Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20230425034010.3789376-1-quic_bjorande@quicinc.com>
 <20230425034010.3789376-6-quic_bjorande@quicinc.com> <ZFD8oQETtLuDH2Xg@hovoldconsulting.com>
 <20230504031354.GE870858@hu-bjorande-lv.qualcomm.com> <ZFNvHgaYsHUc2Y9L@hovoldconsulting.com>
In-Reply-To: <ZFNvHgaYsHUc2Y9L@hovoldconsulting.com>
From:   Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
Date:   Thu, 4 May 2023 11:55:11 +0300
Message-ID: <CAA8EJpr3VaVuXro6yqfUtAOrF9fcQgsy9kbSGhhrA6sTP5sLaA@mail.gmail.com>
Subject: Re: [PATCH 5/7] phy: qcom-qmp-combo: Introduce drm_bridge
To:     Johan Hovold <johan@kernel.org>
Cc:     Bjorn Andersson <quic_bjorande@quicinc.com>,
        Vinod Koul <vkoul@kernel.org>,
        Kishon Vijay Abraham I <kishon@kernel.org>,
        Bjorn Andersson <andersson@kernel.org>,
        Konrad Dybcio <konrad.dybcio@linaro.org>,
        Rob Herring <robh+dt@kernel.org>,
        Krzysztof Kozlowski <krzysztof.kozlowski+dt@linaro.org>,
        linux-arm-msm@vger.kernel.org, linux-phy@lists.infradead.org,
        devicetree@vger.kernel.org, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, 4 May 2023 at 11:38, Johan Hovold <johan@kernel.org> wrote:
>
> On Wed, May 03, 2023 at 08:13:54PM -0700, Bjorn Andersson wrote:
> > On Tue, May 02, 2023 at 02:05:53PM +0200, Johan Hovold wrote:
> > > On Mon, Apr 24, 2023 at 08:40:08PM -0700, Bjorn Andersson wrote:
> > > > The QMP combo PHY sits in an of_graph connected between the DisplayPort
> > > > controller and a USB Type-C connector (or possibly a redriver).
> > > >
> > > > The TCPM needs to be able to convey the HPD signal to the DisplayPort
> > > > controller, but no directly link is provided by DeviceTree so the signal
> > > > needs to "pass through" the QMP combo phy.
> > > >
> > > > Handle this by introducing a drm_bridge which upon initialization finds
> > > > the next bridge (i.e. the usb-c-connector) and chain this together. This
> > > > way HPD changes in the connector will propagate to the DisplayPort
> > > > driver.
> > > >
> > > > The connector bridge is resolved lazily, as the TCPM is expected to be
> > > > able to resolve the typec mux and switch at probe time, so the QMP combo
> > > > phy will probe before the TCPM.
> > > >
> > > > Signed-off-by: Bjorn Andersson <quic_bjorande@quicinc.com>
> > > > ---
> > > >  drivers/phy/qualcomm/phy-qcom-qmp-combo.c | 36 +++++++++++++++++++++++
> > > >  1 file changed, 36 insertions(+)
> > > >
> > > > diff --git a/drivers/phy/qualcomm/phy-qcom-qmp-combo.c b/drivers/phy/qualcomm/phy-qcom-qmp-combo.c
> > > > index 5d6d6ef3944b..84bc08002537 100644
> > > > --- a/drivers/phy/qualcomm/phy-qcom-qmp-combo.c
> > > > +++ b/drivers/phy/qualcomm/phy-qcom-qmp-combo.c
>
> > > > @@ -3196,6 +3200,34 @@ static int qmp_combo_register_clocks(struct qmp_combo *qmp, struct device_node *
> > > >   return devm_add_action_or_reset(qmp->dev, phy_clk_release_provider, dp_np);
> > > >  }
> > > >
> > > > +static int qmp_combo_bridge_attach(struct drm_bridge *bridge,
> > > > +                            enum drm_bridge_attach_flags flags)
> > > > +{
> > > > + struct qmp_combo *qmp = container_of(bridge, struct qmp_combo, bridge);
> > > > + struct drm_bridge *next_bridge;
> > > > +
> > > > + if (!(flags & DRM_BRIDGE_ATTACH_NO_CONNECTOR))
> > > > +         return -EINVAL;
> > > > +
> > > > + next_bridge = devm_drm_of_get_bridge(qmp->dev, qmp->dev->of_node, 0, 0);
> > > > + if (IS_ERR(next_bridge))
> > > > +         return dev_err_probe(qmp->dev, PTR_ERR(next_bridge), "failed to acquire drm_bridge\n");
> > >
> > > Using dev_err_probe() in an attach callback looks wrong as these
> > > functions should not be returning -EPROBE_DEFER (and this is not a probe
> > > function).
> >
> > The problem is that this might return EPROBE_DEFER, and at least today
> > propagates out to returning EPROBE_DEFER from our DP controller's
> > bind().
>
> Due to the known issue with the MSM driver panel lookup, or due to some
> more fundamental problem with the stack?

Ideally MSM DP driver should call component_add() only when the next
bridge is available. This is how we handle it for the DSI case.
However I'm yet to see the changes to dp_display_probe() which make
actual use of the done_probing callback. And even that will only fix
the eDP case. For the normal DP case we have no way of being properly
notified when the next bridge becomes available. So the driver will
try to drm_bridge_attach() from the component's bind() callback and
return an error if the chain is not (yet) fully available.

>
> At least in the former case, I don't think we should hide the fact that
> we have an unresolved issue with the MSM driver this way even if it
> means printing an extra error message until it has been resolved (cf.
> the panel lookup errors that we've intentionally kept in place).
>
> > This is not optimal, but unfortunately we have a two way dependency
> > across the of_graph, so we need to make one of the sides lazy...
>
> But this comments seems to suggest this is a bigger issue than the panel
> lookup.
>
> Could you describe the issue in some more detail (e.g. when would you
> see -EPROBE_DEFER here)?
>
> Johan


-- 
With best wishes
Dmitry