Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752736AbbLXPOQ (ORCPT ); Thu, 24 Dec 2015 10:14:16 -0500 Received: from netrider.rowland.org ([192.131.102.5]:41363 "HELO netrider.rowland.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with SMTP id S1751503AbbLXPOO (ORCPT ); Thu, 24 Dec 2015 10:14:14 -0500 Date: Thu, 24 Dec 2015 10:14:13 -0500 (EST) From: Alan Stern X-X-Sender: stern@netrider.rowland.org To: Oliver Neukum cc: Hayes Wang , "linux-kernel@vger.kernel.org" , "linux-usb@vger.kernel.org" , "netdev@vger.kernel.org" , "peter@lekensteyn.nl" Subject: Re: [PATCH v2] r8152: fix lockup when runtime PM is enabled In-Reply-To: <1450941280.28243.8.camel@suse.com> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2347 Lines: 54 On Thu, 24 Dec 2015, Oliver Neukum wrote: > On Wed, 2015-12-23 at 20:32 -0500, Alan Stern wrote: > > > I don't understand why the wakeup conditions are different. It seems > > to me that the choice of which packets will generate a wakeup ought to > > depend on the user's selection, not on the kind of suspend. For > > instance, if the user says that only a magic packet should cause a > > wakeup then that should be true for both runtime suspend and system > > suspend. > > > > To put it another way, as far as the device is concerned a suspend is > > just a suspend -- there's no different between a runtime suspend and a > > system suspend. > > This literally true, but the host and the driver care. > If we autosuspend a running network device, any packet > (maybe filtered for MAC) should cause a remote wake up, > else we'd lose packets. That's also true during system suspend. > But you cannot keep that setting if the system goes down > or any broadcast packet would resume the whole system. > Yet you cannot just disable remote wake up, as WoL packages > still must trigger a remote wake up. This means that sometimes you want to avoid losing packets and other times you do want to lose packets. That is a policy decision, and therefore it should be made by the user, not the kernel. > So there are drivers which must change settings on devices > as the system goes to sleep, even if their devices have > already been autosuspended. We could use the notifier chains > for that. But can this solution be called elegant? Instead of the driver trying to do this automatically, you could rely on userspace telling the driver which packets should cause a wakeup. The setting could be updated immediately before and after each system suspend. I admit this is more awkward than having the driver make a choice based on the type of suspend. This is a case where the resources provided by the PM core aren't adequate for what the driver needs. The PM core distinguishes between wakeup enabled or disabled; it doesn't distinguish among different levels of wakekup. 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/