Re: [Qemu-devel] [PATCH v6 4/4] hw/intc/arm_gicv3_kvm: Reset GICv3 cpu interface registers

[Peter Maydell]

On 28 November 2016 at 16:01, Vijay Kilari <address@hidden> wrote:
> On Mon, Nov 28, 2016 at 6:31 PM, Peter Maydell <address@hidden> wrote:
>> On 23 November 2016 at 12:39,  <address@hidden> wrote:
>>> From: Vijaya Kumar K <address@hidden>
>>>
>>> Reset CPU interface registers of GICv3 when CPU is reset.
>>> For this, object interface is used, which is called from
>>> arm_cpu_reset function.
>>>
>>> Signed-off-by: Vijaya Kumar K <address@hidden>
>>
>> This approach doesn't handle the SMP case correctly --
>> when a CPU is reset then the CPU interface for that CPU
>> (and only that CPU) should be reset. Your code will
>> reset every CPU interface every time any CPU is reset.
>
> arm_cpu_reset is not called when particular cpu is reset?.
> Is it called for all cpus?.

It's called to reset a particular CPU (so it will be called
once for each CPU).

> OR object_child_foreach_recursive() is calling to reset cpu
> interfaces of
> all cpus?.

It does "look through the whole graph of objects in the
simulation and call the function on anything in the
graph that implements the interface". I've just seen that
your code is doing "ignore the call if the CPU that
triggered this isn't the one we care about", though --
I missed that the first time reading the code.

Still I would prefer it if we did this with the same
mechanism for both TCG and KVM. A generic mechanism for
"let the CPU reset trigger reset of many other devices in the
system" isn't widely useful because real hardware doesn't
have that kind of action-at-a-distance behaviour.

thanks
-- PMM