Date: Wed, 15 Feb 2006 17:24:38 -0500 (EST)
From: Alan Stern <stern@rowland.harvard.edu>
To: James Bottomley <James.Bottomley@HansenPartnership.com>
cc: Kernel development list <linux-kernel@vger.kernel.org>,
       Greg KH <greg@kroah.com>
Subject: Re: [linux-usb-devel] Re: Linux 2.6.16-rc3
In-Reply-To: <1140033491.2883.11.camel@mulgrave.il.steeleye.com>
Message-ID: <Pine.LNX.4.44L0.0602151722250.4817-100000@iolanthe.rowland.org>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1322
Lines: 29

On Wed, 15 Feb 2006, James Bottomley wrote:

> On Wed, 2006-02-15 at 11:07 -0500, Alan Stern wrote:
> > Could we perhaps make this safer and more general?
> > 
> > For instance, add to struct device a "pending puts" counter and a list
> > header (both protected by a global spinlock), and have a kernel thread
> > periodically check the list, doing put_device wherever needed.  How does
> > that sound?
> 
> That's what I've been discussing with Jens elsewhere on this list.
> However, I think what you're proposing is overly complex.  All we really
> need is for a way of flagging a kobject (or kref) so the final put will
> be in user context.  Then we can use storage within the kobject or
> device (or something else) for the purpose.

That's more or less what I was suggesting.  The problems are: How do you 
know which put is the last?  (Answer: you don't.  So every put has to be 
done in process context.)  And how do you flag the data structure and tell 
some process thread to do the put?  (Answer: by putting the object on a 
list, as in my suggestion.)

Alan Stern

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/