Re: [RFC v1] arm/virt: Add memory hot remove support

[Auger Eric]

Hi Shameer,

On 3/26/20 12:14 PM, Shameerali Kolothum Thodi wrote:
> Hi Eric,
> 
>> -----Original Message-----
>> From: Auger Eric [mailto:address@hidden]
>> Sent: 26 March 2020 11:01
>> To: Shameerali Kolothum Thodi <address@hidden>;
>> address@hidden; address@hidden
>> Cc: address@hidden; address@hidden; address@hidden;
>> xuwei (O) <address@hidden>; Zengtao (B) <address@hidden>;
>> Linuxarm <address@hidden>; Anshuman Khandual
>> <address@hidden>
>> Subject: Re: [RFC v1] arm/virt: Add memory hot remove support
>>
>> Hi Shameer,
>>
>> On 3/18/20 1:37 PM, Shameer Kolothum wrote:
>>> This adds support for memory hot remove on arm/virt that
>>> uses acpi ged device.
>>
>> I gave this a try and it works fine if the PCDIMM slot was initially
>> hotplugged:
>> (QEMU) object-add qom-type=memory-backend-ram id=mem1
>> props.size=4294967296
>> {"return": {}}
>> (QEMU) device_add driver=pc-dimm  id=pcdimm1 memdev=mem1
>> (QEMU) device_del id=pcdimm1
>> {"return": {}}
>>
>> on guest I can see:
>> [   82.466321] Offlined Pages 262144
>> [   82.541712] Offlined Pages 262144
>> [   82.589236] Offlined Pages 262144
>> [   82.969166] Offlined Pages 262144
>>
>> However I noticed that if qemu is launched directly with
>>
>> -m 16G,maxmem=32G,slots=2 \
>> -object memory-backend-ram,id=mem1,size=4G \
>> -device pc-dimm,memdev=mem1,id=dimm1,driver=pc-dimm -device
>>
>> and then in the qmp shell:
>> (QEMU) device_del id=dimm1
>>
>> the hot-unplug fails in guest:
>>
>> [   78.897407] Offlined Pages 262144
>> [   79.260811] Offlined Pages 262144
>> [   79.308105] Offlined Pages 262144
>> [   79.333675] page:fffffe00137d1f40 refcount:1 mapcount:0
>> mapping:ffff0004ea9f20b1 index:0xaaab11c6e
>> [   79.335927] anon flags: 0x17ffff8000080024(uptodate|active|swapbacked)
>> [   79.337571] raw: 17ffff8000080024 dead000000000100
>> dead000000000122
>> ffff0004ea9f20b1
>> [   79.339502] raw: 0000000aaab11c6e 0000000000000000 00000001ffffffff
>> ffff0004fd4e3000
>> [   79.341701] page dumped because: unmovable page
>> [   79.342887] page->mem_cgroup:ffff0004fd4e3000
>> [   79.354729] page:fffffe00137d1f40 refcount:1 mapcount:0
>> mapping:ffff0004ea9f20b1 index:0xaaab11c6e
>> [   79.357012] anon flags: 0x17ffff8000080024(uptodate|active|swapbacked)
>> [   79.358658] raw: 17ffff8000080024 dead000000000100
>> dead000000000122
>> ffff0004ea9f20b1
>> [   79.360611] raw: 0000000aaab11c6e 0000000000000000 00000001ffffffff
>> ffff0004fd4e3000
>> [   79.362560] page dumped because: unmovable page
>> [   79.363742] page->mem_cgroup:ffff0004fd4e3000
>> [   79.368636] memory memory20: Offline failed.
>>
>> I did not expect this. The PCDIMM slot in that case does not seem to be
>> interpreted as a hot-unpluggable one (?). I added Anshuman in cc.
> 
> Could you please try adding "movable_node" to qemu guest kernel command line 
> params.
> This will prevent any kernel allocation from hotplugable memory nodes which I 
> think is
> causing the behavior you are seeing.

Effectively, when adding the movable_node option in the guest kernel
parameters, I get the following traces:

[   29.581418] Offlined Pages 262144
[   29.663605] Offlined Pages 262144
[   29.714225] Offlined Pages 262144
[   30.222953] Offlined Pages 262144
[   30.314288] Built 1 zonelists, mobility grouping on.  Total pages:
4076898
[   30.316067] Policy zone: Normal

