ar_osal's MMZ backing is write-combine (correct but slow). A cached fast-path needs a kernel patch exporting arch_sync_dma_for_{device,cpu}. Before optimizing, validate frame integrity on hardware with the current WC backing. The memremap(MEMREMAP_WC) on the no-map mmz@29400000 region is confirmed (Tier-0), so this is about the cached path, not the mapping.
ar_osal's MMZ backing is write-combine (correct but slow). A cached fast-path needs a kernel patch exportingarch_sync_dma_for_{device,cpu}. Before optimizing, validate frame integrity on hardware with the current WC backing. Thememremap(MEMREMAP_WC)on the no-mapmmz@29400000region is confirmed (Tier-0), so this is about the cached path, not the mapping.