iOS 26 issue with getting IP address with NCM

[TheHogfather562]

Operating System

Others

Commit SHA

41a00f4

Board

multiple custom boards, esp32-s3 based and nrf52833 based

Firmware

examples/device/net_lwip_webserver

What happened ?

On the newest iOS version 26.3 and ever since the release of iOS 26, the success rate of obtaining a IP Address from the built in DHCP server is about 30%.

How to reproduce ?

Upload example to a nrf52833dk, plug into an iPad or iPhone running iOS 26 and the ip address won't show up most of the time; in settings->Ethernet->TinyUSB Device.

Debug Log as txt file (LOG/CFG_TUSB_DEBUG=2)

I can't currently produce logs as I'm on a custom board.

Screenshots

I think it might be that the iPhone requests an ip address before the usb device is fully ready and then never asks again. Similar to this potentially #3445.

I've got the issue on both a nrf52833 based board and an esp32-s3 based board.

However I don't see this problem at all on a RP2350 based board

I have checked existing issues, discussion and documentation

• I confirm I have checked existing issues, discussion and documentation.

[lonkenhout]

I cant tell you the exact issue, but the organisation I work for was using a nRF52840 with the NCM stack and it has also been suffering considerably.

Whether this is a TinyUSB issue or an issue on the Apple side I cant tell you. They do sometimes restrict implementations on their side. I had similar issues with the NCM stack inside Zephyr..

What I can tell you is the ECM stack seems to work well still. So if performance is no problem and you need a product quickly, temporarily switching to ECM may be a timely alternative.

[HiFiPhile]

AFAIK none of the mainteniers has iOS device to work with.

[lonkenhout]

If it helps, I have also seen the issue on MacOS and iPadOS. So its not restricted to only iOS.. I doubt I would be able to contribute to this project. But if no progress is made yet tomorrow, I'll bring a dev kit home and see if I can produce some relevant dumps (Wireshark USB, Mac logs from NCM driver, and logs from TinyUSB firmware).

[TheHogfather562]

Thanks for the suggestion @lonkenhout, we have already temporarily switched to the ECM driver on our nrf based product, but for the ESP based product I don't think they support that driver mode anymore.

espressif/esp-usb#72

I'm away currently but if there are any logs I can produce or general suggestions of things to try when I get back then I'm happy to investigate.

[HiFiPhile]

I'm wondering if #3438 helps

[ceedriic]

Hello @HiFiPhile,
I just tried to plug my board (stm32h7, NCM) on an iPad, and I see the same problem. My success rate is higher (~80%) but it's not 100%. There is no such problem with my Mac though. I will need to retest everything with and without #3438, on different OSes, now that my hardware is fixed. I hope to be able to do that this week.

[TheHogfather562]

Previous tests I've performed seem to have narrowed down the problems on my boards to only iOS 26 and above on both iPhones and iPads.

I've tested both my boards successfully on:

• Mac
• Linux
• Windows 10 and 11
• iOS 18

And have only had issues with iOS 26.

[TheHogfather562]

I'm wondering if #3438 helps

I will check if this solves my issue when I'm back on the 2nd

[lonkenhout]

Appreciate the activity guys. I made a couple logs/dumps using the unmodified master branch (except for TU_LOG_BUF(3, ...) debug statements in ncm_device.c causing compilation errors). And added some very simple print statements in dhserver.c.

(Un)successful connection may be a bit misleading, I mean the final result is not a network interface through which I can access the device. As a general description: it seems as though after the NCM interface is detected, it is not completely finalized. Afterwards you never see any IP traffic on either side (which I guess makes sense if NCM hasnt fully setup). I haven't looked deep into the code, so I dont know what state transitions are required to go from NCM to DHCP.

This was all with an nRF52840 devkit btw.

TinyUSB RTT output (LOG=3)
successful_connection1.log
unsuccessful_connection1_rtt.log
successful_connection_difference_rtt.log
multiple_connections_rtt.log

MacBook perspective
AppleUSBLogs.log

USB captures
Unfortunately I cant drop in PCAPNG files here, so I'll briefly show the (what I think are) the relevant parts as images:

Successful connection:

Unsuccessful connection:

I have another longer capture with multiple connections/disconnections, if you want to look at it yourselves let me know where to drop these.

[s09289728096]

It happends on my side. I'm using STH32U5 + lwip + tinyUSB as NCM device, and test:

• Windows 10/11
• Android 15/16
• IOS 18/26
• Linux 6.14

Only IOS 26 got this issue, and it can't be reproduced stably (~20% success rate)
BTW, IOS 18 is observed slower than other OS on initialization, it takes about 7~10 seconds to start the first ARP connection.

I can test more if you guys need.

[HiFiPhile]

What happens if you use tud_network_link_state to send additional interface up event to host ?

[TheHogfather562]

I'm wondering if #3438 helps

This doesn't appear to have made any difference

[TheHogfather562]

use tud_network_link_state to send additional interface up event to host ?

This also appears to have not made any difference, just to check this is what you meant:

during net setup we create a callback that sets the tud_network_link_state?

static void usb_net_link_up_cb(void *ctx)
{
    tud_network_link_state(0, true);
}

// ... in setup

tinyusb_net_config_t _net_config = {
    .on_recv_callback = netif_recv_callback,
    .on_init_callback = usb_net_link_up_cb,
};

[HiFiPhile]

use tud_network_link_state to send additional interface up event to host ?

This also appears to have not made any difference, just to check this is what you meant:

Try to toggle link state manually like in net_lwip_webserver example:

tinyusb/examples/device/net_lwip_webserver/src/main.c

Line 221 in 13e0b0c

static void handle_link_state_switch(void) {