On Friday, August 05, 2011, Rafael J. Wysocki wrote: > On Friday, August 05, 2011, Alan Stern wrote: > > On Fri, 5 Aug 2011, Rafael J. Wysocki wrote: > > > > > > > I don't know how to fix it yet, but I think I know what the problem is. > > > > > Namely, a runtime suspend of the Ethernet adapter as occured in parallel with > > > > > the system-wide suspend and they clashed. The runtime suspend has probably > > > > > been provoked by detaching the Ethernet cable from the box. > > > > > > > > For them to clash in that way would mean that the PM workqueue didn't > > > > get frozen. > > > > > > Not necessarily, I think. It's sufficient that system suspend is started > > > while the runtime PM operation is in progress. > > > > How so? The system suspend doesn't call down to the subsystems or > > drivers until after all the threads are frozen. In this case the stack > > dump shows that the runtime PM operation was asynchronous, running in a > > workqueue thread (pm_runtime_work is one of the routines near the end > > of the stack dump). The thread wouldn't have frozen until the runtime > > PM operation was complete. > > You're right, sorry. System suspend was not involved, so the problem > was triggerend by runtime suspend alone, resulting from unplugging the > Ethernet cable. > > It looks like the crash started in pci_disable_msi(). OTOH, I'm not sure how pci_legacy_suspend_late() was called. In theory it is only called by the system suspend code and it surely is not called for e1000e. Thanks, Rafael _______________________________________________ linux-pm mailing list linux-pm@xxxxxxxxxxxxxxxxxxxxxxxxxx https://lists.linux-foundation.org/mailman/listinfo/linux-pm
