On Fri, Apr 09, 2021 at 08:01:15AM +0000, Thinh Nguyen wrote:
> Greg Kroah-Hartman wrote:
> > On Thu, Apr 08, 2021 at 06:42:32PM -0700, Thinh Nguyen wrote:
> >> We can check for host quirks at runtime base on the controller IP and
> >> version check. Set the following quirks for the DWC_usb31 IP host mode
> >> before creating a platform device for the xHCI driver:
> >>
> >> * XHCI_ISOC_BLOCKED_DISCONNECT
> >> * XHCI_LIMIT_FS_BI_INTR_EP
> >> * XHCI_LOST_DISCONNECT_QUIRK
> >>
> >> Signed-off-by: Thinh Nguyen <Thinh.Nguyen@xxxxxxxxxxxx>
> >> ---
> >> drivers/usb/dwc3/host.c | 21 +++++++++++++++++++++
> >> 1 file changed, 21 insertions(+)
> >>
> >> diff --git a/drivers/usb/dwc3/host.c b/drivers/usb/dwc3/host.c
> >> index f29a264635aa..a486d7fbb163 100644
> >> --- a/drivers/usb/dwc3/host.c
> >> +++ b/drivers/usb/dwc3/host.c
> >> @@ -9,6 +9,7 @@
> >>
> >> #include <linux/acpi.h>
> >> #include <linux/platform_device.h>
> >> +#include <linux/usb/xhci-quirks.h>
> >>
> >> #include "core.h"
> >>
> >> @@ -42,6 +43,17 @@ static int dwc3_host_get_irq(struct dwc3 *dwc)
> >> return irq;
> >> }
> >>
> >> +static void dwc3_host_init_quirks(struct dwc3 *dwc, struct xhci_plat_priv *priv)
> >> +{
> >> + memset(priv, 0, sizeof(*priv));
> >> +
> >> + if (DWC3_VER_IS_WITHIN(DWC31, ANY, 190A)) {
> >> + priv->quirks |= XHCI_ISOC_BLOCKED_DISCONNECT;
> >> + priv->quirks |= XHCI_LIMIT_FS_BI_INTR_EP;
> >> + priv->quirks |= XHCI_LOST_DISCONNECT_QUIRK;
> >> + }
> >> +}
> >> +
> >> int dwc3_host_init(struct dwc3 *dwc)
> >> {
> >> struct property_entry props[4];
> >> @@ -49,6 +61,7 @@ int dwc3_host_init(struct dwc3 *dwc)
> >> int ret, irq;
> >> struct resource *res;
> >> struct platform_device *dwc3_pdev = to_platform_device(dwc->dev);
> >> + struct xhci_plat_priv dwc3_priv;
> >
> > Tying the dwc3 code to the xhci code like this feels really wrong to me,
> > are you sure this is the correct resolution?
> >
> > greg k-h
> >
>
> Can you clarify what feels wrong? The way it's originally implemented
> already tied them in that way. What we're doing here simply takes
> advantage of what xhci-plat glue layer can use to set the xhci quirks.
> With this, we don't have to create new and duplicate DT properties for
> dwc3 and xhci to set some quirks. With the expanding list of dwc3 DT, I
> see this as a plus.
Ok, still feels odd, but I'll let the xhci maintainer make the decision
on it as it's their code to maintain, not mine :)
thanks,
greg k-h
