linux-kernel - Re: [RFC PATCH] arm: Don't trap conditional UDF instructions

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20200513181209.GM1551@shell.armlinux.org.uk>
Date:   Wed, 13 May 2020 19:12:10 +0100
From:   Russell King - ARM Linux admin <linux@...linux.org.uk>
To:     Fredrik Strupe <fredrik@...upe.net>
Cc:     linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
        Oleg Nesterov <oleg@...hat.com>,
        Richard Fontana <rfontana@...hat.com>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        Allison Randal <allison@...utok.net>,
        Kate Stewart <kstewart@...uxfoundation.org>,
        Enrico Weigelt <info@...ux.net>
Subject: Re: [RFC PATCH] arm: Don't trap conditional UDF instructions

On Wed, May 13, 2020 at 05:41:58PM +0200, Fredrik Strupe wrote:
> Hi,
> 
> This is more of a question than a patch, but I hope the attached patch makes
> the issue a bit clearer.
> 
> The arm port of Linux supports hooking/trapping of undefined instructions. Some
> parts of the code use this to trap UDF instructions with certain immediates in
> order to use them for other purposes, like 'UDF #16' which is equivalent to a
> BKPT instruction in A32.
> 
> Moreover, most of the undef hooks on UDF instructions assume that UDF is
> conditional and mask out the condition prefix during matching. The attached
> patch shows the locations where this happens. However, the Arm architecture
> reference manual explicitly states that UDF is *not* conditional, making
> any instruction encoding with a condition prefix other than 0xe (always
> execute) unallocated.

The latest version of the ARM architecture reference manual may say
that, but earlier versions say different things. The latest reference
manual does not apply to earlier architectures, so if you're writing
code to cover multiple different architectures, you must have an
understanding of each of those architectures.

So, from the code:

	ARM:   xxxx 0111 1111 xxxx xxxx xxxx 1111 xxxx

>From DDI0100E:

3.13.1 Undefined instruction space
       Instructions with the following opcodes are undefined
       instruction space:

       opcode[27:25] = 0b011
       opcode[4] = 1

       31 28 27 26 25 24                                     5 4 3     0
       cond  0  1  1  x  x x x x x x x x x x x x x x x x x x x 1 x x x x

So, in this version of the architecture, undefined instructions may
be conditional - and indeed that used to be the case.  The condition
code was always respected, and cond=1111 meant "never" (NV).

Hence, trapping them if the condition code is not 1110 (AL) is
entirely reasonable, legal and safe.  If an ARM CPU defines an
instruction coding that matches the above, then it won't take the
undefined instruction trap, and we'll never see it.

Now, as for UDF usage in the kernel, it may be quite correct that we
always use the AL condition code for them, but it would be very odd
for there to be an instruction implemented with a different (non-NV)
condition code that can't also have it's AL condition code encoding.
You could never execute such an instruction unconditionally.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 10.2Mbps down 587kbps up