Thanks

Eric

> 
> Thanks,
> Shameer
> 
> 
>> Thanks
>>
>> Eric
>>
>>
>>
>>>
>>> Signed-off-by: Shameer Kolothum <address@hidden>
>>> ---
>>>  -RFC because linux kernel support for mem hot remove is just queued
>>>   for 5.7[1].
>>>  -Tested with guest kernel 5.6-rc5 + [1]
>>>
>>> 1. https://patchwork.kernel.org/cover/11419301/
>>> ---
>>>  hw/acpi/generic_event_device.c | 28 +++++++++++++++++
>>>  hw/arm/virt.c                  | 56
>> ++++++++++++++++++++++++++++++++--
>>>  2 files changed, 82 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/hw/acpi/generic_event_device.c
>> b/hw/acpi/generic_event_device.c
>>> index 021ed2bf23..3e28c110fa 100644
>>> --- a/hw/acpi/generic_event_device.c
>>> +++ b/hw/acpi/generic_event_device.c
>>> @@ -182,6 +182,32 @@ static void
>> acpi_ged_device_plug_cb(HotplugHandler *hotplug_dev,
>>>      }
>>>  }
>>>
>>> +static void acpi_ged_unplug_request_cb(HotplugHandler *hotplug_dev,
>>> +                                       DeviceState *dev, Error
>> **errp)
>>> +{
>>> +    AcpiGedState *s = ACPI_GED(hotplug_dev);
>>> +
>>> +    if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
>>> +        acpi_memory_unplug_request_cb(hotplug_dev,
>> &s->memhp_state, dev, errp);
>>> +    } else {
>>> +        error_setg(errp, "acpi: device unplug request for unsupported
>> device"
>>> +                   " type: %s", object_get_typename(OBJECT(dev)));
>>> +    }
>>> +}
>>> +
>>> +static void acpi_ged_unplug_cb(HotplugHandler *hotplug_dev,
>>> +                               DeviceState *dev, Error **errp)
>>> +{
>>> +    AcpiGedState *s = ACPI_GED(hotplug_dev);
>>> +
>>> +    if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
>>> +        acpi_memory_unplug_cb(&s->memhp_state, dev, errp);
>>> +    } else {
>>> +        error_setg(errp, "acpi: device unplug for unsupported device"
>>> +                   " type: %s", object_get_typename(OBJECT(dev)));
>>> +    }
>>> +}
>>> +
>>>  static void acpi_ged_send_event(AcpiDeviceIf *adev, AcpiEventStatusBits
>> ev)
>>>  {
>>>      AcpiGedState *s = ACPI_GED(adev);
>>> @@ -286,6 +312,8 @@ static void acpi_ged_class_init(ObjectClass *class,
>> void *data)
>>>      dc->vmsd = &vmstate_acpi_ged;
>>>
>>>      hc->plug = acpi_ged_device_plug_cb;
>>> +    hc->unplug_request = acpi_ged_unplug_request_cb;
>>> +    hc->unplug = acpi_ged_unplug_cb;
>>>
>>>      adevc->send_event = acpi_ged_send_event;
>>>  }
>>> diff --git a/hw/arm/virt.c b/hw/arm/virt.c
>>> index 94f93dda54..91974e4e80 100644
>>> --- a/hw/arm/virt.c
>>> +++ b/hw/arm/virt.c
>>> @@ -2096,11 +2096,62 @@ static void
>> virt_machine_device_plug_cb(HotplugHandler *hotplug_dev,
>>>      }
>>>  }
>>>
>>> +static void virt_dimm_unplug_request(HotplugHandler *hotplug_dev,
>>> +                                     DeviceState *dev, Error
>> **errp)
>>> +{
>>> +    VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
>>> +    Error *local_err = NULL;
>>> +
>>> +    if (!vms->acpi_dev) {
>>> +        error_setg(errp,
>>> +                   "memory hotplug is not enabled: missing acpi-ged
>> device");
>>> +        goto out;
>>> +    }
>>> +
>>> +    hotplug_handler_unplug_request(HOTPLUG_HANDLER(vms->acpi_dev),
>> dev,
>>> +                                   &local_err);
>>> +out:
>>> +    error_propagate(errp, local_err);
>>> +}
>>> +
>>> +static void virt_dimm_unplug(HotplugHandler *hotplug_dev,
>>> +                             DeviceState *dev, Error **errp)
>>> +{
>>> +    VirtMachineState *vms = VIRT_MACHINE(hotplug_dev);
>>> +    Error *local_err = NULL;
>>> +
>>> +    hotplug_handler_unplug(HOTPLUG_HANDLER(vms->acpi_dev), dev,
>> &local_err);
>>> +    if (local_err) {
>>> +        goto out;
>>> +    }
>>> +
>>> +    pc_dimm_unplug(PC_DIMM(dev), MACHINE(vms));
>>> +    object_property_set_bool(OBJECT(dev), false, "realized", NULL);
>>> +
>>> + out:
>>> +    error_propagate(errp, local_err);
>>> +}
>>> +
>>>  static void virt_machine_device_unplug_request_cb(HotplugHandler
>> *hotplug_dev,
>>>                                            DeviceState *dev, Error
>> **errp)
>>>  {
>>> -    error_setg(errp, "device unplug request for unsupported device"
>>> -               " type: %s", object_get_typename(OBJECT(dev)));
>>> +    if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
>>> +        virt_dimm_unplug_request(hotplug_dev, dev, errp);
>>> +    } else {
>>> +        error_setg(errp, "device unplug request for unsupported device"
>>> +                   " type: %s", object_get_typename(OBJECT(dev)));
>>> +    }
>>> +}
>>> +
>>> +static void virt_machine_device_unplug_cb(HotplugHandler *hotplug_dev,
>>> +                                          DeviceState *dev, Error
>> **errp)
>>> +{
>>> +    if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
>>> +        virt_dimm_unplug(hotplug_dev, dev, errp);
>>> +    } else {
>>> +        error_setg(errp, "virt: device unplug for unsupported device"
>>> +                   " type: %s", object_get_typename(OBJECT(dev)));
>>> +    }
>>>  }
>>>
>>>  static HotplugHandler *virt_machine_get_hotplug_handler(MachineState
>> *machine,
>>> @@ -2181,6 +2232,7 @@ static void virt_machine_class_init(ObjectClass
>> *oc, void *data)
>>>      hc->pre_plug = virt_machine_device_pre_plug_cb;
>>>      hc->plug = virt_machine_device_plug_cb;
>>>      hc->unplug_request = virt_machine_device_unplug_request_cb;
>>> +    hc->unplug = virt_machine_device_unplug_cb;
>>>      mc->numa_mem_supported = true;
>>>      mc->auto_enable_numa_with_memhp = true;
>>>      mc->default_ram_id = "mach-virt.ram";
>>>
>