Re: [PATCH v2] usb: host: xhci-plat: fix support for XHCI_SKIP_PHY_INIT quirk

["Pali Rohár" <pali@kernel.org>]

On Tuesday 26 January 2021 04:27:37 Yoshihiro Shimoda wrote:
> Hi Pali,
> 
> > From: Pali Rohár, Sent: Monday, January 25, 2021 11:20 PM
> > On Friday 15 January 2021 15:32:30 Mathias Nyman wrote:
> > > On 14.1.2021 1.20, Pali Rohár wrote:
> > > > On Thursday 24 December 2020 05:59:05 Peter Chen wrote:
> > > >> On 20-12-23 17:18:47, Pali Rohár wrote:
> > > >>> Currently init_quirk callbacks for xhci platform drivers are called
> > > >>> xhci_plat_setup() function which is called after chip reset completes.
> > > >>> It happens in the middle of the usb_add_hcd() function.
> > > >>>
> > > >>> But XHCI_SKIP_PHY_INIT quirk is checked in the xhci_plat_probe() function
> > > >>> prior calling usb_add_hcd() function. Therefore this XHCI_SKIP_PHY_INIT
> > > >>> currently does nothing as prior xhci_plat_setup() it is not set.
> > > >>>
> > > >>> Quirk XHCI_SKIP_PHY_INIT is only setting hcd->skip_phy_initialization value
> > > >>> which really needs to be set prior calling usb_add_hcd() as this function
> > > >>> at its beginning skips PHY init if this member is set.
> > > >>>
> > > >>> This patch fixes implementation of the XHCI_SKIP_PHY_INIT quirk by calling
> > > >>> init_quirk callbacks (via xhci_priv_init_quirk()) prior checking if
> > > >>> XHCI_SKIP_PHY_INIT is set. Also checking if either xhci->quirks or
> > > >>> priv->quirks contains this XHCI_SKIP_PHY_INIT quirk.
> > > >>>
> > > >>> Signed-off-by: Pali Rohár <pali@kernel.org>
> > > >>>
> > > >>> ---
> > > >>> Changes in v2:
> > > >>> * Check also xhci->quirks as xhci_priv_init_quirk() callbacks are setting xhci->quirks
> > > >>> * Tested with "usb: host: xhci: mvebu: make USB 3.0 PHY optional for Armada 3720" patch
> > > >>> * Removed Fixes: line
> > > >>> ---
> > > >>>  drivers/usb/host/xhci-plat.c | 16 ++++++++--------
> > > >>>  1 file changed, 8 insertions(+), 8 deletions(-)
> > > >>>
> > > >>> diff --git a/drivers/usb/host/xhci-plat.c b/drivers/usb/host/xhci-plat.c
> > > >>> index 4d34f6005381..0eab7cb5a767 100644
> > > >>> --- a/drivers/usb/host/xhci-plat.c
> > > >>> +++ b/drivers/usb/host/xhci-plat.c
> > > >>> @@ -89,13 +89,6 @@ static void xhci_plat_quirks(struct device *dev, struct xhci_hcd *xhci)
> > > >>>  /* called during probe() after chip reset completes */
> > > >>>  static int xhci_plat_setup(struct usb_hcd *hcd)
> > > >>>  {
> > > >>> -	int ret;
> > > >>> -
> > > >>> -
> > > >>> -	ret = xhci_priv_init_quirk(hcd);
> > > >>> -	if (ret)
> > > >>> -		return ret;
> > > >>> -
> > > >>>  	return xhci_gen_setup(hcd, xhci_plat_quirks);
> > > >>>  }
> > > >>>
> > > >>> @@ -330,7 +323,14 @@ static int xhci_plat_probe(struct platform_device *pdev)
> > > >>>
> > > >>>  	hcd->tpl_support = of_usb_host_tpl_support(sysdev->of_node);
> > > >>>  	xhci->shared_hcd->tpl_support = hcd->tpl_support;
> > > >>> -	if (priv && (priv->quirks & XHCI_SKIP_PHY_INIT))
> > > >>> +
> > > >>> +	if (priv) {
> > > >>> +		ret = xhci_priv_init_quirk(hcd);
> > > >>> +		if (ret)
> > > >>> +			goto disable_usb_phy;
> > > >>> +	}
> > > >>> +
> > > >>> +	if ((xhci->quirks & XHCI_SKIP_PHY_INIT) || (priv && (priv->quirks & XHCI_SKIP_PHY_INIT)))
> > > >>>  		hcd->skip_phy_initialization = 1;
> > > >>
> > > >> I am not sure if others agree with you move the position of
> > > >> xhci_priv_init_quirk, Let's see Mathias opinion.
> > > >
> > > > Hello! Do you have an opinion how to handle this issue? As currently it
> > > > is needed for another patch which is fixing issue/regression in xhci-mvebu:
> > > >
> <snip>
> > > >
> > >
> > > I can see the benefit in this.
> > > In the xhci-plat case usb_create_hcd and usb_add_hcd are separate steps, and
> > > we could both copy the xhci_plat_priv .quirks and run the .init_qurks before
> > > adding the hcd.
> > > I guess the current way is inherited from pci case where the earliest place
> > > to do this after hcd is created is the hcd->driver->reset callback (which is
> > > set to xhci_pci_setup() or xhci_plat_setup()).
> > >
> > > xhci-rcar.c is using the .init_quirk to load firmware, we need to check with
> > > them if this change is ok. (added Yoshihiro Shimoda to cc)
> > 
> > Yoshihiro, is this change OK?
> > 
> > Can we move forward? I really need to now how to handle regression in
> > xhci-mvebu driver. And one option is with this patch...
> 
> Thank you for asking me about this topic. I tested the patch, but unfortunately,
> this patch is possible to break a rcar platform because a phy initialization is
> needed before the firmware loading if the platform uses the phy. (Note that
> upstream code (salvator-common.dtsi) doesn't use the phy for xhci. But,
> if we use the phy on other board with this patch, the xhci will not work.)
> 
> So, I think we need to add a new function pointer for your case.

Ok, thank you for testing! I will try to come up with other solution to
mentioned mvebu-xhci issue.

> Best regards,
> Yoshihiro Shimoda
> 
> > > Their firmware would be loaded before phy parts are initialized, usb bus
> > > registered, or roothub device allocated.
> > >
> > > Thanks
> > > -Mathias