Hi,
it looks like much of the perceived Linux slowness on CI20 NAND is due to BCH driver using PIO? Or is it UBI/FS?
Is there a fundamental limitation of NEMC / BCH or MTD that requires CPU to read / write each data word or this could be changed to use DMA engine / driver?
I wrote some (proprietary) drivers in Linux v2.0 days, but complete lost touch with the kernel developments after v2.4.9. It's much more complicated set of frameworks now, but I may want to play with it on 3.18 or 6.1 if speed improvement is realistic. The only doc I have is JZ4780 Programming Manual though (JZ4780_pm.pdf).