Re: [PATCH 2/3] usb: xhci-mtk: fix UAS issue by XHCI_BROKEN_STREAMS quirk

[Chunfeng Yun]

On Mon, 2020-12-21 at 12:35 -0700, Rob Herring wrote:
> On Thu, Dec 17, 2020 at 02:23:50PM +0800, Chunfeng Yun wrote:
> > On Thu, 2020-12-17 at 11:32 +0800, Nicolas Boichat wrote:
> > > On Thu, Dec 17, 2020 at 10:19 AM Chunfeng Yun <chunfeng.yun@xxxxxxxxxxxx> wrote:
> > > >
> > > > On Wed, 2020-12-16 at 20:28 +0800, Nicolas Boichat wrote:
> > > > > On Wed, Dec 16, 2020 at 7:53 PM Chunfeng Yun <chunfeng.yun@xxxxxxxxxxxx> wrote:
> > > > > >
> > > > > > The 0.96 xHCI controller on some platforms does not support
> > > > > > bulk stream even HCCPARAMS says supporting, due to MaxPSASize
> > > > > > is set a non-zero default value by mistake, here use
> > > > > > XHCI_BROKEN_STREAMS quirk to fix it.
> > > > > >
> > > > > > Fixes: 94a631d91ad3 ("usb: xhci-mtk: check hcc_params after adding primary hcd")
> > > > > > Signed-off-by: Chunfeng Yun <chunfeng.yun@xxxxxxxxxxxx>
> > > > > > ---
> > > > > >  drivers/usb/host/xhci-mtk.c | 7 ++++++-
> > > > > >  drivers/usb/host/xhci-mtk.h | 1 +
> > > > > >  2 files changed, 7 insertions(+), 1 deletion(-)
> > > > > >
> > > > > > diff --git a/drivers/usb/host/xhci-mtk.c b/drivers/usb/host/xhci-mtk.c
> > > > > > index 8f321f39ab96..08dab974d847 100644
> > > > > > --- a/drivers/usb/host/xhci-mtk.c
> > > > > > +++ b/drivers/usb/host/xhci-mtk.c
> > > > > > @@ -395,6 +395,9 @@ static void xhci_mtk_quirks(struct device *dev, struct xhci_hcd *xhci)
> > > > > >         xhci->quirks |= XHCI_SPURIOUS_SUCCESS;
> > > > > >         if (mtk->lpm_support)
> > > > > >                 xhci->quirks |= XHCI_LPM_SUPPORT;
> > > > > > +
> > > > > > +       if (mtk->broken_streams)
> > > > > > +               xhci->quirks |= XHCI_BROKEN_STREAMS;
> > > > > >  }
> > > > > >
> > > > > >  /* called during probe() after chip reset completes */
> > > > > > @@ -460,6 +463,8 @@ static int xhci_mtk_probe(struct platform_device *pdev)
> > > > > >                 return ret;
> > > > > >
> > > > > >         mtk->lpm_support = of_property_read_bool(node, "usb3-lpm-capable");
> > > > > > +       mtk->broken_streams =
> > > > > > +               of_property_read_bool(node, "mediatek,broken_streams_quirk");
> > > > >
> > > > > Would it be better to add a data field to struct of_device_id
> > > > > mtk_xhci_of_match, and enable this quirk on mediatek,mt8173-xhci only?
> > > > This is the common issue for all SoCs (before 2016.06) with 0.96 xHCI
> > > > when the controller don't support bulk stream. If enable this quirk only
> > > > for mt8173, then for other SoCs, the compatible need include
> > > > "mediatek,mt8173-xhci" in dts, this may be not flexible for some cases,
> > > > e.g. a new SoC has the broken stream as mt8173, but also has another
> > > > different quirk, the way you suggested will not handle it.
> > > 
> > > It can, we do this regularly for many other components. One example:
> > > https://elixir.bootlin.com/linux/latest/source/drivers/i2c/busses/i2c-mt65xx.c#L402
> > > 
> > Got it. Indeed works when add compatible private data.
> > 
> > Due to many SoCs supports USB and not upstream, I'd prefer to avoid
> > adding new compatible in driver when support new SoCs, and leave the
> > code as simple as possible.
> 
> No. The problem is adding new fixes requires updating the DT. That would 
> be fine if you knew all possible issues and quirks up front. You may for 
> some, but you won't catch them all.
Yes.
But all SoCs use the generic compatible "mediatek,mtk-xhci" except
mtk8173. as following:
compatible = "mediatek,mt<SoC>-xhci", "mediatek,mtk-xhci";

It may be a little messy when the driver fill many compatibles after
supporting several tens or hundreds SoCs.

I'll think about it again. 
> 
> Save DT properties for per board quirks/features, not per SoC ones.
This is for USB IP quirks/feature, not board or SoC

Thanks

> 
> > 
> > > > And I plan to remove "mediatek,mt8173-xhci" in mtk_xhci_of_match after
> > > > converting the binding to YMAL.
> > > >
> > > > >
> > > > > (IMHO usb3-lpm-capable detection should also be done in the same way)
> 
> I tend to agree, but am more tolerable of standard USB features than 
> specific IP block quirks.
> 
> Rob