linux-kernel - Re: [PATCH] arm64: export tishift functions to modules

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-next>] [day] [month] [year] [list]

Message-ID: <CAHmME9oO_5X6M3NLe2tWp=ZY5EQY17GQ6BUomHLBscMejiXTvw@mail.gmail.com>
Date:   Tue, 24 Apr 2018 15:43:04 +0200
From:   "Jason A. Donenfeld" <Jason@...c4.com>
To:     Will Deacon <will.deacon@....com>
Cc:     linux-arm-kernel@...ts.infradead.org,
        LKML <linux-kernel@...r.kernel.org>,
        Ard Biesheuvel <ard.biesheuvel@...aro.org>,
        PaX Team <pageexec@...email.hu>, stable@...r.kernel.org
Subject: Re: [PATCH] arm64: export tishift functions to modules

On Tue, Apr 24, 2018 at 3:34 PM, Will Deacon <will.deacon@....com> wrote:
> I've not run into any build issues here -- is this specifically with some
> out-of-tree module?

I received a bug report email about this. I'm not sure which specific
module, and I assumed from the email that it was actually a result of
in-tree configuration options rather than an out-of-tree module, but
I'm not sure exactly. Either way, I was able to reproduce the problem
by coding up a little PoC out-of-tree module, so it is certainly a
real problem.

> It would be better not to introduce a new header file just for this, I
> think. How about compiler.h instead?

I could, but actually after I wrote this email I noticed that this is
a widespread convention:

zx2c4@...nkpad ~/Projects/linux $ subfind asm-prototypes
./arch/s390/include/asm/asm-prototypes.h
./arch/alpha/include/asm/asm-prototypes.h
./arch/powerpc/include/asm/asm-prototypes.h
./arch/m68k/include/asm/asm-prototypes.h
./arch/mips/include/asm/asm-prototypes.h
./arch/x86/include/asm/asm-prototypes.h
./arch/sparc/include/asm/asm-prototypes.h
./arch/ia64/include/asm/asm-prototypes.h
./arch/um/include/asm/asm-prototypes.h
./include/asm-generic/asm-prototypes.h

>
> We normally export asm symbols via arm64ksyms.c. In fact, would doing that
> remove the need for the explicit declarations completely?

I'm pretty sure it still needs the declaration; otherwise the module
hashing will get confused. Also, the EXPORT_SYMBOL macro is a
different one when called from assembly versus from C, though not sure
that makes a substantive difference. It seems like this is what other
architectures are doing:

zx2c4@...nkpad ~/Projects/linux $ rg 'EXPORT_SYMBOL.*__.*[std]i[0-9]' -g '*.S'
arch/m68k/lib/modsi3.S
111:    EXPORT_SYMBOL(__modsi3)

arch/m68k/lib/umodsi3.S
108:    EXPORT_SYMBOL(__umodsi3)

arch/m68k/lib/udivsi3.S
157:    EXPORT_SYMBOL(__udivsi3)

arch/m68k/lib/divsi3.S
123:    EXPORT_SYMBOL(__divsi3)

arch/m68k/lib/mulsi3.S
105:    EXPORT_SYMBOL(__mulsi3)

arch/powerpc/kernel/misc_32.S
529:EXPORT_SYMBOL(__ashrdi3)
541:EXPORT_SYMBOL(__ashldi3)
553:EXPORT_SYMBOL(__lshrdi3)
569:EXPORT_SYMBOL(__cmpdi2)
584:EXPORT_SYMBOL(__ucmpdi2)
596:EXPORT_SYMBOL(__bswapdi2)

arch/powerpc/kernel/misc_64.S
211:EXPORT_SYMBOL(__bswapdi2)

arch/sparc/lib/lshrdi3.S
30:EXPORT_SYMBOL(__lshrdi3)

arch/sparc/lib/muldi3.S
78:EXPORT_SYMBOL(__muldi3)

arch/sparc/lib/divdi3.S
283:EXPORT_SYMBOL(__divdi3)

arch/sparc/lib/ashrdi3.S
40:EXPORT_SYMBOL(__ashrdi3)

arch/sparc/lib/multi3.S
36:EXPORT_SYMBOL(__multi3)

arch/sparc/lib/ashldi3.S
38:EXPORT_SYMBOL(__ashldi3)