RE: connector_bad_edid() is broken (was: Re: [PATCH] drm/edid: Fix crash with zero/invalid EDID)

[Zuo, Jerry]

[AMD Official Use Only]

> -----Original Message-----
> From: Doug Anderson <dianders@xxxxxxxxxxxx>
> Sent: October 5, 2021 11:14 AM
> To: Zuo, Jerry <Jerry.Zuo@xxxxxxx>
> Cc: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx>; dri-
> devel@xxxxxxxxxxxxxxxxxxxxx; geert@xxxxxxxxxxxxxx; oliver.sang@xxxxxxxxx;
> Daniel Vetter <daniel@xxxxxxxx>; David Airlie <airlied@xxxxxxxx>; Jani Nikula
> <jani.nikula@xxxxxxxxx>; Linus Walleij <linus.walleij@xxxxxxxxxx>; Maarten
> Lankhorst <maarten.lankhorst@xxxxxxxxxxxxxxx>; Maxime Ripard
> <mripard@xxxxxxxxxx>; Sam Ravnborg <sam@xxxxxxxxxxxx>; Thomas
> Zimmermann <tzimmermann@xxxxxxx>; linux-kernel@xxxxxxxxxxxxxxx;
> Wentland, Harry <Harry.Wentland@xxxxxxx>; Siqueira, Rodrigo
> <Rodrigo.Siqueira@xxxxxxx>; Kuogee Hsieh <khsieh@xxxxxxxxxxxxxx>
> Subject: Re: connector_bad_edid() is broken (was: Re: [PATCH] drm/edid:
> Fix crash with zero/invalid EDID)
>
> Hi,
>
> On Tue, Oct 5, 2021 at 6:33 AM Zuo, Jerry <Jerry.Zuo@xxxxxxx> wrote:
> >
> > > BTW I believe connector_bad_edid() itself is broken since commit
> > > e11f5bd8228f ("drm: Add support for DP 1.4 Compliance edid
> > > corruption test"). Before we've even allocated the memory for the
> > > extension blocks that code now assumes edid[0x7e] is to be 100%
> > > trusted and goes and calculates the checksum on a block based on
> > > that. So that's likely going to be pointing somewhere beyond the
> > > base block into memory we've not even allocated. So anyone who
> > > wanted could craft a bogus EDID and maybe get something interesting to
> happen.
> > >
> > > Would be good if someone could fix that while at it. Or just revert
> > > the offending commit if there is no simple solution immediately in sight.
> > >
> > > The fact that we're parsing entirely untrustworthy crap in the
> > > kernel always worries me. Either we need super careful review of all
> > > relevant code, and/or we need to think about moving the parser out of
> the kernel.
> > > I was considering playing around with the usermode helper stuff.
> > > IIRC there is a way to embed the userspace binary into the kernel
> > > and just fire it up when needed. But so far it's been the usual -ENOTIME
> for me...
> > >
> > [AMD Official Use Only]
> >
> > Hi Ville:
> >
> >      Yhea, it is pretty old change from two years ago, and it is no long valid
> anymore. Please simply drop it.
> >
> > Regards,
> > Jerry
>
> I've cut out other bits from this email and changed the subject line since I
> think this is an issue unrelated to the one my original patch was fixing.
>
> I don't actually know a ton about DP compliance testing, but I attempted to
> try to be helpful and revert commit e11f5bd8228f ("drm:
> Add support for DP 1.4 Compliance edid corruption test"). It wasn't too hard
> to deal with the conflicts in the revert itself, but then things didn't compile
> because there are two places that use `real_edid_checksum` and that goes
> away if I revert the patch.
>
> I've made an attempt to fix the problem by just adding a bounds check.
> Perhaps you can see if that looks good to you:
>
> https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Flore.
> kernel.org%2Fr%2F20211005081022.1.Ib059f9c23c2611cb5a9d760e7d0a700c1
> 295928d%40changeid&amp;data=04%7C01%7CJerry.Zuo%40amd.com%7C90
> b948659454400cedd308d98812c339%7C3dd8961fe4884e608e11a82d994e183d
> %7C0%7C0%7C637690436453163864%7CUnknown%7CTWFpbGZsb3d8eyJWIj
> oiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1
> 000&amp;sdata=OtSngWlYyDc1NbNSgAeALqN3nF%2Bnw08nJ068cpAKZJk%3
> D&amp;reserved=0
>
> -Doug

The patch used for DP1.4 compliance edid corruption test. Let me double check if edid corruption test could be passed without the patch.