[PATCH v5] usb: xhci-mtk: fix unreleased bandwidth data

[Ikjoon Jang]

On Thu, Jan 14, 2021 at 4:30 PM Chunfeng Yun <chunfeng.yun at mediatek.com> wrote:
>
> Hi Ikjoon,
>
> On Tue, 2021-01-12 at 13:48 +0800, Ikjoon Jang wrote:
> > On Fri, Jan 8, 2021 at 10:44 PM Mathias Nyman
> > <mathias.nyman at linux.intel.com> wrote:
> > >
> > > On 8.1.2021 8.11, Chunfeng Yun wrote:
> > > > On Thu, 2021-01-07 at 13:09 +0200, Mathias Nyman wrote:
> > > >> On 29.12.2020 8.24, Ikjoon Jang wrote:
> > > >>> xhci-mtk has hooks on add_endpoint() and drop_endpoint() from xhci
> > > >>> to handle its own sw bandwidth managements and stores bandwidth data
> > > >>> into internal table every time add_endpoint() is called,
> > > >>> so when bandwidth allocation fails at one endpoint, all earlier
> > > >>> allocation from the same interface could still remain at the table.
> > > >>>
> > > >>> This patch adds two more hooks from check_bandwidth() and
> > > >>> reset_bandwidth(), and make mtk-xhci to releases all failed endpoints
> > > >>> from reset_bandwidth().
> > > >>>
> > > >>> Fixes: 08e469de87a2 ("usb: xhci-mtk: supports bandwidth scheduling with multi-TT")
> > > >>> Signed-off-by: Ikjoon Jang <ikjn at chromium.org>
> > > >>>
> > > >>
> > > >> ...
> > > >>
> > > >>>
> > > >>> diff --git a/drivers/usb/host/xhci.c b/drivers/usb/host/xhci.c
> > > >>> index d4a8d0efbbc4..e1fcd3cf723f 100644
> > > >>> --- a/drivers/usb/host/xhci.c
> > > >>> +++ b/drivers/usb/host/xhci.c
> > > >>> @@ -2882,6 +2882,12 @@ static int xhci_check_bandwidth(struct usb_hcd *hcd, struct usb_device *udev)
> > > >>>     xhci_dbg(xhci, "%s called for udev %p\n", __func__, udev);
> > > >>>     virt_dev = xhci->devs[udev->slot_id];
> > > >>>
> > > >>> +   if (xhci->quirks & XHCI_MTK_HOST) {
> > > >>> +           ret = xhci_mtk_check_bandwidth(hcd, udev);
> > > >>> +           if (ret < 0)
> > > >>> +                   return ret;
> > > >>> +   }
> > > >>> +
> > > >>
> > > >> Just noticed that XHCI_MTK_HOST quirk is only set in xhci-mtk.c.
> > > >> xhci-mtk.c calls xhci_init_driver(..., xhci_mtk_overrides) with a .reset override function.
> > > >>
> > > >> why not add override functions for .check_bandwidth and .reset_bandwidth to xhci_mtk_overrides instead?
> > > >>
> > > >> Another patch to add similar overrides for .add_endpoint and .drop_endpoint should probably be
> > > >> done so that we can get rid of the xhci_mtk_add/drop_ep_quirk() calls in xhci.c as well
> > > > You mean, we can export xhci_add/drop_endpoint()?
> > >
> > > I think so, unless you have a better idea.
> > > I prefer exporting the generic add/drop_endpoint functions rather than the vendor specific quirk functions.
> > >
> >
> > When moving out all MTK_HOST quirks and unlink xhci-mtk-sch from xhci,
> > xhci-mtk-sch still needs to touch the xhci internals, at least struct
> > xhci_ep_ctx.
> >
> > My naive idea is just let xhci export one more function to expose xhci_ep_ctx.
> > But I'm not sure whether this is acceptable:
> I find that xhci_add_endpoint() ignores some errors with return 0, for
> these cases we needn't call xhci_mtk_add_ep-quirk(), so may be not a
> good way to just export xhci_add_endpoint().

yeah, maybe that's from ep0 case?

And I've thought that we could also unlink xhci-mtk-sch from the xhci module
if MTK_HOST quirk functions are moved out to mtk platform driver's overrides.
I guess I've gone too far.

If we keep xhci-mtk-sch being built with the xhci module,
xhci-mtk-sch can directly access input control context and its drop/add flags,
so I think we can simply remove {add|drop}_endpoint() quirks and just handle
them all in {check|reset}_bandwidth() overrides.

>
> >
> > +struct xhci_ep_ctx* xhci_get_ep_contex(struct xhci_hcd *xhci, struct
> > usb_host_endpoint *ep)
> > +{ ... }
> > +EXPORT_SYMBOL(xhci_get_ep_context);
> >
> > But for v6, I'm going to submit a patch with {check|reset}_bandwidth()
> > quirk function
> >  switched into xhci_driver_overrides first. (and preserve existing
> > MTK_HOST quirk functions).
> >
> > Thanks!
> >
> > > -Mathias
> > >
>