LayerscapeのトピックRe: LS1046A DPAA Jumbo Frame using upstream ethernet driver

LS1046A DPAA Jumbo Frame using upstream ethernet driver

fredericlefranc — Fri, 04 Feb 2022 16:35:03 GMT

Hello,

I'm using the linux kernel from LSDK_21_08 (kernel 5.10). I choose to use DPAA upstream driver instead of the private one.

Everything is working properly except the fact that I can't change MTU to something bigger than the default 1500.

I've tried to use the max-frame-size property in the device tree for each ethernet, but it doesn't change anything.
It seems that to achieve a higher MTU value, I need to configure frame size in FMAN, but with the upstream driver, I have no way to do that.

Do you guys have a solution to allow jumbo frames on LS1046A using upstream driver?

Side question, what are the advantages/disadvantages of using private driver instead of mainline one?

Regards,
Fred.

Re: LS1046A DPAA Jumbo Frame using upstream ethernet driver

yipingwang — Mon, 07 Feb 2022 07:32:45 GMT

To enable Jumbo frame, the user is required to set the L2 maximum frame size to 9600, otherwise the configuration is not valid.

This can be achieved by either setting fsl_fm_max_frm=9600 in the bootargs, or configuring CONFIG_FSL_FM_MAX_FRAME_SIZE from menuconfig.

Re: LS1046A DPAA Jumbo Frame using upstream ethernet driver

fredericlefranc — Mon, 07 Feb 2022 10:08:47 GMT

Hi,

Yes this procedure is the one that should work when using private driver. I choose to use mainline driver, so these are either not available (CONFIG_FSL_FM_MAX_FRAME_SIZE), or doesn't work (fsl_fm_max_frm=9600 in kernel's command line).

I see that I could create a new devicetree value (like max_l2_size) and modify fman driver to use it, but I was wondering why jumbo frame was not available as is in mainline driver.

Regards,
Frederic.

Re: LS1046A DPAA Jumbo Frame using upstream ethernet driver

GregUngerer1 — Mon, 01 Aug 2022 05:55:45 GMT

I am experiencing this same problem with the DPAA/FMAN drivers in the mainline kernel. I am using linux-5.18 kernel. So the problem still exits as far as I can tell.

Federic, did you ever get a solution to fix this on mainline kernels?