linux-kernel - Re: [PATCH] USB: Don't set USB_PORT_FEAT

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CABTNMG21xp6TA8SGJhamfM9D6JGvQHwg8AMySSCh09-DnAZ5qQ@mail.gmail.com>
Date:   Wed, 14 Apr 2021 13:07:43 +0800
From:   Chris Chiu <chris.chiu@...onical.com>
To:     Alan Stern <stern@...land.harvard.edu>
Cc:     gregkh@...uxfoundation.org, m.v.b@...box.com, hadess@...ess.net,
        linux-usb@...r.kernel.org,
        Linux Kernel <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] USB: Don't set USB_PORT_FEAT_SUSPEND on WD19's Realtek Hub

On Tue, Apr 13, 2021 at 10:44 PM Alan Stern <stern@...land.harvard.edu> wrote:
>
> On Tue, Apr 13, 2021 at 03:52:14PM +0800, Chris Chiu wrote:
> > On Mon, Apr 12, 2021 at 11:12 PM Alan Stern <stern@...land.harvard.edu> wrote:
> > >
> > > On Mon, Apr 12, 2021 at 11:00:06PM +0800, chris.chiu@...onical.com wrote:
> > > > The USB_PORT_FEAT_SUSPEND is not really necessary due to the
> > > > "global suspend" in USB 2.0 spec. It's only for many hub devices
> > > > which don't relay wakeup requests from the devices connected to
> > > > downstream ports. For this realtek hub, there's no problem waking
> > > > up the system from connected keyboard.
> > >
> > > What about runtime suspend?  That _does_ require USB_PORT_FEAT_SUSPEND.
> >
> > It's hard to reproduce the same thing with runtime PM. I also don't
> > know the aggressive
> > way to trigger runtime suspend. So I'm assuming the same thing will happen in
> > runtime PM case because they both go the same usb_port_resume path. Could
> > you please suggest a better way to verify this for runtime PM?
>
> To put a USB device into runtime suspend, do this:
>
>         echo 0 >/sys/bus/usb/devices/.../bConfigurationValue
>         echo auto >/sys/bus/usb/devices/.../power/control
>
> where ... is the pathname for the device you want to suspend.  (Note
> that this will unbind the device from its driver, so make sure there's
> no possibility of data loss before you do it.)
>
> To resume the device, write "on" to the power/control file.  You can
> verify the runtime-PM status by reading the files in the power/
> subdirectory.
>
Thanks for the instructions. I can hit the same timeout problem with
runtime PM. The
fail rate seems the same as normal PM. (around 1/4 ~ 1/7)
root@:/sys/bus/usb/devices/3-4.3# echo auto > power/control
root@:/sys/bus/usb/devices/3-4.3# echo on > power/control
root@:/sys/bus/usb/devices/3-4.3# dmesg -c
[ 2789.679807] usb 3-4: kworker/7:0 timed out on ep0out len=0/0
[ 2789.679812] usb 3-4-port3: can't suspend, status -110
[ 2789.680078] usb 3-4.3: Failed to suspend device, error -110


> > > > This commit bypasses the USB_PORT_FEAT_SUSPEND for the quirky hub.
> > > >
> > > > Signed-off-by: Chris Chiu <chris.chiu@...onical.com>
> > > > ---
> > >
> > >
> > > > diff --git a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
> > > > index 7f71218cc1e5..8478d49bba77 100644
> > > > --- a/drivers/usb/core/hub.c
> > > > +++ b/drivers/usb/core/hub.c
> > > > @@ -3329,8 +3329,11 @@ int usb_port_suspend(struct usb_device *udev, pm_message_t msg)
> > > >        * descendants is enabled for remote wakeup.
> > > >        */
> > > >       else if (PMSG_IS_AUTO(msg) || usb_wakeup_enabled_descendants(udev) > 0)
> > > > -             status = set_port_feature(hub->hdev, port1,
> > > > -                             USB_PORT_FEAT_SUSPEND);
> > > > +             if (udev->quirks & USB_QUIRK_NO_SET_FEAT_SUSPEND)
> > >
> > > You should test hub->hdev->quirks, here, not udev->quirks.  The quirk
> > > belongs to the Realtek hub, not to the device that's plugged into the
> > > hub.
> > >
> >
> > Thanks for pointing that out. I'll verify again and propose a V2 after
> > it's done.
>
> Another thing to consider: You shouldn't return 0 from usb_port_suspend
> if the port wasn't actually suspended.  We don't want to kernel to have
> a false idea of the hardware's current state.
>
So we still need the "really_suspend=false". What if I replace it with
the following?
It's a little verbose but expressive enough. Any suggestions?

+       else if (!(hub->hdev->quirks & USB_QUIRK_NO_SET_FEAT_SUSPEND) &&
+               (PMSG_IS_AUTO(msg) || usb_wakeup_enabled_descendants(udev) > 0))
+               status = set_port_feature(hub->hdev, port1,
+                               USB_PORT_FEAT_SUSPEND);
        else {
                really_suspend = false;
                status = 0;

Chris

> Alan Stern