Re: [Enable Designware XGMAC VLAN Stripping Feature 2/2] net: stmmac: dwxgmac2: Add support for HW-accelerated VLAN Stripping

[Florian Fainelli]

On 7/21/2023 9:12 AM, Ng, Boon Khai wrote:
> > -----Original Message-----
> > From: Florian Fainelli <f.fainelli@xxxxxxxxx>
> > Sent: Friday, July 21, 2023 11:59 PM
> > To: Ng, Boon Khai <boon.khai.ng@xxxxxxxxx>; Krzysztof Kozlowski
> > <krzk@xxxxxxxxxx>; Boon@xxxxxxxxxxxxxxxxxxxx; Khai@xxxxxxxxxxxxxxxxxxxx;
> > Ng, Boon Khai <boon.khai.ng@xxxxxxxxx>; Giuseppe Cavallaro
> > <peppe.cavallaro@xxxxxx>; Alexandre Torgue <alexandre.torgue@xxxxxxxxxxx>;
> > Jose Abreu <joabreu@xxxxxxxxxxxx>; David S . Miller <davem@xxxxxxxxxxxxx>;
> > Eric Dumazet <edumazet@xxxxxxxxxx>; Jakub Kicinski <kuba@xxxxxxxxxx>;
> > Paolo Abeni <pabeni@xxxxxxxxxx>; Maxime Coquelin
> > <mcoquelin.stm32@xxxxxxxxx>; netdev@xxxxxxxxxxxxxxx; linux-stm32@st-md-
> > mailman.stormreply.com; linux-arm-kernel@xxxxxxxxxxxxxxxxxxx; linux-
> > kernel@xxxxxxxxxxxxxxx
> > Cc: Shevchenko, Andriy <andriy.shevchenko@xxxxxxxxx>; Tham, Mun Yew
> > <mun.yew.tham@xxxxxxxxx>; Swee, Leong Ching
> > <leong.ching.swee@xxxxxxxxx>; G Thomas, Rohan
> > <rohan.g.thomas@xxxxxxxxx>; Shevchenko Andriy
> > <andriy.shevchenko@xxxxxxxxxxxxxxx>
> > Subject: Re: [Enable Designware XGMAC VLAN Stripping Feature 2/2] net:
> > stmmac: dwxgmac2: Add support for HW-accelerated VLAN Stripping
> >
> >
> >
> > On 7/21/2023 8:30 AM, Ng, Boon Khai wrote:
> > > > -----Original Message-----
> > > > From: Krzysztof Kozlowski <krzk@xxxxxxxxxx>
> > > > Sent: Friday, July 21, 2023 6:11 PM
> > > > To: Boon@xxxxxxxxxxxxxxxxxxxx; Khai@xxxxxxxxxxxxxxxxxxxx; "Ng
> > > > <boon.khai.ng"@intel.com; Giuseppe Cavallaro
> > > > <peppe.cavallaro@xxxxxx>; Alexandre Torgue
> > > > <alexandre.torgue@xxxxxxxxxxx>; Jose Abreu <joabreu@xxxxxxxxxxxx>;
> > > > David S . Miller <davem@xxxxxxxxxxxxx>; Eric Dumazet
> > > > <edumazet@xxxxxxxxxx>; Jakub Kicinski <kuba@xxxxxxxxxx>; Paolo Abeni
> > > > <pabeni@xxxxxxxxxx>; Maxime Coquelin <mcoquelin.stm32@xxxxxxxxx>;
> > > > netdev@xxxxxxxxxxxxxxx; linux-stm32@st- md-mailman.stormreply.com;
> > > > linux-arm-kernel@xxxxxxxxxxxxxxxxxxx; linux- kernel@xxxxxxxxxxxxxxx
> > > > Cc: Ng, Boon Khai <boon.khai.ng@xxxxxxxxx>; Shevchenko, Andriy
> > > > <andriy.shevchenko@xxxxxxxxx>; Tham, Mun Yew
> > > > <mun.yew.tham@xxxxxxxxx>; Swee, Leong Ching
> > > > <leong.ching.swee@xxxxxxxxx>; G Thomas, Rohan
> > > > <rohan.g.thomas@xxxxxxxxx>; Shevchenko Andriy
> > > > <andriy.shevchenko@xxxxxxxxxxxxxxx>
> > > > Subject: Re: [Enable Designware XGMAC VLAN Stripping Feature 2/2] net:
> > > > stmmac: dwxgmac2: Add support for HW-accelerated VLAN Stripping
> > > >
> > > > On 21/07/2023 08:26, Boon@xxxxxxxxxxxxxxxxxxxx wrote:
> > > > > From: Boon Khai Ng <boon.khai.ng@xxxxxxxxx>
> > > > >
> > > > > Currently, VLAN tag stripping is done by software driver in
> > > > > stmmac_rx_vlan(). This patch is to Add support for VLAN tag
> > > > > stripping by the MAC hardware and MAC drivers to support it.
> > > > > This is done by adding rx_hw_vlan() and set_hw_vlan_mode() callbacks
> > > > > at stmmac_ops struct which are called from upper software layer.
> > > > ...
> > > >
> > > > >    	if (priv->dma_cap.vlhash) {
> > > > >    		ndev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
> > > > >    		ndev->features |= NETIF_F_HW_VLAN_STAG_FILTER; diff --
> > > > git
> > > > > a/drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c
> > > > > b/drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c
> > > > > index 23d53ea04b24..bd7f3326a44c 100644
> > > > > --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c
> > > > > +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_platform.c
> > > > > @@ -543,6 +543,12 @@ stmmac_probe_config_dt(struct platform_device
> > > > *pdev, u8 *mac)
> > > > >    			plat->flags |= STMMAC_FLAG_TSO_EN;
> > > > >    	}
> > > > >
> > > > > +	/* Rx VLAN HW Stripping */
> > > > > +	if (of_property_read_bool(np, "snps,rx-vlan-offload")) {
> > > > > +		dev_info(&pdev->dev, "RX VLAN HW Stripping\n");
> > > >
> > > > Why? Drop.
> > >
> > > This is an dts option export to dts for user to choose whether or not
> > > they Want a Hardware stripping or a software stripping.
> > >
> > > May I know what is the reason to drop this?
> >
> > Because the networking stack already exposes knobs for drivers to advertise and
> > control VLAN stripping/insertion on RX/TX using ethtool and feature bits
> > (NETIF_F_HW_VLAN_CTAG_RX, NETIF_F_HW_VLAN_CTAG_TX).
>
> Hi Florian,
>
> Understood, but how does user choose to have the default option
> either hardware strip or software strip, when the device just boot up?

You need the hardware to advertise it and decide as a maintainer of that 
driver whether it makes sense to have one or the other behavior by default.

> I don’t think ethool can "remember" the setting once the device get rebooted?

If by "device" you mean a system that incorporates a XGMAC core, then I 
suppose that is true, though you could have some user-space logic that 
does remember the various ethtool options and re-applies them as soon as 
the device is made available to user-space, this would not be too far 
fetched.

> Any other suggestion of doing it other than using the dts method?

Let me ask you this question: what are you trying to solve by making 
this configurable? HW stripping should always be more efficient, should 
not it, if so, what would be the reasons for not enabling that by 
default? If not, then leave it off and let users enable it if they feel 
like they want it.
--
Florian