Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
To:     Koba Ko <koba.ko@canonical.com>
Cc:     "David S. Miller" <davem@davemloft.net>,
        Jakub Kicinski <kuba@kernel.org>, netdev@vger.kernel.org,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
References: <20210608032207.2923574-1-koba.ko@canonical.com>
 <84eb168e-58ff-0350-74e2-c55249eb258c@gmail.com>
 <CAJB-X+XFYa1cZgtJEL1KCNWviL3Y4X6EbN--rE8CD_9oD9EFyA@mail.gmail.com>
From:   Heiner Kallweit <hkallweit1@gmail.com>
Subject: Re: [PATCH] [v2] r8169: Use PHY_POLL when RTL8106E enable ASPM
Message-ID: <7a36c032-38fa-6aa6-fa0f-c3664850d8ea@gmail.com>
Date:   Tue, 8 Jun 2021 15:44:57 +0200
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:78.0) Gecko/20100101
 Thunderbird/78.11.0
MIME-Version: 1.0
In-Reply-To: <CAJB-X+XFYa1cZgtJEL1KCNWviL3Y4X6EbN--rE8CD_9oD9EFyA@mail.gmail.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 08.06.2021 12:43, Koba Ko wrote:
> On Tue, Jun 8, 2021 at 4:00 PM Heiner Kallweit <hkallweit1@gmail.com> wrote:
>>
>> On 08.06.2021 05:22, Koba Ko wrote:
>>> For RTL8106E, it's a Fast-ethernet chip.
>>> If ASPM is enabled, the link chang interrupt wouldn't be triggered
>>> immediately and must wait a very long time to get link change interrupt.
>>> Even the link change interrupt isn't triggered, the phy link is already
>>> established.
>>>
>>> Use PHY_POLL to watch the status of phy link and disable
>>> the link change interrupt when ASPM is enabled on RTL8106E.
>>>
>>> v2: Instead use PHY_POLL and identify 8106E by RTL_GIGA_MAC_VER_39.
>>>
>>
>> Still the issue description doesn't convince me that it's a hw bug
>> with the respective chip version. What has been stated so far:
>>
>> 1. (and most important) Issue doesn't occur in mainline because ASPM
>>    is disabled in mainline for r8169. Issue occurs only with a
>>    downstream kernel with ASPM enabled for r8169.
> 
> mainline kernel and enable L1, the issue is also observed.
> 
Yes, but enabling L1 via sysfs is at own risk.

>> 2. Issue occurs only with ASPM L1.1 not disabled, even though this chip
>>    version doesn't support L1 sub-states. Just L0s/L1 don't trigger
>>    the issue.
>>    The NIC doesn't announce L1.1 support, therefore PCI core won't
>>    enable L1 sub-states on the PCIe link between NIC and upstream
>>    PCI bridge.
> 
> More precisely, when L1 is enabled, the issue would be triggered.
> For RTL8106E,
> 1. Only disable L0s, pcie_aspm_enabled return 1, issue is triggered.
> 2. Only disable L1_1, pcie_aspm_enabled return 1, issue is triggered.
> 
> 3. Only disable L1, pcie_aspm_enabled return 0, issue is not triggered.
> 
>>
>> 3. Issue occurs only with a GBit-capable link partner. 100MBit link
>>    partners are fine. Not clear whether issue occurs with a specific
>>    Gbit link partner only or with GBit-capable link partners in general.
>>
>> 4. Only link-up interrupt is affected. Not link-down and not interrupts
>>    triggered by other interrupt sources.
>>
>> 5. Realtek couldn't confirm that there's such a hw bug on RTL8106e.
>>
>> One thing that hasn't been asked yet:
>> Does issue occur always if you re-plug the cable? Or only on boot?
>> I'm asking because in the dmesg log you attached to the bugzilla issue
>> the following looks totally ok.
>>
>> [   61.651643] r8169 0000:01:00.0 enp1s0: Link is Down
>> [   63.720015] r8169 0000:01:00.0 enp1s0: Link is Up - 100Mbps/Full - flow control rx/tx
>> [   66.685499] r8169 0000:01:00.0 enp1s0: Link is Down
> 
> Once the link is up,
> 1. If cable is unplug&plug immediately,  you wouldn't see the issue.
> 2. Unplug cable and wait a long time (~1Mins), then plug the cable,
> the issue appears again.
> 
This sounds runtime-pm-related. After 10s the NIC runtime-suspends,
and once the cable is re-plugged a PME is triggered that lets the
PCI core return the PCIe link from D3hot to D0.
If you re-plug the cable after such a longer time, do you see the
PCIe PME immediately in /proc/interrupts?

