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
References: <20200210225806.249297-1-trishalfonso@google.com>
 <13b0ea0caff576e7944e4f9b91560bf46ac9caf0.camel@sipsolutions.net>
 <CAKFsvUKaixKXbUqvVvjzjkty26GS+Ckshg2t7-+erqiN2LVS-g@mail.gmail.com> <e8a45358b273f0d62c42f83d99c1b50a1608929d.camel@sipsolutions.net>
In-Reply-To: <e8a45358b273f0d62c42f83d99c1b50a1608929d.camel@sipsolutions.net>
From:   Dmitry Vyukov <dvyukov@google.com>
Date:   Thu, 13 Feb 2020 09:44:37 +0100
Message-ID: <CACT4Y+ZB3QwzeogxVFVXW_z=eE2n5fQxj7iYq9-Jw68zdS=mUA@mail.gmail.com>
Subject: Re: [RFC PATCH v2] UML: add support for KASAN under x86_64
To:     Johannes Berg <johannes@sipsolutions.net>
Cc:     Patricia Alfonso <trishalfonso@google.com>,
        Jeff Dike <jdike@addtoit.com>,
        Richard Weinberger <richard@nod.at>,
        anton.ivanov@cambridgegreys.com,
        Andrey Ryabinin <aryabinin@virtuozzo.com>,
        David Gow <davidgow@google.com>,
        Brendan Higgins <brendanhiggins@google.com>,
        kasan-dev <kasan-dev@googlegroups.com>,
        LKML <linux-kernel@vger.kernel.org>, linux-um@lists.infradead.org
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Feb 13, 2020 at 9:19 AM Johannes Berg <johannes@sipsolutions.net> wrote:
>
> On Wed, 2020-02-12 at 16:37 -0800, Patricia Alfonso wrote:
> >
> > > That also means if I have say 512MB memory allocated for UML, KASAN will
> > > use an *additional* 64, unlike on a "real" system, where KASAN will take
> > > about 1/8th of the available physical memory, right?
> > >
> > Currently, the amount of shadow memory allocated is a constant based
> > on the amount of user space address space in x86_64 since this is the
> > host architecture I have focused on.
>
> Right, but again like below - that's just mapped, not actually used. But
> as far as I can tell, once you actually start running and potentially
> use all of your mem=1024 (MB), you'll actually also use another 128MB on
> the KASAN shadow, right?
>
> Unlike, say, a real x86_64 machine where if you just have 1024 MB
> physical memory, the KASAN shadow will have to fit into that as well.

Depends on what you mean by "real" :)
Real user-space ASAN will also reserve 1/8th of 47-bit VA on start
(16TB). This implementation seems to be much closer to user-space ASAN
rather than to x86_64 KASAN (in particular it seems to be mostly
portable across archs and is not really x86-specific, which is good).
I think it's reasonable and good, but the implementation difference
with other kernel arches may be worth noting somewhere in comments.