Hi Sarah and Alan, Thanks for the comments. I will make the following revisions: 1. Call pm_runtime_get_noresume only when the first device is connected, and call pm_runtime_put when the last device is disconnected. 2. Wrap the calls in an ifdef CONFIG_USB_DEFAULT_PERSIST. 3. Make sure that all pm_runtime_get_noresume calls have a corresponding pm_runtime_put somewhere (originally the intent was to disable runtime suspend "forever", but no longer). In principle, would the patch be acceptable with these revisions? On Sat, May 25, 2013 at 7:11 AM, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote: > On Fri, 24 May 2013, Sarah Sharp wrote: > >> On Fri, May 24, 2013 at 11:12:57AM -0700, Shawn Nematbakhsh wrote: >> > If a USB controller with XHCI_RESET_ON_RESUME goes to runtime suspend, >> > a reset will be performed upon runtime resume. Any previously suspended >> > devices attached to the controller will be re-enumerated at this time. >> > This will cause problems, for example, if an open system call on the >> > device triggered the resume (the open call will fail). >> >> That's ugly, but disabling runtime PM is going to have a big impact on >> the power consumption of those systems. >> >> Alan, do you think this is really the right thing to be doing here? It >> feels like userspace should just be able to deal with devices >> disconnecting on resume. After all, there are lots of USB devices that >> can't handle USB device suspend at all. > > This is a complicated issue. It depends on the runtime PM settings for > both the device and the host controller. > > As just one aspect, consider the fact that if it wants to, userspace > can already prevent the controller from going into runtime suspend. > Always preventing this at the kernel level, even when no devices are > plugged in, does seem too heavy-handed. > >> Shouldn't userspace just disable runtime PM for the USB device classes >> that don't have a reset resume callback? > > That's not so easy, because the kernel changes over time. Userspace > has no general way to tell which drivers have reset-resume support. > >> > Note that this change is only relevant when persist_enabled is not set >> > for USB devices. >> >> Could we at least wrap the call in an ifdef CONFIG_USB_DEFAULT_PERSIST? >> That way if people have USB persist turned off in their configuration, >> their host will still be able to suspend. > > Not just that; the patch is incorrect on the face of it... > >> > @@ -4687,6 +4687,12 @@ int xhci_gen_setup(struct usb_hcd *hcd, xhci_get_quirks_t get_quirks) >> > >> > get_quirks(dev, xhci); >> > >> > + /* If we are resetting upon resume, we must disable runtime PM. >> > + * Otherwise, an open() syscall to a device on our runtime suspended >> > + * controller will trigger controller reset and device re-enumeration */ >> > + if (xhci->quirks & XHCI_RESET_ON_RESUME) >> > + pm_runtime_get_noresume(dev); >> > + > > It adds a pm_runtime_get call with no corresponding pm_runtime_put. > > Alan Stern > -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
