Thank you for your contribution! Sashiko AI review found 4 potential issues to consider: - [High] drm/panel: missing prepare_prev_first = true in nt36525_probe() - [Medium] drm/panel: vpos, vneg, and backlight in nt36525_probe() mismatch DT binding - [Medium] drm/panel: inverted gpiod logical polarity for reset_gpio in nt36525 - [Low] drm/panel: negative return and open-coded helper in nt36525_get_modes() --
commit bba87e5acae9fff577a4366212e7273dead1f8e4 Author: Adil Hüsrev Tepe <[email protected]> drm/panel: add Novatek NT36525 panel driver This commit adds a DRM panel driver for Novatek NT36525-based MIPI-DSI video-mode panels used in the Samsung Galaxy XCover5. > diff --git a/drivers/gpu/drm/panel/panel-novatek-nt36525.c > b/drivers/gpu/drm/panel/panel-novatek-nt36525.c > new file mode 100644 > index 0000000000000..246476af49b15 > --- /dev/null > +++ b/drivers/gpu/drm/panel/panel-novatek-nt36525.c [ ... ] > +static int nt36525_prepare(struct drm_panel *panel) > +{ > + struct nt36525 *p = panel_to_nt36525(panel); > + int ret; > + > + /* Power-up sequence from Samsung decon_board_nt36525 > panel_power_enable */ > + ret = regulator_enable(p->vdd); > + if (ret) > + return ret; > + usleep_range(1000, 1500); > + ret = regulator_enable(p->vpos); > + if (ret) > + goto disable_vdd; > + usleep_range(1000, 1500); > + ret = regulator_enable(p->vneg); > + if (ret) > + goto disable_vpos; > + usleep_range(1000, 1500); > + > + /* Reset pulse, Samsung sequence from panel_reset */ > + usleep_range(10000, 11000); > + gpiod_set_value_cansleep(p->reset_gpio, 1); > + usleep_range(1000, 1500); > + gpiod_set_value_cansleep(p->reset_gpio, 0); > + usleep_range(3500, 4000); > + gpiod_set_value_cansleep(p->reset_gpio, 1); > + usleep_range(10000, 11000); > + > + ret = nt36525_on(p); > + if (ret) > + goto fail; > + > + return 0; > + > +fail: > + gpiod_set_value_cansleep(p->reset_gpio, 0); > + regulator_disable(p->vneg); > + usleep_range(1000, 1500); > +disable_vpos: > + regulator_disable(p->vpos); > + usleep_range(5000, 5500); > +disable_vdd: > + regulator_disable(p->vdd); > + return ret; > +} > + > +static int nt36525_unprepare(struct drm_panel *panel) > +{ > + struct nt36525 *p = panel_to_nt36525(panel); > + struct mipi_dsi_multi_context ctx = { .dsi = p->dsi }; > + > + mipi_dsi_dcs_set_display_off_multi(&ctx); > + mipi_dsi_msleep(&ctx, 20); > + mipi_dsi_dcs_enter_sleep_mode_multi(&ctx); > + mipi_dsi_msleep(&ctx, 60); > + > + usleep_range(3000, 3500); > + regulator_disable(p->vneg); > + usleep_range(1000, 1500); > + regulator_disable(p->vpos); > + usleep_range(5000, 5500); > + regulator_disable(p->vdd); > + msleep(30); > + gpiod_set_value_cansleep(p->reset_gpio, 0); > + return 0; > +} [Severity: Medium] Are the gpiod logical states for p->reset_gpio inverted across nt36525_prepare(), nt36525_unprepare(), and nt36525_probe()? As noted in the comment in nt36525_probe() ("Request RESX released (high) rather than asserting it here"), the hardware RESX pin is active-low (physical 0 asserts reset, physical 1 releases it). Under the gpiod API and panel-common.yaml, logical 1 (GPIOD_OUT_HIGH) means reset asserted and logical 0 (GPIOD_OUT_LOW) means reset deasserted, with the physical inversion described via GPIO_ACTIVE_LOW in Device Tree. Passing physical levels directly means a Device Tree using standard GPIO_ACTIVE_LOW polarity holds the panel in reset during nt36525_on() and releases reset