In trying to get 36-bit physical working on 85xx I ran into an issue with the NAND driver. Its given a virtual address and trying to match it to which chipselect it maps to. The assumption in the code is that we have a 1:1 mapping and clearly that may not be true w/36-bit phys.

So it seems we need a virt_to_phys() function. However there are a few questions I have about this:

1. what should the prototype of this function look like:

phys_addr_t virt_to_phys (void * addr);

or

int virt_to_phys(void *p, phys_addr_t *addr);

I think the vast majority of implementations will be just 1:1 mapping, however there is a question in a dynamic mapping case of how to report back not finding a mapping.

2. what impact is there on existing use on:

cpu/mips/au1x00_eth.c: fifo_tx[next_tx].addr = (virt_to_phys(packet))| TX_DMA_ENABLE; cpu/mips/au1x00_eth.c: fifo_rx[next_rx].addr = (virt_to_phys(NetRxPackets[next_rx]))|RX_DMA_ENABLE; cpu/mips/au1x00_eth.c: fifo_tx[i].addr = virt_to_phys(&txbuf[0]); cpu/mips/au1x00_eth.c: fifo_rx[i].addr = (virt_to_phys(NetRxPackets[i]))|RX_DMA_ENABLE;

- k