next iteration on display, still some artifacting issues, and ux with buttons states and updating

main/drivers/display/display_rgb.c

@@ -108,14 +108,24 @@ static void rgb_flush_cb(lv_display_t *display, const lv_area_t *area, uint8_t *
     lv_display_flush_ready(display);
 }
 
-esp_err_t display_init(void)
+/*
+ * Phase 1: allocate everything, scan nothing.
+ *
+ * Called BEFORE wifi_init_sta(), on a heap nothing else has touched. The
+ * bounce buffers need two contiguous 25.6KB DMA-capable internal blocks, and
+ * WiFi's dynamic RX buffers churn that same pool hard enough to fragment it:
+ * observed on hardware (2026-08-04) as "no mem for bounce buffer" with 84KB
+ * of total free heap -- plenty of bytes, no block big enough, and retries
+ * useless because the fragmentation is not transient.
+ *
+ * Allocating first is deterministic where retrying was a dice roll. The
+ * panel is NOT scanning after this returns -- scan-out (and with it the
+ * PSRAM-reading refill ISR that flash writes would fault) starts only in
+ * display_start().
+ */
+esp_err_t display_alloc(void)
 {
-    ESP_LOGI(TAG, "Initializing 800x480 RGB panel (CrowPanel Advance)");
-
-    /* lv_display_create() below allocates from LVGL's heap. Without lv_init()
-     * that heap is a null pool and the call faults with LoadProhibited. */
-    ESP_RETURN_ON_FALSE(lv_is_initialized(), ESP_ERR_INVALID_STATE, TAG,
-                        "lv_init() must run before display_init()");
+    ESP_LOGI(TAG, "Allocating 800x480 RGB panel (CrowPanel Advance)");
 
     const esp_lcd_rgb_panel_config_t panel_config = {
         .clk_src = LCD_CLK_SRC_DEFAULT,
@@ -175,6 +185,29 @@ esp_err_t display_init(void)
 
     ESP_RETURN_ON_ERROR(esp_lcd_new_rgb_panel(&panel_config, &panel), TAG,
                         "esp_lcd_new_rgb_panel failed");
+
+    ESP_LOGI(TAG, "Panel buffers allocated; free internal heap: %u KB",
+             (unsigned)(heap_caps_get_free_size(MALLOC_CAP_INTERNAL) / 1024));
+    return ESP_OK;
+}
+
+/*
+ * Phase 2: start scan-out and hand the framebuffers to LVGL.
+ *
+ * From esp_lcd_panel_init() onward the bounce-refill ISR is reading PSRAM
+ * continuously, so every flash write (NVS, WiFi PHY calibration) must already
+ * be done -- see the init-order comment in main.c.
+ */
+esp_err_t display_start(void)
+{
+    ESP_RETURN_ON_FALSE(panel != NULL, ESP_ERR_INVALID_STATE, TAG,
+                        "display_alloc() must run before display_start()");
+
+    /* lv_display_create() below allocates from LVGL's heap. Without lv_init()
+     * that heap is a null pool and the call faults with LoadProhibited. */
+    ESP_RETURN_ON_FALSE(lv_is_initialized(), ESP_ERR_INVALID_STATE, TAG,
+                        "lv_init() must run before display_start()");
+
     ESP_RETURN_ON_ERROR(esp_lcd_panel_reset(panel), TAG, "panel reset failed");
     ESP_RETURN_ON_ERROR(esp_lcd_panel_init(panel), TAG, "panel init failed");
 

main/drivers/display/display_rgb.h

@@ -9,9 +9,22 @@
 #define DISPLAY_H_RES 800
 #define DISPLAY_V_RES 480
 
-/* Requires lv_init() to have already run -- call lvgl_port_init() first.
- * Backlight is NOT owned here on the Advance; see drivers/backlight. */
-esp_err_t display_init(void);
+/*
+ * Two-phase bring-up, and the order is load-bearing:
+ *
+ *   display_alloc()  -- BEFORE wifi_init_sta(). Grabs the framebuffers and
+ *                       the two contiguous bounce buffers while the internal
+ *                       heap is still unfragmented; WiFi's RX buffer churn
+ *                       destroys the needed contiguity within seconds.
+ *                       The panel is not scanning after this.
+ *   display_start()  -- AFTER WiFi (all flash writes done) and after
+ *                       lvgl_port_init(). Starts scan-out and registers the
+ *                       display with LVGL.
+ *
+ * Backlight is NOT owned here on the Advance; see drivers/backlight.
+ */
+esp_err_t display_alloc(void);
+esp_err_t display_start(void);
 
 lv_display_t *display_get_default(void);