when powering off, or else has to misdeclare the active-low pin as GPIO_ACTIVE_HIGH. > +static int nt36525_get_modes(struct drm_panel *panel, struct drm_connector > *conn) > +{ > + struct drm_display_mode *m; > + > + m = drm_mode_duplicate(conn->dev, &nt36525_xcover5_mode); > + if (!m) > + return -ENOMEM; > + drm_mode_set_name(m); > + drm_mode_probed_add(conn, m); > + conn->display_info.width_mm = m->width_mm; > + conn->display_info.height_mm = m->height_mm; > + return 1; > +} [Severity: Low] Can nt36525_get_modes() use drm_connector_helper_get_modes_fixed(conn, &nt36525_xcover5_mode) instead of open-coding mode duplication? drm_helper_probe_get_modes() expects .get_modes() callbacks to return the number of modes (0 on failure) and logs a drm_err() if a negative error code such as -ENOMEM is returned. drm_connector_helper_get_modes_fixed() handles duplicating the mode, setting its name, updating display_info dimensions, and returning 0 on failure. [ ... ] > +static int nt36525_probe(struct mipi_dsi_device *dsi) > +{ > + struct device *dev = &dsi->dev; > + struct nt36525 *p; > + int ret; > + > + p = devm_drm_panel_alloc(dev, struct nt36525, panel, &nt36525_funcs, > + DRM_MODE_CONNECTOR_DSI); > + if (IS_ERR(p)) > + return PTR_ERR(p); > + > + p->dsi = dsi; > + ret = drm_panel_of_backlight(&p->panel); > + if (ret) > + return ret; > + > + p->vdd = devm_regulator_get(dev, "vdd"); > + if (IS_ERR(p->vdd)) > + return dev_err_probe(dev, PTR_ERR(p->vdd), "vdd supply\n"); > + > + p->vpos = devm_regulator_get(dev, "vpos"); > + if (IS_ERR(p->vpos)) > + return dev_err_probe(dev, PTR_ERR(p->vpos), "vpos supply\n"); > + > + p->vneg = devm_regulator_get(dev, "vneg"); > + if (IS_ERR(p->vneg)) > + return dev_err_probe(dev, PTR_ERR(p->vneg), "vneg supply\n"); [Severity: Medium] Do the vpos, vneg, and backlight properties requested here match the DT binding in Documentation/devicetree/bindings/display/panel/novatek,nt36525.yaml? The binding schema only defines compatible, reg, reset-gpios, vdd-supply, and port with additionalProperties: false. A Device Tree conforming to that schema will lack vpos-supply, vneg-supply, and backlight, causing devm_regulator_get() to fall back to dummy regulators with a warning (leaving the positive and negative LCD bias supplies unswitched) and drm_panel_of_backlight() to find no backlight, while a Device Tree that includes them will fail dtbs_check. [ ... ] > + dsi->mode_flags = MIPI_DSI_MODE_VIDEO /* video (non-burst) */ > + | MIPI_DSI_MODE_VIDEO_SYNC_PULSE > + | MIPI_DSI_MODE_LPM /* force LP for all > transfers */ > + | MIPI_DSI_CLOCK_NON_CONTINUOUS; > + > + drm_panel_add(&p->panel); [Severity: High] Should p->panel.prepare_prev_first = true be set before calling drm_panel_add(&p->panel)? drm_panel_add() copies panel->prepare_prev_first into panel->bridge.pre_enable_prev_first. Because it is left false, drm_atomic_bridge_chain_pre_enable() calls nt36525_prepare() -> nt36525_on() before the upstream DSI host bridge's atomic_pre_enable() callback, and drm_atomic_bridge_chain_post_disable() disables the upstream DSI host bridge before nt36525_unprepare() sends display-off and sleep-in commands. On bridge-based DSI hosts such as samsung-dsim, transfers sent while DSIM_STATE_ENABLED is cleared are rejected by samsung_dsim_host_transfer() with -EINVAL, causing panel bring-up and power-down to fail. -- Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=2
