linux-kernel - Re: [RFC 5/8] x86: refcount: prevent gcc distortions

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]

Date:   Wed, 16 May 2018 16:44:00 +0000
From:   Nadav Amit <namit@...are.com>
To:     Jan Beulich <JBeulich@...e.com>
CC:     "keescook@...omium.org" <keescook@...omium.org>,
        the arch/x86 maintainers <x86@...nel.org>,
        "tglx@...utronix.de" <tglx@...utronix.de>,
        Josh Poimboeuf <jpoimboe@...hat.com>,
        "mingo@...hat.com" <mingo@...hat.com>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "hpa@...or.com" <hpa@...or.com>
Subject: Re: [RFC 5/8] x86: refcount: prevent gcc distortions

Jan Beulich <JBeulich@...e.com> wrote:

>>>> On 15.05.18 at 16:11, <namit@...are.com> wrote:
>> --- a/arch/x86/include/asm/refcount.h
>> +++ b/arch/x86/include/asm/refcount.h
>> @@ -14,34 +14,43 @@
>>  * central refcount exception. The fixup address for the exception points
>>  * back to the regular execution flow in .text.
>>  */
>> -#define _REFCOUNT_EXCEPTION				\
>> -	".pushsection .text..refcount\n"		\
>> -	"111:\tlea %[counter], %%" _ASM_CX "\n"		\
>> -	"112:\t" ASM_UD2 "\n"				\
>> -	ASM_UNREACHABLE					\
>> -	".popsection\n"					\
>> -	"113:\n"					\
>> +
>> +asm ("\n"
>> +	".macro __REFCOUNT_EXCEPTION counter:vararg\n\t"
>> +	".pushsection .text..refcount\n"
>> +	"111:\tlea \\counter, %" _ASM_CX "\n"
>> +	"112:\t" ASM_UD2 "\n\t"
>> +	ASM_UNREACHABLE
>> +	".popsection\n\t"
>> +	"113:\n"
>> 	_ASM_EXTABLE_REFCOUNT(112b, 113b)
>> +	".endm");
> 
> A few comments on assembly code formatting - while gas at present is
> relatively lax in this regard, I wouldn't exclude that there might be a
> more strict mode in the future, and that such a mode might eventually
> become the default. Furthermore these formatting aspects affect
> readability of the assembly produced, should anyone ever find a need
> to look at it (perhaps because of some breakage) - I certainly do every
> once in a while.
> 
> Labels should be placed without any indentation (but of course there
> may be more than one on a line, in which case subsequent ones may
> of course be arbitrarily indented). Instructions and directives, otoh,
> should be placed with at least a single tab or space of indentation
> (unless preceded by a label, in which case the extra white space still
> helps readability).

Writing these patches, I looked at the generated assembly, and there did not
appear to be a standard. IIRC, .pushsection directives were not always
inlined. I will fix it according to your comments.

> I'm also not sure about the purpose of the leading plain newline here.
> gcc annotates code resulting from inline assembly anyway iirc, so
> proper visual separation should already be available.

Right. It was only to get the macro directive not tabulated, but as you
said, it should be tabulated, so I will remove it.

> 
> If I was the maintainer of this code, I would also object to the
> mis-alignment your file scope asm()-s have ("asm (" is 5 characters,
> which doesn't equal a tab's width).

I tried many formats (including the one you propose), and eventually went
with the one that made checkpatch not yell at me. I will revert to the one
you propose, which makes most sense, and ignore checkpatch warnings.

Thanks,
Nadav