linux-kernel - Re: [PATCH v6 1/8] drm/msm/dp: Add eDP support via aux

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAD=FV=XS8h8GwL11CmJou31hTbf=mS77-j1y66uY+2YuGAUpzQ@mail.gmail.com>
Date:   Fri, 8 Apr 2022 06:56:08 -0700
From:   Doug Anderson <dianders@...omium.org>
To:     Dmitry Baryshkov <dmitry.baryshkov@...aro.org>
Cc:     Abhinav Kumar <quic_abhinavk@...cinc.com>,
        "Sankeerth Billakanti (QUIC)" <quic_sbillaka@...cinc.com>,
        quic_kalyant <quic_kalyant@...cinc.com>,
        "open list:OPEN FIRMWARE AND FLATTENED DEVICE TREE BINDINGS" 
        <devicetree@...r.kernel.org>,
        quic_vproddut <quic_vproddut@...cinc.com>,
        David Airlie <airlied@...ux.ie>,
        linux-arm-msm <linux-arm-msm@...r.kernel.org>,
        "Kuogee Hsieh (QUIC)" <quic_khsieh@...cinc.com>,
        freedreno <freedreno@...ts.freedesktop.org>,
        dri-devel <dri-devel@...ts.freedesktop.org>,
        "bjorn.andersson@...aro.org" <bjorn.andersson@...aro.org>,
        Sean Paul <seanpaul@...omium.org>,
        "Aravind Venkateswaran (QUIC)" <quic_aravindh@...cinc.com>,
        Stephen Boyd <swboyd@...omium.org>,
        Sean Paul <sean@...rly.run>,
        LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v6 1/8] drm/msm/dp: Add eDP support via aux_bus

Hi,

On Fri, Apr 8, 2022 at 5:13 AM Dmitry Baryshkov
<dmitry.baryshkov@...aro.org> wrote:
>
> On Fri, 8 Apr 2022 at 03:28, Doug Anderson <dianders@...omium.org> wrote:
> >
> > Hi,
> >
> > On Thu, Apr 7, 2022 at 4:36 PM Dmitry Baryshkov
> > <dmitry.baryshkov@...aro.org> wrote:
> > >
> > > The ps8640 driver looks 'working by coincidence'. It calls
> > > dp_aux_populate, then immediately after the function returns it checks
> > > for the panel. If panel-edp is built as a module, the probe might fail
> > > easily.
> > > The anx7625 driver has the same kind of issue. The DP AUX bus is
> > > populated from the probe() and after some additional work the panel is
> > > being checked.
> > > This design is fragile and from my quick glance it can break (or be
> > > broken) too easy. It reminds me of our drm msm 'probe' loops
> > > preventing the device to boot completely if the dsi bridge/panel could
> > > not be probed in time.
> >
> > I did spend some time thinking about this, at least for ps8640. I
> > believe that as long as the panel's probe isn't asynchronous.
>
> By panel probe (as a probe of any device) is potentially asynchronous.
> As in your example, the PWM might not be present, the regulator probe
> might have been delayed, the panel-edp might be built as a module,
> which is not present for some reason.

Well, in those cases it's not exactly asynchronous, or at least not in
the way I was thinking about. Either it will work now, or we should
try again later when more drivers have probed. The case I was worried
about was:

1. We call devm_of_dp_aux_populate_ep_devices()

2. devm_of_dp_aux_populate_ep_devices() kicks off a probe to the panel
in the background

3. devm_of_dp_aux_populate_ep_devices() returns to us without waiting
for the panel's probe to finish.

I think that's possible if the panel driver sets PROBE_PREFER_ASYNCHRONOUS.

I was less worried about the resources missing since I thought that
would be handled by the normal probe deferral case. IIRC the "infinite
loop" that we used to end up with MSM's probe was because something
about the way the MSM DRM driver worked would cause the deferral
mechanisms to retry instantly each time we deferred. I don't remember
exactly what triggered it, but I don't _think_ that's true for ps8640?


> > Basically if the panel isn't ready then ps8640 should return and we'll
> > retry later. I do remember the probe loops that we used to have with
> > msm and I don't _think_ this would trigger it.
>
> I don't have proof here. But as I wrote, this thing might break at some point.
>
> > That being said, if we need to separate out the AUX bus into a
> > sub-device like we did in sn65dsi86 we certainly could.
>
> Ideally we should separate the "bridge" subdevice, like it was done in
> sn65dsi86.
> So that the aux host probes, registers the EP device, then the bridge
> device can not be probed (and thus the drm_bridge is not created)
> until the next  bridge becomes available.

You're definitely right, that's best. I was hesitant to force the
issue during review of the ps8640 because it adds a bunch of
complexity and didn't _seem_ to be needed. Maybe it makes sense to
just code it up, though...

-Doug