Re: [PATCH 2.5.6] New wireless driver API part 2

[Jean Tourrilhes <jt@bougret.hpl.hp.com>]

On Mon, Mar 11, 2002 at 04:50:11PM -0500, Jeff Garzik wrote:
> Jean Tourrilhes wrote:
> 
> Overall looks good.  My only minor objection would be that this function 
> should return an error value.  Clearly the kmalloc can fail, at least.
> 
>     Jeff

	Thanks for the quick review (as usual), very much appreciated.

	Now, for the return value...
	I've debated this precise point. Here is the comment that I
wrote in the code you just quoted :
		/* Note : we don't return an error to the driver, because
		 * the driver would not know what to do about it. It can't
		 * return an error to the user, because the event is not
		 * initiated by a user request.
		 * The best the driver could do is to log an error message.
		 * We will do it ourselves instead...
		 */
	The failure to deliver an event to the user is not critical,
and I don't really see what the driver code would do with a return
code. In fact, event delivery to user space is not reliable (netlink
may drop it in case its queues are full - this is more likely than
kmalloc failure), and my code only check a few of those failure
conditions, so the driver has no way to know if the message reached
its intended destination.
	In fact, I eliminated the return code *on purpose*, to prevent
driver writer to do stupid things (like shutting down the driver) or
adding additional log message (waste at this point).
	Convincing enough ?

	Have fun...

	Jean