On Mon, Jan 20, 2025 at 4:10 PM Miquel Raynal <miquel.raynal@xxxxxxxxxxx> wrote: > > > >> >> Don't you need a dma_addr_t here instead? You shall remap the resource. > >> >> > >> > Sorry, I don't quite understand. > >> > 'dma_base' is already of type dma_addr_t. > >> > >> I didn't identify where the dma_base was remapped, but if that's already > >> done then we're good. > > > > Perhaps I misunderstand the usage of dma_map_resource(). dma_base is > > the physical address and will be written to the DMA controller > > register at last. > > IIUC, yes and no. It is the remapped address of the physical address you > want to reach, not the physical address itself. The type is dma_addr_t > and not phys_addr_t and you can get a dma address out of a physical > address with a call to dma_map_resource(). The numerical value might be > the same if there is no IO-MMU involved, but the good practice is to > remap anyway. Got it. I will keep dma_map_resource() as it is, which is called in ls1x_nand_probe(). + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "nand-dma"); + host->dma_base = dma_map_resource(dev, res->start, resource_size(res), + DMA_BIDIRECTIONAL, 0); + if (dma_mapping_error(dev, host->dma_base)) + return -ENXIO; > > Thanks, > Miquèl > -- Best regards, Keguang Zhang
