RE: [PATCH v5] net: stmmac: fix 'ethtool -P' return -EBUSY

-----Original Message-----
From: Hao Chen <redacted>
Sent: 2021年7月21日 9:35
To: Heiner Kallweit <hkallweit1@gmail.com>
Cc: alexandre.torgue@foss.st.com; joabreu@synopsys.com;
davem@davemloft.net; kuba@kernel.org; mcoquelin.stm32@gmail.com;
linux@armlinux.org.uk; netdev@vger.kernel.org;
linux-stm32@st-md-mailman.stormreply.com; Joakim Zhang
[off-list ref]; peppe.cavallaro@st.com
Subject: Re: [PATCH v5] net: stmmac: fix 'ethtool -P' return -EBUSY


在 2021/7/19 下午9:57, Heiner Kallweit 写道:

quoted

On 19.07.2021 15:08, Hao Chen wrote:

quoted

The permanent mac address should be available for query when the
device is not up.
NetworkManager, the system network daemon, uses 'ethtool -P' to
obtain the permanent address after the kernel start. When the network
device is not up, it will return the device busy error with 'ethtool
-P'. At that time, it is unable to access the Internet through the
permanent address by NetworkManager.

At first a few formal aspects:
- You don't get a lot of new friends when posting a new version every
   few hours. Consolidate feedback and then post a new version,
   typically not more than one version per day. Mileage of maintainers
   may vary here.
- When posting a new version include a change log.
- Properly annotate the patch as net vs. net-next.
- If you declare something to be a fix, add a Fixes tag.

Thanks for your guidance.

quoted

quoted

I think that the '.begin' is not used to check if the device is up,
it's just a pre hook for ethtool. We shouldn't check if the device is
up there.

Supposedly the check is there for a reason. Your statement leaves the
impression that you're not aware of why the check exists.
Therefore: First understand this, and then explain why it's safe to
remove the check. This drivers uses runtime pm, so device might be in
PCI D3 if interface is down (don't know this driver in detail).
Therefore registers may not be accessible what may impact certain
ethtool ops.

The network manager obtains and stores the permanent MAC address before
the network

card is up, and then the user can specify the MAC address of the device in the
desktop network

manager for networking.


If the driver needs the network card up to obtain the permanent MAC address,
the user will

not be able to specify the device MAC address for networking after power on.

I'm not sure quite understand what's your requirements, AFAIK, NIC driver has
.ndo_set_mac_address to set mac address, and it can be done whether ifup or ifdown.

I don't think we should detect that the network card is up in '. Begin', which will
cause that

all the ethtool commands can't be used when the network card is down. If
some ethtool

commands can only be used in the up state, they should be detected in the
corresponding

ethool OPS function instead of in the pre hook. This is too rude and
unreasonable.


As for safe, remove the '. Begin' implementation. After modification, it
should be safer to

operate Ethernet in the down state than in the up state.

As Heiner mentioned, ethtool ops may access NIC hardware registers which need
clocks enabled. Now we have .begin that means clocks enabled when use ethtool,
if you remove .begin, you need ensure clocks enabled whenever ethtool used. So,
remove .begin directly wound not be acceptable.

I have checked the '. Begin' implementation of other drivers, most of
which support the

submission of NIC driver for the first time. They are too long to know
why '. Begin' is

implemented. I suspect that they have not noticed the usage of '. Begin'.

In cpsw NIC driver, commit ID 7898b1daf0555, a bug was found to be fixed:

"The CPSW might be suspended by RPM if all ethX interfaces are down,

but it still could be accesible through ethtool interfce. In this case
ethtool operations, requiring registers access, will cause L3 errors and
CPSW crash."

My machine uses platform GMAC. I don't have PCI network card. I can't
test the use

of PCI network card. Through cat /sys/devices/... /power/
runtime_status, return

unsupported. Therefore, runtime PM is not supported.

I have a GMAC5.10 at my hand, and check the runtime PM status on net-next tree:
~# cat /sys/class/net/eth0/device/power/runtime_status
Active

Another main you said:
"By browse the code and test it,  I find that stmmac not support runtime_status,

pm_runtime_get_sync will return -13 errno. So shouldn't modify it in this way.

I think we should just delete the implementation of begin, it's a pre hook for ethtool."

I don't know why " pm_runtime_get_sync will return -13 errno ". If so, STMMAC driver still
can work at you side? Clocks always enabled after probed?

Best Regards,
Joakim Zhang

To sum up, I think it's better to remove the '. Begin' implementation here.

Or use the following implementation:

   static int stmmac_ethtool_begin(struct net_device *dev)
{
           struct stmmac_priv *priv = netdev_priv(dev);
           int ret;

           if (strcmp(priv->device->bus_type->name, "pci"))
                   return 0;

           return pm_runtime_resume_and_get(dev);
}

   static void stmmac_ethtool_complete(struct net_device *dev)
{
           struct stmmac_priv *priv = netdev_priv(dev);

           if (strcmp(priv->device->bus_type->name, "pci"))
return;

pm_runtime_put(priv->device);
   }

  .begin = stmmac_ethtool_begin,
  .complete = stmmac_ethtool_complete,

quoted

quoted

Signed-off-by: Hao Chen <redacted>
---
  drivers/net/ethernet/stmicro/stmmac/stmmac_ethtool.c | 8 --------
  1 file changed, 8 deletions(-)

diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_ethtool.c

b/drivers/net/ethernet/stmicro/stmmac/stmmac_ethtool.c

quoted

quoted

index d0ce608b81c3..8901dc9f758e 100644

--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_ethtool.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_ethtool.c

@@ -410,13 +410,6 @@ static void stmmac_ethtool_setmsglevel(struct

net_device *dev, u32 level)

quoted

quoted

  }

-static int stmmac_check_if_running(struct net_device *dev)
-{
-	if (!netif_running(dev))
-		return -EBUSY;
-	return 0;
-}
-
  static int stmmac_ethtool_get_regs_len(struct net_device *dev)
  {
  	struct stmmac_priv *priv = netdev_priv(dev);

@@ -1073,7 +1066,6 @@ static int stmmac_set_tunable(struct net_device

*dev,

quoted

quoted

  static const struct ethtool_ops stmmac_ethtool_ops = {
  	.supported_coalesce_params = ETHTOOL_COALESCE_USECS |
  				     ETHTOOL_COALESCE_MAX_FRAMES,
-	.begin = stmmac_check_if_running,
  	.get_drvinfo = stmmac_ethtool_getdrvinfo,
  	.get_msglevel = stmmac_ethtool_getmsglevel,
  	.set_msglevel = stmmac_ethtool_setmsglevel,