Qu Wenruo [off-list ref] writes:

On 2021/9/17 15:18, Punit Agrawal wrote:

quoted

Hi Qu,
Qu Wenruo [off-list ref] writes:

quoted

On 2021/8/30 22:10, Michael Riesch wrote:

quoted

Hi Punit,
On 8/30/21 3:49 PM, Punit Agrawal wrote:

quoted

Hi Michael,

Michael Riesch [off-list ref] writes:

quoted

Hi ChenYu,

On 8/29/21 7:48 PM, Chen-Yu Tsai wrote:

quoted

Hi,

On Mon, Aug 23, 2021 at 10:39 PM Michael Riesch
[off-list ref] wrote:

quoted

This reverts commit 2c896fb02e7f65299646f295a007bda043e0f382
"net: stmmac: dwmac-rk: add pd_gmac support for rk3399" and fixes
unbalanced pm_runtime_enable warnings.

In the commit to be reverted, support for power management was
introduced to the Rockchip glue code. Later, power management support
was introduced to the stmmac core code, resulting in multiple
invocations of pm_runtime_{enable,disable,get_sync,put_sync}.

The multiple invocations happen in rk_gmac_powerup and
stmmac_{dvr_probe, resume} as well as in rk_gmac_powerdown and
stmmac_{dvr_remove, suspend}, respectively, which are always called
in conjunction.

Signed-off-by: Michael Riesch <redacted>

I just found that Ethernet stopped working on my RK3399 devices,
and I bisected it down to this patch.

Oh dear. First patch in a kernel release for a while and I already break
things.

I am seeing the same failure symptoms reported by ChenYu on my RockPro64
with v5.14. Reverting the revert i.e., 2d26f6e39afb ("net: stmmac:
dwmac-rk: fix unbalanced pm_runtime_enable warnings") brings back the
network.

quoted

Cc: Sasha as this patch has just been applied to 5.13-stable.

quoted

The symptom I see is no DHCP responses, either because the request
isn't getting sent over the wire, or the response isn't getting
received. The PHY seems to be working correctly.

Unfortunately I don't have any RK3399 hardware. Is this a custom
board/special hardware or something that is readily available in the
shops? Maybe this is a good reason to buy a RK3399 based single-board
computer :-)

Not sure about the other RK3399 boards but RockPro64 is easily
available.

I was thinking to get one of those anyway ;-)

quoted

quoted

I am working on the RK3568 EVB1 and have not encountered faulty
behavior. DHCP works fine and I can boot via NFS. Therefore, not sure
whether I can be much of help in this matter, but in case you want to
discuss this further please do not hesitate to contact me off-list.

I tried to look for the differences between RK3568 and RK3399 but the
upstream device tree doesn't seem to carry a gmac node in the device
tree for EK3568 EVB1. Do you have a pointer for the dts you're using?

The gmac nodes have been added recently and should enter
5.15-rc1. Until
then, you can check out the dts from linux-rockchip/for-next [0].

Do you have the upstream commit?

As I compiled v5.15-rc1 and still can't get the ethernet work.

Not sure if it's my Uboot->systemd-boot->customer kernel setup not
passing the device tree correctly or something else...

For the RK3568 device tree changes, I think the pull request got
delayed
to the next cycle. So likely to land in v5.16.
In case you're after ethernet on RK3399, there's no solution
yet. Reverting 2d26f6e39afb ("net: stmmac: dwmac-rk: fix unbalanced
pm_runtime_enable warnings") gets you there in the meanwhile.

Thanks, currently I have seen other distros like ManjaroARM is already
reverting that commit.

But even with that commit reverted, I still get some other strange
network behavior.

The most weird one is distcc, when the RK3399 board is the client and
x86_64 desktop acts as a volunteer, after compiling hundreds of files, 
it suddenly no longer work.

I haven't seen something like this - but then I am not a heavy user of
the network on the board. 

Is it just the network that dies or the whole system freezes? Sometimes
I've seen the board lock up if it's under heavy load.

All work can no longer be distributed to the same volunteer.


But on RPI CM4 board, the same kernel (both upstream 5.14.2, even the
binary is the same), the same distro (Manjaro ARM), the same distcc 
setup, the setup works flawless.


Not sure if this is related, but it looks like a network related
problem, and considering both boards are using the same kernel, just 
different ethernet driver, I guess there is something more problematic
here in recent RK3399 code.

If you are only seeing the problem with recent kernels, maybe a bisect
might help narrow things down.

[...]