Re: [GIT, RFC] Killing the Big Kernel Lock

From: Frederic Weisbecker <fweisbec@gmail.com>
To: Roland Dreier <rdreier@cisco.com>
Cc: Arnd Bergmann <arnd@arndb.de>,
	linux-kernel@vger.kernel.org, Matthew Wilcox <matthew@wil.cx>,
	Thomas Gleixner <tglx@linutronix.de>,
	jblunck@suse.de, Alan Cox <alan@linux.intel.com>,
	Ingo Molnar <mingo@elte.hu>
Subject: Re: [GIT, RFC] Killing the Big Kernel Lock
Date: Thu, 1 Apr 2010 00:20:41 +0200	[thread overview]
Message-ID: <20100331222040.GG5163@nowhere> (raw)
In-Reply-To: <adazl1ogn0o.fsf@roland-alpha.cisco.com>

On Wed, Mar 31, 2010 at 03:11:03PM -0700, Roland Dreier wrote:
> Hi Arnd,
> 
> Looking at your tree, I see you have commit 753dd249 ("perf_event: use
> nonseekable_open") that does:
> 
>  > --- a/kernel/perf_event.c
>  > +++ b/kernel/perf_event.c
>  > @@ -2515,6 +2515,8 @@ static int perf_fasync(int fd, struct file *filp, int on)
>  >  }
>  >  
>  >  static const struct file_operations perf_fops = {
>  > +       .open                   = nonseekable_open,
>  > +       .llseek                 = no_llseek,
>  >         .release                = perf_release,
>  >         .read                   = perf_read,
>  >         .poll                   = perf_poll,
> 
> But if I understand this correctly, the assignment to .open is at best
> useless -- these file_operations are only used via anon_inode_getfd()
> and so there is no possible path that can call the .open method.  Or am
> I missing something?
> 
> (The same applies to the kvm_main.c changes too)

Good point, I'll update that.

Thanks.