Michael S. Tsirkin <mst@xxxxxxxxxx> writes:
> On Tue, Sep 24, 2019 at 02:44:33PM +0200, Sergio Lopez wrote:
>> +static void microvm_fix_kernel_cmdline(MachineState *machine)
>> +{
>> + X86MachineState *x86ms = X86_MACHINE(machine);
>> + BusState *bus;
>> + BusChild *kid;
>> + char *cmdline;
>> +
>> + /*
>> + * Find MMIO transports with attached devices, and add them to the kernel
>> + * command line.
>> + *
>> + * Yes, this is a hack, but one that heavily improves the UX without
>> + * introducing any significant issues.
>> + */
>> + cmdline = g_strdup(machine->kernel_cmdline);
>> + bus = sysbus_get_default();
>> + QTAILQ_FOREACH(kid, &bus->children, sibling) {
>> + DeviceState *dev = kid->child;
>> + ObjectClass *class = object_get_class(OBJECT(dev));
>> +
>> + if (class == object_class_by_name(TYPE_VIRTIO_MMIO)) {
>> + VirtIOMMIOProxy *mmio = VIRTIO_MMIO(OBJECT(dev));
>> + VirtioBusState *mmio_virtio_bus = &mmio->bus;
>> + BusState *mmio_bus = &mmio_virtio_bus->parent_obj;
>> +
>> + if (!QTAILQ_EMPTY(&mmio_bus->children)) {
>> + gchar *mmio_cmdline = microvm_get_mmio_cmdline(mmio_bus->name);
>> + if (mmio_cmdline) {
>> + char *newcmd = g_strjoin(NULL, cmdline, mmio_cmdline, NULL);
>> + g_free(mmio_cmdline);
>> + g_free(cmdline);
>> + cmdline = newcmd;
>> + }
>> + }
>> + }
>> + }
>> +
>> + fw_cfg_modify_i32(x86ms->fw_cfg, FW_CFG_CMDLINE_SIZE, strlen(cmdline) + 1);
>> + fw_cfg_modify_string(x86ms->fw_cfg, FW_CFG_CMDLINE_DATA, cmdline);
>> +}
>
> Can we rearrange this somewhat? Maybe the mmio constructor
> would format the device description and add to some list,
> and then microvm would just get stuff from that list
> and add it to kernel command line?
> This way it can also be controlled by a virtio-mmio property, so
> e.g. you can disable it per device if you like.
> In particular, this seems like a handy trick for any machine type
> using mmio.
Disabling it per-device won't be easy, as transport options can't be
specified using the underlying device properties.
But, otherwise, sounds like a good idea to avoid having to traverse the
qtree. I'll give it a try.
Thanks,
Sergio.
Attachment:
signature.asc
Description: PGP signature
