linux-kernel - Re: [PATCH] drm/msm/mdp5: enable autocommit

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: <CAOCk7NosRhRp3vZxg2Nx8106PQ0ryo5b68cUv605XUzCm6gYPA@mail.gmail.com>
Date:   Tue, 12 Nov 2019 08:38:27 -0700
From:   Jeffrey Hugo <jeffrey.l.hugo@...il.com>
To:     Brian Masney <masneyb@...tation.org>
Cc:     Rob Clark <robdclark@...omium.org>,
        Rob Clark <robdclark@...il.com>,
        freedreno <freedreno@...ts.freedesktop.org>,
        lkml <linux-kernel@...r.kernel.org>,
        "open list:DRM PANEL DRIVERS" <dri-devel@...ts.freedesktop.org>,
        MSM <linux-arm-msm@...r.kernel.org>, Sean Paul <sean@...rly.run>
Subject: Re: [PATCH] drm/msm/mdp5: enable autocommit

On Tue, Nov 12, 2019 at 3:49 AM Brian Masney <masneyb@...tation.org> wrote:
>
> Since the introduction of commit 2d99ced787e3 ("drm/msm: async commit
> support"), command-mode panels began throwing the following errors:
>
>     msm fd900000.mdss: pp done time out, lm=0
>
> Let's fix this by enabling the autorefresh feature that's available in
> the MDP starting at version 1.0. This will cause the MDP to
> automatically send a frame to the panel every time the panel invokes
> the TE signal, which will trigger the PP_DONE IRQ. This requires not
> sending a START signal for command-mode panels.
>
> This fixes the error and gives us a counter for command-mode panels that
> we can use to implement async commit support for the MDP5 in a follow up
> patch.
>
> Signed-off-by: Brian Masney <masneyb@...tation.org>
> Suggested-by: Jeffrey Hugo <jeffrey.l.hugo@...il.com>
> ---
>  drivers/gpu/drm/msm/disp/mdp5/mdp5_crtc.c | 15 ++++++++++++++-
>  drivers/gpu/drm/msm/disp/mdp5/mdp5_ctl.c  |  9 +--------
>  2 files changed, 15 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/gpu/drm/msm/disp/mdp5/mdp5_crtc.c b/drivers/gpu/drm/msm/disp/mdp5/mdp5_crtc.c
> index 05cc04f729d6..539348cb6331 100644
> --- a/drivers/gpu/drm/msm/disp/mdp5/mdp5_crtc.c
> +++ b/drivers/gpu/drm/msm/disp/mdp5/mdp5_crtc.c
> @@ -456,6 +456,7 @@ static void mdp5_crtc_atomic_enable(struct drm_crtc *crtc,
>  {
>         struct mdp5_crtc *mdp5_crtc = to_mdp5_crtc(crtc);
>         struct mdp5_crtc_state *mdp5_cstate = to_mdp5_crtc_state(crtc->state);
> +       struct mdp5_pipeline *pipeline = &mdp5_cstate->pipeline;
>         struct mdp5_kms *mdp5_kms = get_kms(crtc);
>         struct device *dev = &mdp5_kms->pdev->dev;
>
> @@ -493,9 +494,21 @@ static void mdp5_crtc_atomic_enable(struct drm_crtc *crtc,
>
>         mdp_irq_register(&mdp5_kms->base, &mdp5_crtc->err);
>
> -       if (mdp5_cstate->cmd_mode)
> +       if (mdp5_cstate->cmd_mode) {
>                 mdp_irq_register(&mdp5_kms->base, &mdp5_crtc->pp_done);
>
> +               /*
> +                * Enable autorefresh so we get regular ping/pong IRQs.
> +                * - Bit 31 is the enable bit
> +                * - Bits 0-15 represent the frame count, specifically how many
> +                *   TE events before the MDP sends a frame.
> +                */
> +               mdp5_write(mdp5_kms,
> +                          REG_MDP5_PP_AUTOREFRESH_CONFIG(pipeline->mixer->pp),
> +                          BIT(31) | BIT(0));
> +               crtc_flush_all(crtc);
> +       }
> +
>         mdp5_crtc->enabled = true;
>  }
>
> diff --git a/drivers/gpu/drm/msm/disp/mdp5/mdp5_ctl.c b/drivers/gpu/drm/msm/disp/mdp5/mdp5_ctl.c
> index 030279d7b64b..aee295abada3 100644
> --- a/drivers/gpu/drm/msm/disp/mdp5/mdp5_ctl.c
> +++ b/drivers/gpu/drm/msm/disp/mdp5/mdp5_ctl.c
> @@ -187,14 +187,7 @@ static bool start_signal_needed(struct mdp5_ctl *ctl,
>         if (!ctl->encoder_enabled)
>                 return false;
>
> -       switch (intf->type) {
> -       case INTF_WB:
> -               return true;
> -       case INTF_DSI:
> -               return intf->mode == MDP5_INTF_DSI_MODE_COMMAND;
> -       default:
> -               return false;
> -       }
> +       return intf->type == INTF_WB;
>  }

I don't think this fully works.

The whole "flush" thing exists because the configuration is double
buffered.  You write to the flush register to tell the hardware to
pickup the new configuration, but it doesn't do that automatically.
It only picks up the new config on the next "vsync".  When you have a
video mode panel, you have the timing engine running, which drives
that.  With a command mode panel, you have either the start signal, or
the auto refresh to do the same, but you have a bit of a chicken and
egg situation where if you are programming the hardware from scratch,
autorefresh isn't already enabled to then pickup the config to enable
autorefresh. In this case, you'll need a single start to kick
everything off.  However, if say the bootloader already configured
things and has autorefresh running, then you need to not do that start
because you'll overload the DSI like you saw.

Nothing is simple, is it?  :)