On 30 August 2017 at 03:45, Philippe Mathieu-Daudé <f4...@amsat.org> wrote: > I think they might be issues if you start QEMU without -serial and then use > a firmware polling for an uart, the device won't be mapped and the memory > accesses are mostly ignored. > > I'd rather use: > > for (i = 0; i < MSF2_NUM_UARTS && i < MAX_SERIAL_PORTS; i++) { > static const char *serial[] = {"serial0", "serial1"}; > > if (!serial_hds[i]) { > serial_hds[i] = qemu_chr_new(serial[i], "null"); > > } > >> + serial_mm_init(get_system_memory(), uart_addr[i], 2, >> + qdev_get_gpio_in(armv7m, uart_irq[i]), >> + 115200, serial_hds[i], DEVICE_NATIVE_ENDIAN); >> + } >> + }
It would be better to fix serial_mm_init() to handle having a NULL chardev pointer, because we already have a lot of SoC code that just passes it serial_hds[] regardless. I'd leave this code as it is and we can fix serial_mm_init separately (somebody pointed out this issue for a xilinx board recently). thanks -- PMM