Re: [PATCH V8 3/8] hw/acpi: Update ACPI GED framework to support vCPU Hotplug

[Zhao Liu <zhao1.liu@intel.com>]

Hi Salil,

On Fri, May 03, 2024 at 07:59:32PM +0000, Salil Mehta wrote:
> Date: Fri, 3 May 2024 19:59:32 +0000
> From: Salil Mehta <salil.mehta@huawei.com>
> Subject: RE: [PATCH V8 3/8] hw/acpi: Update ACPI GED framework to support
>  vCPU Hotplug
> 
> Hello,
> 
> Sorry, I missed this earlier.
>
> >  From: Zhao Liu <zhao1.liu@intel.com>
> >  Sent: Wednesday, March 13, 2024 6:14 AM
> >  To: Salil Mehta <salil.mehta@huawei.com>
> >  
> >  Hi Salil,
> >  
> >  It seems my comment [1] in v7 was missed, but I still hit the same issue. Pls
> >  let me paste the previous comment here again.
> >  
> >  [1]: https://lore.kernel.org/qemu-devel/ZXCqp32ggIFvUweu@intel.com/
> 
> Yes, I have this in my mind. 
> 
> >  
> >  [snip]
> >  
> >  > @@ -400,6 +411,12 @@ static void acpi_ged_initfn(Object *obj)
> >  >      memory_region_init_io(&ged_st->regs, obj, &ged_regs_ops, ged_st,
> >  >                            TYPE_ACPI_GED "-regs", ACPI_GED_REG_COUNT);
> >  >      sysbus_init_mmio(sbd, &ged_st->regs);
> >  > +
> >  > +    memory_region_init(&s->container_cpuhp, OBJECT(dev), "cpuhp
> >  container",
> >  > +                       ACPI_CPU_HOTPLUG_REG_LEN);
> >  > +    sysbus_init_mmio(SYS_BUS_DEVICE(dev), &s->container_cpuhp);
> >  > +    cpu_hotplug_hw_init(&s->container_cpuhp, OBJECT(dev),
> >  > +                        &s->cpuhp_state, 0);
> >  >  }
> >  >
> >  
> >  I find this cpu_hotplug_hw_init() can still cause qtest errors (for v8) on x86
> >  platforms as you mentioned in v6:
> >  https://lore.kernel.org/qemu-devel/15e70616-6abb-63a4-17d0-
> >  820f4a254607@opnsrc.net/T/#m108f102b2fe92b7dd7218f2f942f7b233a9d6a
> >  f3
> >  
> >  IIUC, microvm machine has its own 'possible_cpus_arch_ids' and that is
> >  inherited from its parent x86 machine.
> >  
> >  The above error is because device-introspect-test sets the none-machine:
> >  
> >  # starting QEMU: exec ./qemu-system-i386 -qtest unix:/tmp/qtest-
> >  3094820.sock -qtest-log /dev/null -chardev socket,path=/tmp/qtest-
> >  3094820.qmp,id=char0 -mon chardev=char0,mode=control -display none -
> >  audio none -nodefaults -machine none -accel qtest
> >  
> >  So what about just checking mc->possible_cpu_arch_ids instead of an
> >  assert in cpu_hotplug_hw_init()?
> >  
> >  diff --git a/hw/acpi/cpu.c b/hw/acpi/cpu.c index
> >  4b24a2500361..303f1f1f57bc 100644
> >  --- a/hw/acpi/cpu.c
> >  +++ b/hw/acpi/cpu.c
> >  @@ -221,7 +221,10 @@ void cpu_hotplug_hw_init(MemoryRegion *as,
> >  Object *owner,
> >       const CPUArchIdList *id_list;
> >       int i;
> >  
> >  -    assert(mc->possible_cpu_arch_ids);
> >  +    if (!mc->possible_cpu_arch_ids) {
> >  +        return;
> >  +    }
> >  +
> 
> 
> Yes, we can do this with some debug print or trace maybe.

Here it is just to return early without touching mc->possible_cpu_arch_ids().
If you adopt this workaround, then in the meantime I suggest adding a comment
to this "if" to clarify that it is for compatibility with certain machines
that do not implement mc->possible_cpu_arch_ids().
 
> >       id_list = mc->possible_cpu_arch_ids(machine);
> >       state->dev_count = id_list->len;
> >       state->devs = g_new0(typeof(*state->devs), state->dev_count);
> >  
> >  This check seems to be acceptable in the general code path? Not all
> >  machines have possible_cpu_arch_ids, after all.
> 
> True. BTW, have you tested this with Qtest?

Yes, by "make check" on x86 platform. This workaround can help us pass
the x86 tests.

Regards,
Zhao