Re: Adding device support in aarch64 virt board

[Peter Maydell]

On Thu, 22 Oct 2020 at 01:19, root artifice <artifice37@outlook.com> wrote:
> I’d like to add I2C support for aarch64 virt board. So I modified 
> hw/arm/virt.c as follows.

> Add [VIRT_I2C] = { 0x09100000, 0x00001000 } entry in base_memmap to define a 
> memory region
> Use the following to emulate a I2C device
>
> dev = sysbus_create_simple(TYPE_VERSATILE_I2C, vms->memmap[VIRT_I2C].base, 
> NULL);
>
> i2c = (I2CBus *)qdev_get_child_bus(dev, "i2c");
>
> i2c_slave_create_simple(i2c, "sii9022", 0x39);
>
> Use the following to add device tree file entry
>
> nodename = g_strdup_printf("/i2c@%" PRIx64, base);
>
> qemu_fdt_add_subnode(vms->fdt, nodename);
>
> qemu_fdt_setprop_string(vms->fdt, nodename, "compatible", 
> "arm,versatile-i2c");
>
> qemu_fdt_setprop_sized_cells(vms->fdt, nodename, "reg", 1, base, 1, size);
>
>
>
> After modification, QEMU can compile successfully. However, when I run it 
> (qemu-system-aarch64 -M virt -m 2048 -cpu cortex-a53 -kernel ./Image 
> -nographic -hda rootfs.ext2 --append "root=/dev/vda"), I got the following 
> error:
>
> qemu-system-aarch64: qemu-5.1.0/softmmu/memory.c:2388: 
> memory_region_add_subregion_common: Assertion `!subregion->container' failed
>
>
>
> The assertion exists in qemu-5.1.0\softmmu\memory.c: 
> memory_region_add_subregion_common() and is triggered by dev = 
> sysbus_create_simple(TYPE_VERSATILE_I2C, vms->memmap[VIRT_I2C].base, NULL);. 
> I think it means I’m using a memory region that has been used. But I think 
> the memory region in base_memap I’ve chosen is available. Besides, the 
> assertion “subregion->container” is derived from dev = qdev_new(name) 
> (located in hw/core/sysbus.c), which I have no control over when calling 
> sysbus_create_simple(). Could someone tell me which part is wrong?

This assertion means "you tried to put a MemoryRegion into a
container twice". It's not related to the specific guest addresses
involved but to what is being done with the MemoryRegion objects.
It shouldn't happen with just the code you quote above, so
my conclusion is that you haven't showed us all your code and
the bug is in something you haven't told us about.

PS: the best way to add I2C support to the virt board would
be to have a PCI device which was an I2C controller. Then
you wouldn't need to modify the virt.c code at all.

thanks
-- PMM