Re: [PATCH] r8169: fix RTL8117 Wake-on-Lan in DASH mode

[Heiner Kallweit <hkallweit1@gmail.com>]

On 12/2/2025 12:26 AM, René Rebe wrote:
> Hi,
> 
>> On 1. Dec 2025, at 22:12, Heiner Kallweit <hkallweit1@gmail.com> wrote:
>>
>> On 12/1/2025 9:31 PM, René Rebe wrote:
>>> Hi
>>>
>>>> On 1. Dec 2025, at 21:15, Heiner Kallweit <hkallweit1@gmail.com> wrote:
>>>>
>>>> On 12/1/2025 8:17 PM, René Rebe wrote:
>>>>> Wake-on-Lan does currently not work in DASH mode, e.g. the ASUS Pro WS
>>>>> X570-ACE with RTL8168fp/RTL8117.
>>>>>
>>>>> Fix by not returning early in rtl_prepare_power_down when dash_enabled.
>>>> Good
>>>>
>>>>> While this fixes WOL, it still kills the OOB RTL8117 remote management
>>>>> BMC connection. Fix by not calling rtl8168_driver_stop if wol is enabled.
>>>>>
>>>> You mean remote management whilst system is powered down and waiting
>>>> for a WoL packet? Note that link speed is reduced to a minimum then,
>>>> and DMA is disabled. Who would drive the MAC?
>>>> Realtek doesn't provide any chip documentation, therefore it's hard to
>>>> say what is expected from the MAC driver in DASH case.
>>>
>>> This RTL8117 has a 250 or 400 MHz MIPS cpu inside that runs
>>> a out-of-band linux kernel. Pretty sketchy low-quality setup unfortunately:
>>>
>>> https://www.youtube.com/watch?v=YqEa8Gd1c2I&t=1695s
>>>>
>>>>> While at it, enable wake on magic packet by default, like most other
>>>>> Linux drivers do.
>>>>>
>>>> It's by intent that WoL is disabled per default. Most users don't use WoL
>>>> and would suffer from higher power consumption if system is suspended
>>>> or powered down.
>>>
>>> It was just a suggestion, I can use ethtool, it is the only driver that does
>>> not have it on by default in all the systems I have.
>>>
>>>> Which benefit would you see if WoL would be enabled by default
>>>> (in DASH and non-DASH case)?
>>>
>>> So it just works when pro-sumers want to wake it up, not the most
>>> important detail of the patch.
>>>
>>>>> Signed-off-by: René Rebe <rene@exactco.de>
>>>>
>>>> Your patch apparently is meant to be a fix. Therefore please add Fixes
>>>> tag and address to net tree.
>>>> https://www.kernel.org/doc/Documentation/networking/netdev-FAQ.rst
>>>> And please add all netdev maintainers when re-submitting.
>>>> scripts/get_maintainer.pl provides all needed info.
>>>
>>> Yes, I realized after sending. The only Fixes: would be the original
>>> change adding the DASH support I assume?
>>>
>>> Any opinion re not stopping DASH on if down? IMHO taking a
>>> link down should not break the remote management connection.
>>>
>> I have no clue how the OOB BMC interacts with MAC/PHY, and I have no
>> hw supporting DASH to test. So not really a basis for an opinion.
>> However: DASH has been existing on Realtek hw for at least 15 yrs,
>> and I'm not aware of any complaint related to what you mention.
>> So it doesn't seem to be a common use case.
> 
> Well the Asus Control Center Express is so bad and barely working
> it does not surprise me nobody is using it. We reversed the protocol
> and wrote some script and hacked VNC client to make it useful for us.
> 
> The Asus GPL compliance code dump for the MIPS Linux BMC system
> has some rtl8168_oob or so BMC side driver for it to learn more details.
> 
Do you have a link to this code?

