On Tue, Aug 9, 2022 at 6:47 AM Mathias Nyman <mathias.nyman@xxxxxxxxxxxxxxx> wrote: > > On 21.7.2022 0.53, Evan Green wrote: > > On Tue, Jun 14, 2022 at 3:06 AM Mathias Nyman > > <mathias.nyman@xxxxxxxxxxxxxxx> wrote: > >> > >> On 9.6.2022 18.08, Evan Green wrote: > >>> On Thu, Jun 9, 2022 at 12:58 AM Mathias Nyman > >>> <mathias.nyman@xxxxxxxxxxxxxxx> wrote: > >>>> > >>>> On 8.6.2022 16.43, Alan Stern wrote: > >>>>> On Wed, Jun 08, 2022 at 02:47:22PM +0300, Mathias Nyman wrote: > >>>>>> On 8.6.2022 11.19, Oliver Neukum wrote: > >>>>>>> > >>>>>>> > >>>>>>> On 07.06.22 15:58, Mathias Nyman wrote: > >>>>>>> > >>>>>>> Hi, > >>>>>>> > >>>>>>>> In shutdown (S5), with xHCI as host, this can be solved fairly easily > >>>>>>>> by turning off roothub port power in the .shutdown path. > >>>>>>> > >>>>>>> That would suck for the people who charge their phone from their > >>>>>>> computer. > >>>>>> > >>>>>> Good point. > >>>>>> My guess is that xHC port power bits won't actually turn off VBus for those > >>>>>> Sleep-and-charge, or Always-on ports. > >>>>>> VBus is allowed to be on even if port is in power-off state, but usb link state > >>>>>> should be forced to ss.Disabled from other states, like U3. > >>>>>> > >>>>>> Need to try it out, it's possible this turns off VBus for some usb-A ports > >>>>>> on some older systems that earlier (accidentally?) supplied VBus on > >>>>>> "normal" ports after shutdown. > >>>>> > >>>>> How about turning off port power _only_ in the shutdown or unbind path, > >>>>> and setting the port link states to ss.Disabled in the poweroff or > >>>>> poweroff_noirq stage of hibernation (if wakeup is disabled)? Would that > >>>>> solve the problem of the firmware needing to time out on reboot? > >>>>> > >>>> > >>>> That would be optimal, but unfortunately xHCI doesn't support setting link > >>>> state directly to ss.Disabled. Only way is to clear port power (PP) bit. > >>>> > >>>> To avoid turning off VBus in hibernate we could limit port power bit clearing > >>>> to xHC hosts that don't have the Port Power Control (PPC) capability flag. > >>>> > >>>> We know these xHC hosts don't control power switches, and clearing PP won't turn > >>>> off VBus (xhci 5.4.8, PORTRSC) > >>>> > >>>> This could be a solution for some hosts, but probably not cover all. > >>>> Not sure if the hardware this was reported on has PPC flag set. > >>> > >>> It appears it does not, HCCPARAMS1 for both USB controllers seems to > >>> be 0x20007fc1 (missing bit 3). You can check my work in case I made an > >>> error here: https://pastebin.com/9raZc63N > >>> -Evan > >> > >> Thanks, good to know. > >> So if disabling ports in hibernate doesn't work then we could turn off port power for > >> hosts with PPC==0. It should at least solve the issue for this particular system, > >> and not change current VBus policy in hibernate. > > > > Did a new version of this ever make it out? > > -Evan > > started writing a patch before vacation that sets port link to ss.Disabled instead of > turning off power power. > > Still has some FIXME lines and extra prints, but should be testable. > pushed to a fix_port_disable_s4 topic branch on top of 5.19: > > git://git.kernel.org/pub/scm/linux/kernel/git/mnyman/xhci.git fix_port_disable_s4 > https://git.kernel.org/pub/scm/linux/kernel/git/mnyman/xhci.git/log/?h=fix_port_disable_s4 Thanks! I was OOO last week, but I'm back and should be able to give it a try shortly. > > There was also one reported regression with powering off ports in shutdown S5, probably need > to sort that out before pushing this. Uh oh, I was just made aware of a regression as well on 4.19 that got narrowed down to that. I'm trying to learn more now. > > Thanks > -Mathias
