linux-kernel - Re: [PATCH v2 00/10] x86: use gcc 6+ asm flag output feature

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, 08 Jun 2016 14:30:51 -0700
From:	"H. Peter Anvin" <hpa@...or.com>
To:	"H. Peter Anvin" <hpa@...ux.intel.com>,
	Ingo Molnar <mingo@...nel.org>,
	Thomas Gleixner <tglx@...utronix.de>,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
CC:	Andy Lutomirski <luto@...capital.net>,
	Borislav Petkov <bp@...en8.de>
Subject: Re: [PATCH v2 00/10] x86: use gcc 6+ asm flag output feature

On June 8, 2016 12:38:36 PM PDT, "H. Peter Anvin" <hpa@...ux.intel.com> wrote:
>From: "H. Peter Anvin" <hpa@...or.com>
>
>gcc 6+ has the ability to let flags (actually, conditions, which are
>specific combinations of flags) to be used directly as asm() outputs.
>The syntax for that is "=@cc<cc>" where <cc> is the same set of
>letters that would be used in a j<cc> or set<cc> instruction
>(e.g. "=@ccz" to test the ZF flag.)
>
>This patchset by itself reduces the size of the x86-64 kernel by
>0.12%, from a baseline of 4.7-rc2 built with gcc 6.1 (first line is
>with the patchset, the second one is without):
>
>     text       data       bss        dec       hex filename
>
>68245656   41004339  20533248  129783243   7bc55cb
>o.i386-allconfig/vmlinux
>68355716   41008499  20533248  129897463   7be13f7
>o.i386-allconfig/vmlinux
>
>127384005  129742359  38150144  295276508  11998fdc
>o.x86_64-allconfig/vmlinux
>127538765  129742295  38150144  295431204  119bec24
>o.x86_64-allconfig/vmlinux
>
>v2: fix a conflict between <linux/random.h> and <asm/archrandom.h>
>    discovered by Ingo Molnar.  There are a few places in x86-specific
>    code where we need all of <arch/archrandom.h> even when
>    xCONFIG_ARCH_RANDOM is disabled, so <linux/random.h> does not
>    suffice.  Only the <asm/archrandom.h> patch has been changed;
>    since it is a little bit different from the others I have moved it
>    to the end.

I should probably clarify one thing: the output type *must* be bool for the flags output.  Fortunately, not only does it also work for the older compilers, but it sometimes allows GCC to generate better code, which is why I have made the change unconditionally.
-- 
Sent from my Android device with K-9 Mail. Please excuse brevity and formatting.