto@kernel.org>, X86 ML <x86@kernel.org>,
        Vitaly Kuznetsov <vkuznets@redhat.com>,
        Ingo Molnar <mingo@kernel.org>, Mike Galbraith <efault@gmx.de>,
        Dave Hansen <dave.hansen@intel.com>, will.daecon@arm.com,
        Catalin Marinas <catalin.marinas@arm.com>,
        Benjamin Herrenschmidt <benh@kernel.crashing.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sat, Jul 28, 2018 at 2:53 PM, Rik van Riel <riel@surriel.com> wrote:
> Conditionally skip lazy TLB mm refcounting. When an architecture has
> CONFIG_ARCH_NO_ACTIVE_MM_REFCOUNTING enabled, an mm that is used in
> lazy TLB mode anywhere will get shot down from exit_mmap, and there
> in no need to incur the cache line bouncing overhead of refcounting
> a lazy TLB mm.

Unless I've misunderstood something, this patch results in idle tasks
whose active_mm has been freed still having active_mm pointing at
freed memory.  This isn't strictly speaking a bug, but it's extremely
confusing and risks all kinds of nasty errors.  That's why I prefer
the approach of actually removing the active_mm field on x86 rather
than merely removing the refcount.

I realize that this will add more ifdeffery and make the patch a bit
bigger, but I think it'll be much more robust.  Not to mention that it
will save a pointer an a refcount per mm_struct, but that barely
matters.