Re: [PATCH 1/1] pcie: Do not set power state for some hot-plugged devices

["Annie.li" <annie.li@oracle.com>]

On 12/15/2021 1:05 AM, Gerd Hoffmann wrote:
> On Tue, Dec 14, 2021 at 09:53:12PM +0000, Annie Li wrote:
>> After the PCIe device is hot-plugged, the device's power state is
>> initialized as ON. However, the device isn't powered on yet, i.e.
>> the PCI_EXP_SYSCTL_PCC bit isn't set to PCI_EXP_SLTCTL_PWR_ON.
>> Later on, its power state will set back to OFF due to the non
>> PCI_EXP_SLTCTL_PWR_ON state. The device is invisible until
>> PCI_EXP_SLTCTL_PWR_ON is set.
>>
>> This may be appropriate for general PCIe hot-plug cases. However,
>> if the device is hot-plugged when the VM is in VM_STATE_PRELAUNCH
>> state, especially the system disk device, the firmware will fail
>> to find the system disk. As a result, the guest fails to boot.
> Maybe we should just not set DeviceState->hotplugged = true for devices
> added in VM_STATE_PRELAUNCH?  It's not actual hotplug (i.e. device added
> while the system is running) after all ...
Simply not setting "DeviceState->hotplugged" doesn't work. Devices 
created in
PHASE_MACHINE_READY phase are treated as hot-plugged devices. So I just 
tried
following change for the quick test, the device is still invisible to 
the firmware with
this change.

static void device_initfn(Object *obj)
{
...snip...
-     if (phase_check(PHASE_MACHINE_READY)) {
+    if (phase_check(PHASE_MACHINE_READY) && 
!runstate_check(RUN_STATE_PRELAUNCH)) {
         dev->hotplugged = 1;
         qdev_hot_added = true;
     }
...snip...
}

Thanks
Annie
>
> There are lots of places checking DeviceState->hotplugged, and I suspect
> we have similar issues elsewhere.
>
> take care,
>    Gerd
>