linux-kernel - Re: [PATCH] blkdev: use an efficient way to check merge flags

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] [day] [month] [year] [list]

Date:	Thu, 3 Apr 2014 14:45:42 +0200
From:	Jan Kara <jack@...e.cz>
To:	Zhan Jianyu <nasa4836@...il.com>
Cc:	Jan Kara <jack@...e.cz>, axboe@...nel.dk, hch@....de,
	viro@...iv.linux.org.uk, kmo@...erainc.com,
	namjae.jeon@...sung.com, LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] blkdev: use an efficient way to check merge flags

On Thu 03-04-14 16:00:44, Zhan Jianyu wrote:
> On Thu, Apr 3, 2014 at 2:13 AM, Jan Kara <jack@...e.cz> wrote:
> >   OK, but have you checked the generated code is actually any better? This
> > is something I'd expect a compiler might be able to optimize anyway. And the
> > original code looks more readable to me.
> 
> Hi, Jan,
> 
> I've disassemble the code on my x86_64 box
> (it's inline though, I just look at its call site),
> and found that this patch DOES make it more efficient.
> 
> Orig asm snippt                                               with
> patch asm snippt
> ============                                              ================
> 
> mov    %edx,%ecx                                          mov    %rdx,%r9
> xor    %r8d,%ecx                                             xor    %r8d,%r8d
> test   $0x80,%cl                                              and    $0x380,%r9d
> jne    14c5 <blk_rq_merge_ok+0x15>                test   $0x380,%ecx
> and    $0x3,%ch                                              sete   %r8b
> jne    14c5 <blk_rq_merge_ok+0x15>                cmp    %r8,%r9
> 
> je     14b5 <blk_rq_merge_ok+0x15>
> 
> This saves a branch.
> 
> Furthermore,  I found that gcc is smart enough to try to optimize the
> code, so if we do
> like this, it will generate the most optimal and smallest code :
> 
> 
> static inline bool blk_check_merge_flags(unsigned int flags1,
>                                         ¦unsigned int flags2)
> {
>         return ((flags1 ^ flags2) &
>                 (REQ_DISCARD | REQ_SECURE | REQ_WRITE_SAME))
>                 == 0;
> }
> 
> this gives out  :
> 
> mov    %edx,%r8d
> xor    %ecx,%r8d
> and    $0x380,%r8d
> jne    14a5 <blk_rq_merge_ok+0x15>
> 
> But yes, it compromises readibility.
  OK, I'd expect gcc is more clever ;). Thanks for the comparison. Anyway
if that function is performance sensitive, we can use your optimization.
Just add a comment there that we want to check whether the three flags are
the same in both flags and that checking your way generates better code.

								Honza
-- 
Jan Kara <jack@...e.cz>
SUSE Labs, CR
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/