linux-kernel - Re: [PATCH] drm: sun8i-ui/vi: Fix layer zpos change/atomic modesetting

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]

Date:   Mon, 30 Sep 2019 17:59:37 +0200
From:   Ondřej Jirman <megous@...ous.com>
To:     Maxime Ripard <mripard@...nel.org>,
        David Airlie <airlied@...ux.ie>,
        Daniel Vetter <daniel@...ll.ch>, Chen-Yu Tsai <wens@...e.org>,
        dri-devel@...ts.freedesktop.org,
        linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] drm: sun8i-ui/vi: Fix layer zpos change/atomic
 modesetting

Hi,

On Tue, Sep 24, 2019 at 02:40:54PM +0200, megous hlavni wrote:
> > >  On first run of the X server, only the black screen and the layer
> > >  containing the cursor is visible. Switching to console and back
> > >  corrects the situation.
> > >
> > >  I have dumped registers, and found out this:
> > >
> > >  (In both cases there are two enabled planes, plane 1 with zpos 0 and
> > >  plane 3 with zpos 1).
> > >
> > >  1) First Xorg run:
> > >
> > >    0x01101000 : 00000201
> > >    0x01101080 : 00000030
> > >
> > >    BLD_FILL_COLOR_CTL: (aka SUN8I_MIXER_BLEND_PIPE_CTL)
> > >      P1_EN
> > >
> > >    BLD_CH_RTCTL: (aka SUN8I_MIXER_BLEND_ROUTE)
> > >      P0_RTCTL channel0
> > >      P1_RTCTL channel3
> > >
> > >  2) After switch to console and back to Xorg:
> > >
> > >  0x01101000 : 00000301
> > >  0x01101080 : 00000031
> > >
> > >    BLD_FILL_COLOR_CTL:
> > >      P1_EN and P0_EN
> > >
> > >    BLD_CH_RTCTL:
> > >      P0_RTCTL channel1
> > >      P1_RTCTL channel3
> > >
> > >  What happens is that sun8i_ui_layer_enable() function may disable
> > >  blending pipes even if it is no longer assigned to its layer, because
> > >  of incorrect state/zpos tracking in the driver.
> > >
> > >  In particular, layer 1 is configured to zpos 0 and thus uses pipe 0.
> > >  When layer 3 is enabled by X server, sun8i_ui_layer_enable() will get
> > >  called with old_zpos=0 zpos=1, which will lead to disabling of pipe 0.
> > >
> > >  In general this issue can happen to any layer during enable or zpos
> > >  changes on multiple layers at once.
> > >
> > >  To correct this we now pass previous enabled/disabled state of the
> > >  layer, and pass real previous zpos of the layer to
> > >  sun8i_ui_layer_enable() and rework the sun8i_ui_layer_enable() function
> > >  to react to the state changes correctly. In order to not complicate
> > >  the atomic_disable callback with all of the above changes, we simply
> > >  remove it and implement all the chanes as part of atomic_update, which
> > >  also reduces the code duplication.
> > 
> > I'm not even sure why we need that old state. Can't we just reset
> > completely the whole thing and do the configuration all over again?
> 
> That would be nice from a dev standpoint if we can get a complete state for all
> planes at once from DRM core, but how? DRM helper gives callbacks
> for updating individual planes with prev and new state. These individual layer
> change notifications don't map nicely to how pipes are represented in the mixer
> registers.

If anyone wants to pursue this further, feel free to. I'm not planning to
pursue this fix further, at the moment.

regards,
	o.