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: <9fba4d17-d826-6a3a-86fc-60977aa6cc9e@strupe.net>
Date:   Wed, 13 May 2020 20:56:08 +0200
From:   Fredrik Strupe <fredrik@...upe.net>
To:     Russell King - ARM Linux admin <linux@...linux.org.uk>
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 13.05.2020 20:12, Russell King - ARM Linux admin wrote:
> 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.
>

That makes sense. Thank you very much for a great answer!

Fredrik