diff --git a/docs/src/details/integration/chip_vendors/stm32/ltdc.rst b/docs/src/details/integration/chip_vendors/stm32/ltdc.rst
index ffbffcab6c..efbfea4899 100644
--- a/docs/src/details/integration/chip_vendors/stm32/ltdc.rst
+++ b/docs/src/details/integration/chip_vendors/stm32/ltdc.rst
@@ -65,8 +65,9 @@ LTDC layer to create the display for.
 
 For the best visial results, ``optional_other_full_size_buffer`` should be used
 if enough memory is available. Single-buffered mode is what you should use
-if memory is very scarce. If there is almost enough memory for double-buffered
-direct mode, but not quite, then use partial render mode.
+if memory is very scarce. Chips with a CPU data cache have unavoidable visual
+artifacts when using single-buffered direct mode. If there is almost enough
+memory for double-buffered direct mode, but not quite, then use partial render mode.
 
 To clarify what ``my_ltdc_framebuffer_address`` exactly is, it's the value of
 ``pLayerCfg.FBStartAdress`` when the LTDC layer is configured using the STM32 HAL,
@@ -118,6 +119,26 @@ the display width is 800, the display height is 480, and ``1K`` means 1024 bytes
         FLASH  (rx)    : ORIGIN = 0x08000000, LENGTH = 4096K
     }
 
+If the framebuffer is not reserved in the linker script or the framebuffer is not set during
+board initialization, you may absolutely set it at runtime before creating the LVGL display.
+
+.. code-block:: c
+
+    /* like this */
+    HAL_LTDC_SetAddress(&hltdc, (uint32_t)my_ltdc_framebuffer_address, my_ltdc_layer_index);
+
+    /* and then... */
+
+    lv_display_t * disp;
+    disp = lv_st_ltdc_create_direct(my_ltdc_framebuffer_address,
+                                    optional_other_full_size_buffer,
+                                    my_ltdc_layer_index);
+    /* or */
+    disp = lv_st_ltdc_create_partial(partial_buf1,
+                                     optional_partial_buf2,
+                                     partial_buf_size,
+                                     my_ltdc_layer_index);
+
 
 Display Rotation
 ****************
diff --git a/src/drivers/display/st_ltdc/lv_st_ltdc.c b/src/drivers/display/st_ltdc/lv_st_ltdc.c
index 10599d1529..06b56d63bf 100644
--- a/src/drivers/display/st_ltdc/lv_st_ltdc.c
+++ b/src/drivers/display/st_ltdc/lv_st_ltdc.c
@@ -49,6 +49,7 @@ static void flush_cb(lv_display_t * disp, const lv_area_t * area, uint8_t * px_m
 static void flush_wait_cb(lv_display_t * disp);
 static lv_color_format_t get_lv_cf_from_layer_cf(uint32_t cf);
 static void reload_event_callback(LTDC_HandleTypeDef * hltdc);
+static void clean_dcache(void);
 
 #if LV_ST_LTDC_USE_DMA2D_FLUSH
     static void transfer_complete_callback(DMA2D_HandleTypeDef * hdma2d);
@@ -144,7 +145,15 @@ static void flush_cb(lv_display_t * disp, const lv_area_t * area, uint8_t * px_m
     g_data.disp_flushed_in_flush_cb[layer_idx] = false;
 
     if(disp->render_mode == LV_DISPLAY_RENDER_MODE_DIRECT) {
-        if(lv_display_is_double_buffered(disp) && lv_display_flush_is_last(disp)) {
+        bool flush_is_last = lv_display_flush_is_last(disp);
+        if(flush_is_last) {
+            /* there is no ideal time to clean the cache (if present)
+               for **single-buffered** direct mode because the active buffer is drawn to
+               while LTDC is scanning it. Clean it in the last flush, at least,
+               but not every flush because it's expensive for not much visual improvement. */
+            clean_dcache();
+        }
+        if(flush_is_last && lv_display_is_double_buffered(disp)) {
             HAL_LTDC_SetAddress_NoReload(&hltdc, (uint32_t)px_map, layer_idx);
             g_data.layer_interrupt_is_owned[layer_idx] = true;
             HAL_LTDC_Reload(&hltdc, LTDC_RELOAD_VERTICAL_BLANKING);
@@ -172,6 +181,8 @@ static void flush_cb(lv_display_t * disp, const lv_area_t * area, uint8_t * px_m
         lv_display_rotation_t rotation = lv_display_get_rotation(disp);
         if(rotation == LV_DISPLAY_ROTATION_0) {
 #if LV_ST_LTDC_USE_DMA2D_FLUSH
+            clean_dcache();
+
             uint32_t dma2d_input_cf = get_dma2d_input_cf_from_lv_cf(cf);
             uint32_t dma2d_output_cf = get_dma2d_output_cf_from_layer_cf(layer_cfg->PixelFormat);
 
@@ -194,12 +205,14 @@ static void flush_cb(lv_display_t * disp, const lv_area_t * area, uint8_t * px_m
                 fb_p += fb_stride;
                 px_map_p += area_stride;
             }
+            clean_dcache();
             g_data.disp_flushed_in_flush_cb[layer_idx] = true;
 #endif
         }
         else {
             uint32_t area_stride = px_size * area_width;
             lv_draw_sw_rotate(px_map, first_pixel, area_width, area_height, area_stride, fb_stride, rotation, cf);
+            clean_dcache();
             g_data.disp_flushed_in_flush_cb[layer_idx] = true;
         }
     }
@@ -242,6 +255,15 @@ static void reload_event_callback(LTDC_HandleTypeDef * hltdc)
     }
 }
 
+static void clean_dcache(void)
+{
+#if defined(__CORTEX_M) && __CORTEX_M == 7
+    SCB_CleanDCache();
+#elif defined(__CORTEX_A) && __CORTEX_A == 7
+    L1C_CleanDCacheAll();
+#endif
+}
+
 #if LV_ST_LTDC_USE_DMA2D_FLUSH
 static void transfer_complete_callback(DMA2D_HandleTypeDef * hdma2d)
 {