ESP32 Ethernet does not always work

[cbrake]

Hi, We are using an Olimex ESP32-POE module and occasionally, the Ethernet will not work. If configured for DHCP, will not get an address, and with static IP, network operations do not work.

Observations:

• Z-MR boots, claims L4 network connectivity (NET_EVENT_L4_CONNECTED), but network does not work
• Rebooting usually fixes it
• Tried 2 olimex modules
• Two different switches (of the same module)
• Tried direct connection with crossover cable to a Macbook
• We were using Zephyr pre-release < 4.2.0. Upgrading to 4.2.0 did not make a difference.

Documentation for the board: https://www.olimex.com/Products/IoT/ESP32/ESP32-POE-ISO/open-source-hardware

The Phy clock is driven by GPIO0 on the ESP32 (I've read in the documentation that may be a concern).

Appreciate any pointers or ideas if anyone else has seen anything like this. We'll try to update this as we collect more data.

See this discussion for information on previous efforts with this device:

#89419

[sylvioalves]

@cbrake, I can test current release using a esp_ethernet_kit, where PHY is not driven by GPIO. AT least that will give us a pointer whether the issue is about clocking. But before I do so, have you done it? Also, how can I reproduce the issue? Restarting the board multiple times and wait for DHCP?

[jsnapoli1]

Hi @sylvioalves , @cbrake is away this week but I am working on this with him. We have not tried switching our board configuration due to being close to production. Was hoping it was software first. Here is a copy/paste of another issue I made, but have since closed and redirected here. Unfortunately, there is no easy way to reproduce this issue without just continuously pressing reboot.

Here is a link to the other issue I made: #96024 (comment)

[sylvioalves]

@jsnapoli1 Is samples/boards/espressif/ethernet enough to reproduce the issue? I am keen to say that if Rebooting usually fixes it, there may be some missing/needed code restart/re-config needed.

[jsnapoli1]

Here is what the sample gives:

uart:~$ net iface
Default interface: 1


Interface eth0 (0x3ffb5f48) (Ethernet) [1]
===================================
Interface is down.
[00:00:31.216,000] <inf> net_config: Running dhcpv4 client...
[00:00:31.216,000] <err> net_config: Timeout while waiting network interface
[00:00:31.216,000] <err> net_config: Network initialization failed (-115)
uart:~$ 

[sylvioalves]

Let me ask you something else: Are you using APLL in any other peripheral like I2S?

[jsnapoli1]

I do not believe we are using APLL in any other peripherals

[sylvioalves]

Can you provide the whole logging output that includes the bootloader as well?

[jsnapoli1]

I believe this is what you are asking for:

[12:11:42.709] Connected to /dev/tty.usbserial-130
ets Jul 29 2019 12:21:46

rst:0x1 (POWERON_RESET),boot:0x1a (SPI_FAST_FLASH_BOOT)
configsip: 0, SPIWP:0xee
clk_drv:0x00,q_drv:0x00,d_drv:0x00,cs0_drv:0x00,hd_drv:0x00,wp_drv:0x00
mode:DIO, clock div:2
load:0x3ffb0000,len:24932
load:0x40080000,len:50160
1150 mmu set 00010000, pos 00010000
entry 0x40087678
I (45) soc_init: ESP Simple boot
I (46) soc_init: compile time Sep 15 2025 11:57:22
W (46) soc_init: Unicore bootloader
I (46) soc_init: chip revision: v3.0
I (49) flash_init: SPI Speed      : 40MHz
I (53) flash_init: SPI Mode       : DIO
I (56) flash_init: SPI Flash Size : 4MB
I (60) boot: DRAM: lma 0x00001020 vma 0x3ffb0000 len 0x6164   (24932)
I (66) boot: IRAM: lma 0x0000718c vma 0x40080000 len 0xc3f0   (50160)
I (72) boot: IRAM: lma 0x00013588 vma 0x00000000 len 0xca70   (51824)
I (78) boot: IMAP: lma 0x00020000 vma 0x400d0000 len 0x1d068  (118888)
I (85) boot: IRAM: lma 0x0003d070 vma 0x00000000 len 0x2f88   (12168)
I (91) boot: DMAP: lma 0x00040000 vma 0x3f400000 len 0x82c4   (33476)
I (97) boot: Image with 6 segments
I (100) boot: IROM segment: paddr=00020000h, vaddr=400d0000h, size=1D068h (118888) map
I (108) boot: DROM segment: paddr=00040000h, vaddr=3f400000h, size=082C4h ( 33476) map
I (127) boot: libc heap size 127 kB.
I (127) spi_flash: detected chip: generic
I (127) spi_flash: flash io: dio
I (127) quad_psram: This chip is ESP32-D0WD
I (132) esp_psram: Found 8MB PSRAM device
I (135) esp_psram: Speed: 40MHz
I (138) esp_psram: PSRAM initialized, cache is in normal (1-core) mode.
W (145) esp_psram: Virtual address not enough for PSRAM, map as much as we can. 4MB is mapped
I (1034) esp_psram: SPI SRAM memory test OK

