On Mon, Jan 13, 2025 at 04:14:07PM +0800, Yo-Jung (Leo) Lin wrote:
> Hi Alan
>
> On Fri, Jan 10, 2025 at 11:44 PM Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote:
> >
> > On Fri, Jan 10, 2025 at 04:44:10PM +0800, Yo-Jung (Leo) Lin wrote:
> > > The commit d9b4067aef50 ("USB: Fix the issue of task recovery failure
> > > caused by USB status when S4 wakes up") fixed an issue where if an USB
> > > port change happens during the entering steps of hibernation, xhci driver
> > > would attempt to resume the root hub, making the hibernation fail.
> > >
> > > System-wide suspend may fail due to the same reason, but this hasn't been
> > > addressed yet. This has been found on HP ProOne 440[1], as well as on
> > > some newer Dell all-in-one models. When suspend fails due to this reason,
> > > the kernel would show the following messages:
> >
> > I believe this problem was discussed on the mailing list before, and it
> > turned out that the issue was caused by a bug in the xhci-hcd driver,
> > not a bug in the USB core.
>
> Could you be more specific on which bug/thread it is?
> If you were mentioning thread about d9b4067aef50 ("USB: Fix the
> issue of task recovery failure caused by USB status when S4 wakes up"),
> the log in that commit message suggests that it happened on ehci, while
> here it happened on xhci. So this may be more general than just the xhci.
I was referring to the discussion in the email thread here:
https://lore.kernel.org/linux-usb/7be0c87a-c00f-4346-8482-f41ef0249b57@xxxxxxxxxxxxxxxxxxx/
> > Basically, suspend is _supposed_ to fail if a wakeup event occurs while
> > the suspend is in progress. As I recall, the bug in xhci-hcd was that
> > it treats some non-wakeup events as if they were wakeup events.
> >
> > In particular, a port change on the root hub should be treated as a
> > wakeup event if and only if the root hub is enabled for wakeup. Does
> > xhci-hcd check for this before failing the suspend?
> >
> > This reasoning shows that your proposed fix is incorrect.
> >
> Thanks for the feedback, This indeed isn't a correct way to address this.
> Will try to figure out some other ways.
Okay, good.
Alan Stern
