[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CACT4Y+bPzRbWw-dPQkLVENPKy_DBdjrbSce0f6XE3=W7RhfhBA@mail.gmail.com>
Date: Thu, 16 Jan 2020 09:44:08 +0100
From: Dmitry Vyukov <dvyukov@...gle.com>
To: Patricia Alfonso <trishalfonso@...gle.com>
Cc: Jeff Dike <jdike@...toit.com>, Richard Weinberger <richard@....at>,
anton.ivanov@...bridgegreys.com,
Andrey Ryabinin <aryabinin@...tuozzo.com>,
David Gow <davidgow@...gle.com>,
Brendan Higgins <brendanhiggins@...gle.com>,
linux-um@...ts.infradead.org,
kasan-dev <kasan-dev@...glegroups.com>,
LKML <linux-kernel@...r.kernel.org>
Subject: Re: [RFC PATCH] UML: add support for KASAN under x86_64
On Wed, Jan 15, 2020 at 7:28 PM Patricia Alfonso
<trishalfonso@...gle.com> wrote:
> +config KASAN_SHADOW_OFFSET
> + hex
> + depends on KASAN
> + default 0x100000000000
> + help
> + This is the offset at which the ~2.25TB of shadow memory is
> + initialized and used by KASAN for memory debugging. The default
> + is 0x100000000000.
What are restrictions on this value?
In user-space we use 0x7fff8000 as a base (just below 2GB) and it's
extremely profitable wrt codegen since it fits into immediate of most
instructions.
We can load and add the base with a short instruction:
2d8c: 48 81 c2 00 80 ff 7f add $0x7fff8000,%rdx
Or even add base, load shadow and check it with a single 7-byte instruction:
1e4: 80 b8 00 80 ff 7f 00 cmpb $0x0,0x7fff8000(%rax)
While with the large base, it takes 10 bytes just to load the const
into a register (current x86 KASAN codegen):
ffffffff81001571: 48 b8 00 00 00 00 00 fc ff df movabs
$0xdffffc0000000000,%rax
Most instructions don't have 8-byte immediates, so then we separately
need to add/load/check.
Powered by blists - more mailing lists