[00:00:01.110,000] <inf> phy_mii: PHY (1) ID 0
*** Booting Zephyr OS build v4.2.0-111168-g8c4d8218de20 ***
[00:00:01.215,000] <inf> net_config: Initializing network
[00:00:01.215,000] <inf> net_config: Waiting interface 1 (0x3ffb5f48) to be up...

Let me know if you need anything else

[jsnapoli1]

Hi @sylvioalves - Any updates/progress/discoveries?

[sylvioalves]

@jsnapoli1 I modified my esp32_ethernet_kit to use GPIO0 as RMII clock. However, the board uses IP101GRI instead of the LAN7810 of yours.
It works good and I can't reproduce the error using when running this code: #96190

1. Can you confirm that you get the PHY ID as in here? If not, something is wrong indeed.
   [00:00:01.126,000] <inf> phy_mii: PHY (1) ID 2430C54

2. Also, make sure you use the proper reg address in here:

	phy: ethernet-phy@1 {
		compatible = "ethernet-phy";
		reg = <1>;
	};

Let me know how it goes, thanks.

[jsnapoli1]

Is that just a matter of changing the reg?

Also, here is a potential issue. Looking at my dtsi:

&mdio {
pinctrl-0 = <&mdio_default>;
pinctrl-names = "default";
ref-clk-output-gpios = <&gpio0 0 0>;

phy: ethernet-phy@1 {
	compatible = "ethernet-phy";
	status = "disabled";
	reg = <1>;
};

};

&eth {
phy-handle = <&phy>;
phy-connection-type = "rmii";
ref-clk-output-gpios = <&gpio0 0 0>;
};

[jsnapoli1]

Appears to me that we have seemingly disabled phy ethernet and created eth instead. @cbrake is out of office, so not sure why we would have done that, but I may look into changing that.

[sylvioalves]

Please, make the change below and provide me that log output again:

phy: ethernet-phy@1 {
		compatible = "ethernet-phy";
		reg = <1>;
	};

Make sure you have this in place as well:
#96190

[cbrake]

@sylvioalves still catching up -- thanks for #96190 -- we had a hack in our application code to work around this. Will pull this in and see if we can remove the hack.

[sylvioalves]

Ok, let me know how I could reproduce the error. I have a esp32-ethernet-kit using gpio0 as clock basis. Not the same board as you do but similar approach.

[jsnapoli1]

Following up on this,

After updating the above, I still see the same issues with both DHCP and Static IP. I am about to test upgrading to zephyr main and also implementing these changes to see if it has to do with other pieces of Zephyr 4.2.0, however, we still see false L4 links or inabilities to sync to DHCP.

[jsnapoli1]

Updating to main does not fix further issues.

[sylvioalves]

Would be good if I can reproduce the issue but I can't. How many times you had to reset the board to get such error? Which sample or configuration you have in there?