Re: [PATCH] ACPI: PM: Re-enable ACPI GPE if it's already enabled

[Kai-Heng Feng <kai.heng.feng@canonical.com>]

> On Nov 24, 2020, at 22:00, Rafael J. Wysocki <rafael@kernel.org> wrote:
> 
> On Tue, Nov 24, 2020 at 8:36 AM Kai-Heng Feng
> <kai.heng.feng@canonical.com> wrote:
>> 
>> Dell Precision 5550 fails to detect Thunderbolt device hotplug events,
>> once the Thunderbolt device and its root port are runtime-suspended to
>> D3cold.
>> 
>> While putting the entire hierarchy to D3cold, the root port ACPI GPE is
>> enabled via acpi_pci_propagate_wakeup() when suspending Thunderbolt
>> bridges/switches. So when putting the root port to D3cold as last step,
>> ACPI GPE is untouched as it's already enabled.
>> 
>> However, platform may need PCI devices to be in D3hot or PME enabled
>> prior enabling GPE to make it work.
> 
> What platforms and why.

Dell Precision 5550. Its thunderbolt port can't detect newly plugged thunderbolt devices.

> 
>> So re-enable ACPI GPE to address this.
>> 
>> Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
>> ---
>> drivers/acpi/device_pm.c | 13 ++++++-------
>> 1 file changed, 6 insertions(+), 7 deletions(-)
>> 
>> diff --git a/drivers/acpi/device_pm.c b/drivers/acpi/device_pm.c
>> index 94d91c67aeae..dc25d9d204ae 100644
>> --- a/drivers/acpi/device_pm.c
>> +++ b/drivers/acpi/device_pm.c
>> @@ -757,11 +757,10 @@ static int __acpi_device_wakeup_enable(struct acpi_device *adev,
>> 
>>        mutex_lock(&acpi_wakeup_lock);
>> 
>> -       if (wakeup->enable_count >= max_count)
>> -               goto out;
>> -
>> -       if (wakeup->enable_count > 0)
>> -               goto inc;
>> +       if (wakeup->enable_count > 0) {
>> +               acpi_disable_gpe(wakeup->gpe_device, wakeup->gpe_number);
>> +               acpi_disable_wakeup_device_power(adev);
>> +       }
> 
> An event occurring at this point may be lost after this patch.

Yes, so this approach is not optimal.

> 
> It looks like you are trying to work around a hardware issue.  

Windows doesn't have this issue. So I don't think it's hardware issue.

> Can you
> please describe that issue in detail?

The GPE used by Thunderbolt root port, was previously enabled by Thunderbolt switches/bridges.
So when the GPE is already enabled when Thunderbolt root port is suspended.
However, the GPE needs to be enabled after root port is suspended, and that's the approach this patch takes.

Is there any actual hardware benefits from acpi_pci_propagate_wakeup()?
If there's no actual device benefits from it, we can remove it and only enable GPE for the root port.
Otherwise we need to quirk off Thunderbolt bridges/switches, since their native PME just work without the need to enable root port GPE.

Kai-Heng

> 
>> 
>>        error = acpi_enable_wakeup_device_power(adev, target_state);
>>        if (error)
>> @@ -777,8 +776,8 @@ static int __acpi_device_wakeup_enable(struct acpi_device *adev,
>>        acpi_handle_debug(adev->handle, "GPE%2X enabled for wakeup\n",
>>                          (unsigned int)wakeup->gpe_number);
>> 
>> -inc:
>> -       wakeup->enable_count++;
>> +       if (wakeup->enable_count < max_count)
>> +               wakeup->enable_count++;
>> 
>> out:
>>        mutex_unlock(&acpi_wakeup_lock);
>> --
>> 2.29.2