Re: [PATCH 00/11] hw/arm/aspeed: Split AspeedSoCState per 2400/2600/10x0

[Philippe Mathieu-Daudé]

On 24/10/23 18:24, Philippe Mathieu-Daudé wrote:
> Hi,
>
> This series is extracted for a bigger work.
>
> Cortex-A MP clusters (TYPE_A15MPCORE_PRIV) should create
> the ARM cores in its own state. Unfortunately we don't do
> it that way, and this model calls qemu_get_cpu().
>
> In order to remove the qemu_get_cpu() call there, we first
> need to rework some SoC users.
>
> This series rework the Aspeed SoC state, so it is clear
> what fields are really used by a SoC type (2400 / 2600 /
> 10x0). It will then be easier to have the MP cluster create
> the core instances.

Being a bit more verbose, as I was trying to explain to Cédric
on IRC.

The fby35 machine creates 2 SoCs:

static void fby35_init(MachineState *machine)
{
    Fby35State *s = FBY35(machine);

    fby35_bmc_init(s);
    fby35_bic_init(s);
}

- bmc is Aspeed2600 (A7 MPCORE)
- bic is Aspeed10x0 (M7)

If we were to create the bic before the bmc, as:

static void fby35_init(MachineState *machine)
{
    Fby35State *s = FBY35(machine);

    fby35_bic_init(s);
    fby35_bmc_init(s);
}

then the MPCORE misbehave as it calls qemu_get_cpu(0) which
returns the M7 from the bic.