linux-kernel - Re: [PATCH] proc: use #pragma once

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: <94db646a-4fd8-9ce3-2111-b2e4852ad324@rasmusvillemoes.dk>
Date:   Fri, 4 May 2018 01:23:12 +0200
From:   Rasmus Villemoes <linux@...musvillemoes.dk>
To:     Al Viro <viro@...IV.linux.org.uk>
Cc:     Andrew Morton <akpm@...ux-foundation.org>,
        Alexey Dobriyan <adobriyan@...il.com>, dsterba@...e.cz,
        Christoph Hellwig <hch@...radead.org>,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH] proc: use #pragma once

On 2018-05-04 00:42, Al Viro wrote:
> On Fri, May 04, 2018 at 12:14:57AM +0200, Rasmus Villemoes wrote:
> 
>> FWIW, it's not just removing some identifiers from cpp's hash tables, it
>> also reduces I/O: Due to our header mess, we have some cyclic includes,
>> e.g mm.h -> memremap.h -> mm.h. While parsing mm.h, cpp sees the #define
>> _LINUX_MM_H, then goes parsing memremap.h, but since it hasn't reached
>> the end of mm.h yet (seeing that there's nothing but comments outside
>> the #ifndef/#endif pair), it hasn't had a chance to set the internal
>> flag for mm.h, so it goes slurping in mm.h again. Obviously, the
>> definedness of _LINUX_MM_H at that point means it "only" has to parse
>> those 87K for comments and matching up #ifs, #ifdefs,#endifs etc. With
>> #pragma once, the flag gets set for mm.h immediately, so the #include
>> from memremap.h is entirely ignored. This can easily be verified with
>> strace. And mm.h is not the only header getting read twice.
> 
> Which gcc version it is?

When I stumbled on this a few years ago (I was stracing gcc for some
reason), 4.something. Verifying today, 5.4.

> Does it *really* read anything twice?

Yes, just do "strace -f" and grep 'open.*= [0-9]'.

> After all, you are going to read (and tokenize) the rest of mm.h anyway, so if it
> throws away the file it has read, only to reread it again, it's bloody
> dumb.

Hm, yes, gcc should be able to hang the buffer off of the structure it
keeps around anyway for tracking the status of mm.h. But it doesn't.

> Note that sequence of preprocessor tokens does not depend upon the ifdefs;
> anything under #if 0 *is* tokenized all the same.  So it's not even that
> "parsing" (tokenizing, actually) has to be repeated.

Are you sure about that? Maybe formally it has to (if the phases are
done strictly sequentially), but can't it go into some fast-forward mode
under #if 0 (or #ifndef $some-defined-macro) where it only looks for
string and char literals, comments and "control-flow" directives?
libcpp/internal.h does have

  /* True if we are skipping a failed conditional group.  */
  unsigned char skipping;

Rasmus