Thank you for your contribution! Sashiko AI review found 8 potential issues to consider: - [High] drm/tegra: broken disable sequence and active state in tegra_dbi_atomic_disable() - [High] drm/tegra: premature bridge add and NULL next_bridge in tegra_dbi_attach() - [Medium] drm/tegra: duplicate next_bridge enable call in tegra_dbi_atomic_enable() - [Medium] drm/tegra: dangling priv->next_bridge pointer in tegra_dbi_host_detach() - [Medium] drm/tegra: device_node refcount leak on dbi_np in tegra_dc_dbi_probe() - [Medium] drm/tegra: unbalanced of_node_put() on remote panel node in tegra_dc_dbi_probe() - [Medium] drm/tegra: dbi: missing idle wait after stopping DC in tegra_dbi_atomic_enable() - [Medium] drm/tegra: unconditional PIN_LDC muxing in tegra_dbi_atomic_enable() --
commit c9bf1d91683f8b29d2b2a60164a0b606d5716c0d Author: Svyatoslav Ryhel <[email protected]> drm/tegra: Add support for 8-bit CPU interface This commit adds a MIPI DBI host and DRM bridge driver for the 8-bit CPU display controller interface on Tegra20 and Tegra30 SoCs, and hooks its probe into tegra_dc_probe(). > diff --git a/drivers/gpu/drm/tegra/dbi.c b/drivers/gpu/drm/tegra/dbi.c > new file mode 100644 > index 0000000000000..290e4c51dd3a7 > --- /dev/null > +++ b/drivers/gpu/drm/tegra/dbi.c [ ... ] > +static void tegra_dbi_host_detach(struct mipi_dbi_host *host, struct > mipi_dbi_device *dbidev) > +{ > + struct tegra_dbi_priv *priv = dbi_host_to_tegra_dbi(host); > + > + drm_bridge_put(priv->next_bridge); > +} [Severity: Medium] Does tegra_dbi_host_detach() leave a dangling pointer in priv->next_bridge after dropping its reference? If the DBI peripheral driver is unbound, later accesses in tegra_dbi_atomic_enable() or tegra_dbi_attach() will dereference the freed bridge unless priv->next_bridge is cleared (for example with drm_bridge_clear_and_put()). [ ... ] > +static void tegra_dbi_atomic_enable(struct drm_bridge *bridge, > + struct drm_atomic_commit *state) > +{ > + struct tegra_dbi_priv *priv = bridge_to_tegra_dbi(bridge); > + struct drm_bridge *next_bridge = priv->next_bridge; > + struct tegra_dc *dc = priv->dc; > + u32 value; > + > + value = tegra_dc_readl(dc, DC_CMD_DISPLAY_COMMAND); > + value &= ~DISP_CTRL_MODE_MASK; > + tegra_dc_writel(dc, value, DC_CMD_DISPLAY_COMMAND); > + > + /* Disable all options except cursor if enabled */ > + value = tegra_dc_readl(dc, DC_DISP_DISP_WIN_OPTIONS); > + value &= CURSOR_ENABLE; > + tegra_dc_writel(dc, value, DC_DISP_DISP_WIN_OPTIONS); > + > + tegra_dc_commit(dc); [Severity: Medium] Should tegra_dbi_atomic_enable() wait for the display controller to become idle after committing DISP_CTRL_MODE_STOP? tegra_crtc_atomic_enable() has already started DISP_CTRL_MODE_C_DISPLAY and tegra_rgb_encoder_enable() has committed rgb_enable, so GENERAL_ACT_REQ is deferred until the next frame boundary rather than latching immediately. Because panel enable callbacks sleep across multiple frames while bit-banging DCS commands, the pending frame-boundary latch can overwrite the active DC_COM_PIN_OUTPUT_* GPIO settings with the shadowed video-mode values mid-transfer. [ ... ] > + /* Set LDC pin to V Pulse 1 */ > + value = tegra_dc_readl(dc, DC_COM_PIN_OUTPUT_SELECT(6)); > + value &= ~PIN_OUTPUT_SELECT_MASK(PIN_OUTPUT_SELECT_LDC); > + value |= PIN_OUTPUT_SELECT_F4(PIN_OUTPUT_SELECT_LDC); > + tegra_dc_writel(dc, value, DC_COM_PIN_OUTPUT_SELECT(6)); [Severity: Medium] Does this unconditionally mux V_PULSE1 onto the primary interface D/C pin (PIN_LDC) even when the secondary 8-bit CPU interface is used? On the secondary interface (nvidia,interface = 1), the D/C signal is on PIN_LVP0 in DC_COM_PIN_OUTPUT_SELECT(4), which tegra_rgb_encoder_enable() leaves set to PIN_OUTPUT_SELECT_OUT (0x00210222). Restoring com_pin_backup therefore leaves PIN_LVP0 in fixed GPIO output mode during scanout instead of driving per-frame MIPI_DCS_WRITE_MEMORY_START commands. [ ... ] > + /* Unshadow to set up command and data pins */ > + tegra_dc_writel(dc, WRITE_MUX | READ_MUX, DC_CMD_STATE_ACCESS); > + > + priv->interface->init(priv); > + > + priv->active = true; > + > + /* > + * DBI requires that the next bridge be enabled here in order to > + * send DCS commands, since the interface pins will subsequently be > + * reconfigured for video data. > + * > + * Using the prepare_prev_first flag in the panel will not work, > + * because the DBI bridge requires the panel to be enabled beforehand > + * for proper configuration. > + */ > + if (next_bridge->funcs && next_bridge->funcs->atomic_enable) > + next_bridge->funcs->atomic_enable(next_bridge, state); [Severity: Medium] As reported by Mikko Perttunen on v5 [1], does calling next_bridge->funcs->atomic_enable() here cause it to run twice on every modeset? Because tegra_dbi_attach() links priv->next_bridge into encoder->bridge_chain after priv->bridge, drm_atomic_bridge_chain_enable() immediately advances to priv->next_bridge and calls its atomic_enable callback a second time while priv->active is false. For panel_bridge, this triggers the "Skipping enable of already enabled panel" warning in drm_panel_enable(), and any downstream bridge that sends DBI commands in the second call fails with -ENODEV. > + > + tegra_dbi_pin_set(dc, &priv->interface->cs, 0); > + > + priv->active = false; [ ... ] > +static void tegra_dbi_atomic_disable(struct drm_bridge *bridge, > + struct drm_atomic_commit *state) > +{ > + struct tegra_dbi_priv *priv = bridge_to_tegra_dbi(bridge); > + struct tegra_dc *dc = priv->dc; > + > + tegra_dc_writel(dc, WRITE_MUX | READ_MUX, DC_CMD_STATE_ACCESS); > + > + priv->interface->init(priv); > + priv->active = true; > +} [Severity: High] Does running this setup in atomic_disable prevent panel power-down commands from working and leave the interface marked active after CRTC disable? During disable_outputs(), drm_atomic_bridge_chain_disable() walks the chain in reverse, so next_bridge->funcs->atomic_disable() runs before tegra_dbi_atomic_disable() while priv->active is still false. Immediately after tegra_dbi_atomic_disable() configures the DBI pins with WRITE_MUX | READ_MUX enabled, drm_atomic_helper_commit_encoder_bridge_disable() calls tegra_rgb_encoder_disable(), which writes rgb_disable directly to the active DC_COM_PIN_OUTPUT_SELECT and DC_COM_PIN_OUTPUT_ENABLE registers and tristates the pins before drm_panel_unprepare() runs in post_disable. In addition, nothing clears priv->active or DC_CMD_STATE_ACCESS afterward, so tegra_dbi_dbi_command() remains open for MMIO writes after the display controller is suspended and unclocked. [ ... ] > +static int tegra_dbi_attach(struct drm_bridge *bridge, > + struct drm_encoder *encoder, > + enum drm_bridge_attach_flags flags) > +{ > + struct tegra_dbi_priv *priv = bridge_to_tegra_dbi(bridge); > + > + return drm_bridge_attach(bridge->encoder, priv->next_bridge, bridge, > + flags); > +} [Severity: High] What happens if tegra_dbi_attach() runs before the MIPI DBI panel driver has bound and called tegra_dbi_host_attach()? tegra_dc_dbi_probe() publishes priv->bridge via devm_drm_bridge_add() before devm_mipi_dbi_device_register(), and device_add() succeeds even when the panel driver has not probed yet or deferred probe (for instance, drivers/gpu/drm/Makefile links tegra/ before panel/). In that case priv->next_bridge is still NULL when host1x_drm_probe() calls tegra_dbi_attach(), so drm_bridge_attach() returns -EINVAL instead of -EPROBE_DEFER and aborts DRM initialization permanently. [ ... ] > + priv->bridge.of_node = of_node_get(dbi_np); > + priv->bridge.timings = &default_tegra_dbi_timings; > + > + ret = devm_drm_bridge_add(dev, &priv->bridge); > + if (ret) > + return ret; [Severity: Medium] Does of_node_get(dbi_np) leak a reference to the dbi device_node? dbi_np is already managed with __free(device_node), and neither the DRM bridge core nor a devm action releases the extra reference stored in priv->bridge.of_node on error, probe deferral, or driver unbind. > + > + np = of_graph_get_remote_node(dbi_np, 1, -1); > + if (!np) > + return -EINVAL; > + > + dbidev = devm_mipi_dbi_device_register(dev, &priv->dbi_host, np); > + of_node_put(np); [Severity: Medium] Does calling of_node_put(np) here cause an of_node refcount underflow when dbidev is released? devm_mipi_dbi_device_register() attaches np to dbidev->dev via device_set_node() without taking its own reference, and mipi_dbi_device_release() calls of_node_put(dev->of_node) when the device is freed: drivers/gpu/drm/drm_mipi_dbi_bus.c:devm_mipi_dbi_device_register() { ... dbidev->dev.release = mipi_dbi_device_release; device_set_node(&dbidev->dev, of_fwnode_handle(np)); ... } Dropping np's only reference right after registration leaves dbidev->dev.of_node unreferenced and results in a double of_node_put() on error unwind or device removal. [1] https://lore.kernel.org/r/[email protected] -- Sashiko AI review ยท https://sashiko.dev/#/patchset/[email protected]?part=12
