nxp-ls1028ardb support for MTU 8192 on felix-switch

[hittzt]

Hi experts,

Is there any plan for nxp-ls1028ardb board to support jumbo frame(like set MTU=8192) in future?

Or is it possible to support the jumbo frame on felix-switch on ls1028ardb?

This may help to promote the performance of the switch.

Would you please help to verify this issue?

Thanks,

Zhantao

[lomah69097]

Is there any plan for nxp-ls1028ardb board to support jumbo frame(like set MTU=8192) in future?

Or is it possible to support the jumbo frame on felix-switch on ls1028ardb?

This may help to promote the performance of the switch.myaccountaccess

me and my friend face a problem so i am looking for a solution 

thank you 

[yipingwang]

Jumbo frame for felix is going to be supported for kernel 5.4 in the coming sdk release (in December).

[yipingwang]

Jumbo frame for felix is going to be supported for kernel 5.4 in the coming sdk release (in December).

This feature is already supported upstream for newer kernels.

[hittzt]

Hi Yiping,

You mean the patches for jumbo frame support on felix switch have been merged into

mainline kernel?

The latest stable kernel 5.9.x or under developing 5.10 or other kernel version and it is both implemented on dsa and switchdev?

Thanks,

Zhantao

[hittzt]

I find the related patches are merged in kernel 5.7.x branch, and seems they are only for dsa/ocelot.

That is to say in future, the ocelot/felix switch only supported by dsa?

Thanks,

Zhantao 

[yipingwang]

After 2020, Kernel 4.19 is not supported any more, so ocelot/felix switch is only supported by dsa.

[hittzt]

Hi Yiping,

Thanks for your reply.

As you stated, the future LSDK will release based on 5.x kernels.

And I would like to know, the TSN feature  would not be affacted by a MTU larger or smaller than 1518,

right?

Thanks,

Zhantao

[yipingwang]

According to the LSDK documentation:

“The upper limit for jumbo frame sizes is computed based on the maximum MTU supported by the DSA master interface (an ENETC Ethernet interface in this case). Namely, the maximum allowed Felix DSA port MTU is equal to the maximum MTU of the DSA master Ethernet interface (ENETC) minus the injection header overhead (32 bytes with the long prefix).”

And id you go to the ENETC driver chapter, sub-chapter “Scatter-gather and jumbo frame support”:

“The MTU of an ENETC netdevice interface (both PF and VF) can be increased to up to (9600 – L2 header size – FCS size) bytes.”

So it should be no greater than: 9600 – 14 (ETH_HLEN) – 4 (FCS) – 32 (TAG) = 9550 B

In practice I think a vlan tag also needs to be subtracted from this value, so it’s more likely 9546B.

Now the DSA header overhead is different in the latest upstream driver, so it’s 20B instead of 32B, so this maximum MTU value varies depending on felix driver implementation/ kernel version.

Bottom line is that it’s over 9500B (and less than 9600B), the approximate value given by the LSDK documentation should be good enough for practical purposes, I don’t think anyone really needs exact value at byte precision.  You can also test for the maximum allowed value in Linux, at runtime (try to set mtu and see what message you get).

[hittzt]

Hi Yiping,

OK, got it. Thanks for your reply!

BTW, the larger or smaller MTU value than 1518B will not affact the TSN functionality on felix switch, right?

Thanks,

Zhantao

[yipingwang]

Yes, you are right.

[hittzt]

OK, got it.

Thanks,

Zhantao

[hittzt]

Hi Yiping,

Thanks for your reply.

And I wander the MTU range of felix switch on LS1028ARDB,

I found in the current newest stable kernel branch v5.9.x, the max mtu of felix switch port is:

65535 - ETH_HLEN(14) - ETH_FCS_LEN(4) - OCELOT_TAG_LEN(16)

But in reallity, the switch ports max mtu should smaller than the above value.

Would you please help to check the precise value of max mtu?

Thanks,

Zhantao