> Maybe I should backup a fork of it on my GitHub to archive it.
> 
> Given the BMC should be reachable, would be acceptable if I work
> out a patch to not take the phy down and always keep it up even
> for if down and module unload when in “dash” mode?
> 
Yes

>> There are different generations of DASH in RTL8168DP, RTL8168EP,
>> RTL8117, variants of RTL8125, RTL8127 etc. Having said that,
>> there's a certain chance of a regression, even if the patch works
>> correctly on your system. Therefore I'd prefer to handle any additional
>> changes in separate patches, to facilitate bisecting in case of a
>> regression.
> 
> Of course, will sent separate patches for each topic.
> 
> What about defaulting to wol by magic like most other linux
> drivers? IMHO the drivers should all be have similar and not
> all have some other defaults. In theory it could be made
> a tree-wide kconfig if users and distros care enough what the
> global default should be.
> 
I'm not in favor of enabling WoL per default, as it prevents
powering down the PHY if system is suspended / shut down.
This impacts especially users of mobile devices.
"Others do it too" for me is a weak argument, if no one can
explain why it's a good thing what they're doing.

> 	René
> 
>>> I probably would need to single step thru the driver init to find out
>>> what reset stops the out of band traffic there, too.
>>>
>>> René
>>>
>>>>> ---
>>>>>
>>>>> There is still another issue that should be fixed: the dirver init
>>>>> kills the OOB BMC connection until if up, too. We also should probaly
>>>>> not even conditionalize rtl8168_driver_stop on wol_enabled as the BMC
>>>>> should always be accessible. IMHO even on module unload.
>>>>>
>>>>> ---
>>>>> drivers/net/ethernet/realtek/r8169_main.c | 9 +++++----
>>>>> 1 file changed, 5 insertions(+), 4 deletions(-)
>>>>>
>>>>> diff --git a/drivers/net/ethernet/realtek/r8169_main.c b/drivers/net/ethernet/realtek/r8169_main.c
>>>>> index 853aabedb128..e2f9b9027fe2 100644
>>>>> --- a/drivers/net/ethernet/realtek/r8169_main.c
>>>>> +++ b/drivers/net/ethernet/realtek/r8169_main.c
>>>>> @@ -2669,9 +2669,6 @@ static void rtl_wol_enable_rx(struct rtl8169_private *tp)
>>>>>
>>>>> static void rtl_prepare_power_down(struct rtl8169_private *tp)
>>>>> {
>>>>> - if (tp->dash_enabled)
>>>>> - return;
>>>>> -
>>>>> if (tp->mac_version == RTL_GIGA_MAC_VER_32 ||
>>>>>   tp->mac_version == RTL_GIGA_MAC_VER_33)
>>>>> rtl_ephy_write(tp, 0x19, 0xff64);
>>>>> @@ -4807,7 +4804,7 @@ static void rtl8169_down(struct rtl8169_private *tp)
>>>>> rtl_disable_exit_l1(tp);
>>>>> rtl_prepare_power_down(tp);
>>>>>
>>>>> - if (tp->dash_type != RTL_DASH_NONE)
>>>>> + if (tp->dash_type != RTL_DASH_NONE && !tp->saved_wolopts)
>>>>> rtl8168_driver_stop(tp);
>>>>> }
>>>>>
>>>>> @@ -5406,6 +5403,7 @@ static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
>>>>> tp->pci_dev = pdev;
>>>>> tp->supports_gmii = ent->driver_data == RTL_CFG_NO_GBIT ? 0 : 1;
>>>>> tp->ocp_base = OCP_STD_PHY_BASE;
>>>>> + tp->saved_wolopts = WAKE_MAGIC;
>>>>>
>>>>> raw_spin_lock_init(&tp->mac_ocp_lock);
>>>>> mutex_init(&tp->led_lock);
>>>>> @@ -5565,6 +5563,9 @@ static int rtl_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
>>>>> if (rc)
>>>>> return rc;
>>>>>
>>>>> + if (tp->saved_wolopts)
>>>>> + __rtl8169_set_wol(tp, tp->saved_wolopts);
>>>>> +
>>>>> rc = register_netdev(dev);
>>>>> if (rc)
>>>>> return rc;
>>>>
>>>
>>
>