Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S964921AbWISOZR (ORCPT ); Tue, 19 Sep 2006 10:25:17 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S964999AbWISOZR (ORCPT ); Tue, 19 Sep 2006 10:25:17 -0400 Received: from iolanthe.rowland.org ([192.131.102.54]:6926 "HELO iolanthe.rowland.org") by vger.kernel.org with SMTP id S964921AbWISOZQ (ORCPT ); Tue, 19 Sep 2006 10:25:16 -0400 Date: Tue, 19 Sep 2006 10:25:15 -0400 (EDT) From: Alan Stern X-X-Sender: stern@iolanthe.rowland.org To: Jiri Kosina cc: David Brownell , Andrew Morton , , , , Subject: Re: [linux-usb-devel] [PATCH] USB: consolidate error values from EHCI, UHCI and OHCI _suspend() In-Reply-To: Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2159 Lines: 47 On Tue, 19 Sep 2006, Jiri Kosina wrote: > On Mon, 18 Sep 2006, David Brownell wrote: > > > > EHCI, UHCI and OHCI USB host drivers are not consistent when returining > > > error values from their _suspend() functions, in case that the device is > > > not in suspended state. This could confuse users, so let all three of them > > > return -EBUSY. > > Shouldn't you also update uhci_suspend()? Currently it just ignores > > hcd->state ... > > You are right that the patch is possibly not fully correct. I was trying > to fix the situation I was getting into with the bug in usb_resume_both() > (see my "[PATCH] 2.6.18-rc6-mm2 - usb_resume_both() - fix suspend/resume" > mail from yesterday), but now it is obvious that the EINVAL from UHCI is > of a "different kind" than EBUSY from OHCI and UHCI (though they are > triggered in the same situations -- when the previous resume was not done > correctly). Sort of. The real trigger is that the PM core tries to suspend the upper-level PCI device but the lower-level root-hub device has not been suspended. You saw this occurring because during the previous resume, the root-hub device's power.power_state.event was not updated, causing the PM core to think the root hub was still suspended when in fact it wasn't. > As far as I can see, the UHCI driver is, strangely enough, not using > hcd->state at all. > > (by the way, EHCI and OHCI seem to have broken (read: missing) locking > when accessing the hcd->state. Should I fix it by per-hcd spinlock, or > does the patch already exist somewhere?) Believe it or not, these two paragraphs are closely connected. The reason uhci-hcd doesn't use hcd->state at all is because there is no locking to protect it! My feeling has long been that hcd->state deserves to disappear. It tries to serve too many functions. Please don't add any code in a futile attempt to preserve it. 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/