Re: [PATCH] drm/bridge: Fix Exynos DSI after making bridge chain a double-linked list

From: Boris Brezillon
Date: Fri Dec 27 2019 - 07:00:13 EST


On Fri, 27 Dec 2019 12:01:35 +0100
Marek Szyprowski <m.szyprowski@xxxxxxxxxxx> wrote:

> Exynos DSI DRM driver uses private calls to out bridge to force certain
> order of operations during init/exit sequences. This no longer works after
> conversion of bridge chain to a double-linked list. To fix the regression
> call bridge related operations manually instead of the generic
> drm_bridge_chain_*() operations.

I think it'd be worth explaining what the problem is (infinite loop
caused by list_for_each_entry() use when the bridge is no longer part
of the chain attached to the encoder).

>
> Signed-off-by: Marek Szyprowski <m.szyprowski@xxxxxxxxxxx>
> Fixes: 05193dc38197 ("drm/bridge: Make the bridge chain a double-linked list")

We also need to fix that in VC4.

> ---
> This patch is a result of the following discussion:
> https://www.spinics.net/lists/dri-devel/msg239256.html
> ---
> drivers/gpu/drm/exynos/exynos_drm_dsi.c | 21 ++++++++++++---------
> 1 file changed, 12 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/gpu/drm/exynos/exynos_drm_dsi.c b/drivers/gpu/drm/exynos/exynos_drm_dsi.c
> index 3955f84dc893..f5905c239a86 100644
> --- a/drivers/gpu/drm/exynos/exynos_drm_dsi.c
> +++ b/drivers/gpu/drm/exynos/exynos_drm_dsi.c
> @@ -255,7 +255,6 @@ struct exynos_dsi {
> struct mipi_dsi_host dsi_host;
> struct drm_connector connector;
> struct drm_panel *panel;
> - struct list_head bridge_chain;
> struct drm_bridge *out_bridge;
> struct device *dev;
>
> @@ -1391,7 +1390,8 @@ static void exynos_dsi_enable(struct drm_encoder *encoder)
> if (ret < 0)
> goto err_put_sync;
> } else {
> - drm_bridge_chain_pre_enable(dsi->out_bridge);
> + if (dsi->out_bridge->funcs->pre_enable)
> + dsi->out_bridge->funcs->pre_enable(dsi->out_bridge);

Okay, so you're calling ->{pre_enable,enable,disable,post_disable}() on
the first bridge element which only works if the chain contains one
bridge (see below). Maybe you should keep exynos_dsi.bridge_chain and
create custom helpers to iterate over chain elements instead of calling
those hooks only on out_bridge.

> }
>
> exynos_dsi_set_display_mode(dsi);
> @@ -1402,7 +1402,8 @@ static void exynos_dsi_enable(struct drm_encoder *encoder)
> if (ret < 0)
> goto err_display_disable;
> } else {
> - drm_bridge_chain_enable(dsi->out_bridge);
> + if (dsi->out_bridge->funcs->enable)
> + dsi->out_bridge->funcs->enable(dsi->out_bridge);
> }
>
> dsi->state |= DSIM_STATE_VIDOUT_AVAILABLE;
> @@ -1427,10 +1428,12 @@ static void exynos_dsi_disable(struct drm_encoder *encoder)
> dsi->state &= ~DSIM_STATE_VIDOUT_AVAILABLE;
>
> drm_panel_disable(dsi->panel);
> - drm_bridge_chain_disable(dsi->out_bridge);
> + if (dsi->out_bridge->funcs->disable)
> + dsi->out_bridge->funcs->disable(dsi->out_bridge);
> exynos_dsi_set_display_enable(dsi, false);
> drm_panel_unprepare(dsi->panel);
> - drm_bridge_chain_post_disable(dsi->out_bridge);
> + if (dsi->out_bridge->funcs->post_disable)
> + dsi->out_bridge->funcs->post_disable(dsi->out_bridge);
> dsi->state &= ~DSIM_STATE_ENABLED;
> pm_runtime_put_sync(dsi->dev);
> }
> @@ -1521,9 +1524,11 @@ static int exynos_dsi_host_attach(struct mipi_dsi_host *host,
>
> out_bridge = of_drm_find_bridge(device->dev.of_node);
> if (out_bridge) {
> - drm_bridge_attach(encoder, out_bridge, NULL);
> + out_bridge->dev = drm;
> + out_bridge->encoder = encoder;
> + if (out_bridge->funcs->attach)
> + out_bridge->funcs->attach(out_bridge);

If you don't want to keep the dsi_exynos.bridge_chain field, you should
probably still use drm_bridge_attach() here and manually reset
encoder->bridge_chain after checking it only contains a single element.
Something like:

drm_bridge_attach(encoder, out_bridge, NULL);

/*
* FIXME: we need to control the enable/disable
* sequence to make sure the DSI host controller is
* ready to send DSI commands before the
* bridge->pre_enable()/panel->prepare() hooks are
* called (those hooks are allowed to send DSI
* commands). The only way to do that right now is to
* reset the bridge chain so that the core thinks it's
* empty and turns drm_bridge_chain_xxx() calls into
* NOPs. This only works if the chain contains only one
* bridge, otherwise we'd be skipping bridge->xxx()
* calls on other bridges which is wrong.
*/
WARN_ON(!list_is_singular(&encoder->bridge_chain));
INIT_LIST_HEAD(&encoder->bridge_chain);

Calling ->attach() directly might lead to an invalid pointer
dereference if out_bridge tries to attach to another bridge
(out_bridge->bridge_node is uninitialized if you don't call
drm_bridge_attach()).

> dsi->out_bridge = out_bridge;
> - list_splice(&encoder->bridge_chain, &dsi->bridge_chain);
> } else {
> int ret = exynos_dsi_create_connector(encoder);
>
> @@ -1589,7 +1594,6 @@ static int exynos_dsi_host_detach(struct mipi_dsi_host *host,
> if (dsi->out_bridge->funcs->detach)
> dsi->out_bridge->funcs->detach(dsi->out_bridge);
> dsi->out_bridge = NULL;
> - INIT_LIST_HEAD(&dsi->bridge_chain);
> }
>
> if (drm->mode_config.poll_enabled)
> @@ -1737,7 +1741,6 @@ static int exynos_dsi_probe(struct platform_device *pdev)
> init_completion(&dsi->completed);
> spin_lock_init(&dsi->transfer_lock);
> INIT_LIST_HEAD(&dsi->transfer_list);
> - INIT_LIST_HEAD(&dsi->bridge_chain);
>
> dsi->dsi_host.ops = &exynos_dsi_ops;
> dsi->dsi_host.dev = dev;