On Tue, Oct 12, 2021 at 08:11:45PM +0800, zhuyinbo wrote:
>
> 在 2021/10/12 下午4:30, Yinbo Zhu 写道:
> > The usb states of ohci controller include UsbOperational, UsbReset,
> > UsbSuspend and UsbResume. Among them, only the UsbOperational state
> > supports launching the start-of-frame for host controller according
> > the ohci protocol spec, but in S3 and S4 (suspend to memory/suspend
> > to disk) press test procedure, it may happen that the start-of-
> > frame was launched in UsbSuspend status and cause ohci works failed
> > that the phenomenon was hc will allways reproduce the SoF interrupt
> > and consider that hc doesn't deal with the ed/td/done list in non-
> > UsbOperational, and this patch was to disable SoF interrupt in ohci
> > _rh_suspend so that it can fix ohci SoF abnormally interrupt issue.
> >
> > Signed-off-by: Yinbo Zhu <zhuyinbo@xxxxxxxxxxx>
> > ---
> > Change in v5:
> > Move the key code change into ohci_rh_suspend.
> > Rework the commit log information.
> >
> >
> > drivers/usb/host/ohci-hub.c | 2 ++
> > 1 file changed, 2 insertions(+)
> >
> > diff --git a/drivers/usb/host/ohci-hub.c b/drivers/usb/host/ohci-hub.c
> > index f474f2f..80a0094 100644
> > --- a/drivers/usb/host/ohci-hub.c
> > +++ b/drivers/usb/host/ohci-hub.c
> > @@ -88,6 +88,8 @@ static int ohci_rh_suspend (struct ohci_hcd *ohci, int autostop)
> > msleep (8);
> > spin_lock_irq (&ohci->lock);
> > }
> > + /* All ED unlinks should be finished, no need for SOF interrupts */
> > + ohci_writel(ohci, OHCI_INTR_SF, &ohci->regs->intrdisable);
> > update_done_list(ohci);
> > ohci_work(ohci);
>
> Hi Alan Stern,
>
> I consider that autostop sometimes was '1' then and HC doesn't stop
> deal with the ed/td/done list , then ohci_wok will called start_ed_unlink
> and
>
> start_ed_unlink will enable INTR_SF as follows, so I think add above two
> line code after ohci_work it is more appropriate, Do you think so?
>
> static void start_ed_unlink (struct ohci_hcd *ohci, struct ed *ed)
> {
>
> ...
>
> 494 /* enable SOF interrupt */
> 495 ohci_writel (ohci, OHCI_INTR_SF,
> &ohci->regs->intrstatus);
> 496 ohci_writel (ohci, OHCI_INTR_SF,
> &ohci->regs->intrenable);
>
> ...
>
> }
Yes, that's a good point. Go ahead and make this change. And be sure
to put a blank line before the new comment line.
I also think the patch description should be rewritten. It talks
about things that are not relevant to the patch. For example, you
don't _know_ that your host controller is sending SOF packets in the
non-UsbOperational state; you only know that it is generating
interrupts with the INTR_SF flag set.
So you should change the description to something like this:
While going into S3 or S4 suspend, an OHCI host controller can
generate interrupt requests if the INTR_SF enable flag is set. The
interrupt handler routine isn't prepared for this and it doesn't turn
off the flag, causing an interrupt storm.
To fix this problem, make ohci_rh_suspend() always disable INTR_SF
interrupts after processing the done list and the ED unlinks but
before the controller goes into the suspended (non-UsbOperational)
state. There's no reason to leave the flag enabled, since a suspended
controller doesn't generate Start-of-Frame packets.
Alan Stern