And if you set /sys/class/net/<if>/power/control to "on", does the
issue still occur?

>>
>>> Signed-off-by: Koba Ko <koba.ko@canonical.com>
>>> ---
>>>  drivers/net/ethernet/realtek/r8169_main.c | 21 +++++++++++++++++++--
>>>  1 file changed, 19 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/net/ethernet/realtek/r8169_main.c b/drivers/net/ethernet/realtek/r8169_main.c
>>> index 2c89cde7da1e..a59cbaef2839 100644
>>> --- a/drivers/net/ethernet/realtek/r8169_main.c
>>> +++ b/drivers/net/ethernet/realtek/r8169_main.c
>>> @@ -4914,6 +4914,19 @@ static const struct dev_pm_ops rtl8169_pm_ops = {
>>>
>>>  #endif /* CONFIG_PM */
>>>
>>> +static int rtl_phy_poll_quirk(struct rtl8169_private *tp)
>>> +{
>>> +     struct pci_dev *pdev = tp->pci_dev;
>>> +
>>> +     if (!pcie_aspm_enabled(pdev))
>>
>> That's the wrong call. According to what you said earlier you want to
>> check for L1 sub-states, not for ASPM in general.
> 
> As per described above, that's why use pcie_aspm_enabled here.
> 
>>
>>> +             return 0;
>>> +
>>> +     if (tp->mac_version == RTL_GIGA_MAC_VER_39)
>>> +             return 1;
>>> +
>>> +     return 0;
>>> +}
>>> +
>>>  static void rtl_wol_shutdown_quirk(struct rtl8169_private *tp)
>>>  {
>>>       /* WoL fails with 8168b when the receiver is disabled. */
>>> @@ -4991,7 +5004,10 @@ static const struct net_device_ops rtl_netdev_ops = {
>>>
>>>  static void rtl_set_irq_mask(struct rtl8169_private *tp)
>>>  {
>>> -     tp->irq_mask = RxOK | RxErr | TxOK | TxErr | LinkChg;
>>> +     tp->irq_mask = RxOK | RxErr | TxOK | TxErr;
>>> +
>>> +     if (!rtl_phy_poll_quirk(tp))
>>> +             tp->irq_mask |= LinkChg;
>>>
>>>       if (tp->mac_version <= RTL_GIGA_MAC_VER_06)
>>>               tp->irq_mask |= SYSErr | RxOverflow | RxFIFOOver;
>>> @@ -5085,7 +5101,8 @@ static int r8169_mdio_register(struct rtl8169_private *tp)
>>>       new_bus->name = "r8169";
>>>       new_bus->priv = tp;
>>>       new_bus->parent = &pdev->dev;
>>> -     new_bus->irq[0] = PHY_MAC_INTERRUPT;
>>> +     new_bus->irq[0] =
>>> +             (rtl_phy_poll_quirk(tp) ? PHY_POLL : PHY_MAC_INTERRUPT);
>>>       snprintf(new_bus->id, MII_BUS_ID_SIZE, "r8169-%x", pci_dev_id(pdev));
>>>
>>>       new_bus->read = r8169_mdio_read_reg;
>>>
>>