Large I/O areas are broken

[Jiri Svoboda]

If you pio_enable() an area larger than page size, writes beyond the first page never go to the device. I observed this when trying to map the GXemul disk device as a single I/O area. Most of the registers are in the first page, while the I/O buffer is at offset 0x4000 (= page size). The writes to the buffer never make it to the device. When the buffer is mapped as a separate I/O area, everything works as expected.

[Jakub Jermář]

After fixing gxe_bd_t like this:

• uint32_t pad5[0x3fc0];

+ uint8_t pad5[0x3fc0];

the whole structure can be pio_enable()'d at once without any problems and the disk behaves as expected.