Wifi regression on esp32-c6

[kaspar030]

Bug description

We're seeing a wifi regression on esp32-c6, where the device seems to delay or drop packets it should receive.

It shows i.e., with weird ping times:

❯ ping 192.168.1.224                 
PING 192.168.1.224 (192.168.1.224) 56(84) bytes of data.               
64 bytes from 192.168.1.224: icmp_seq=4 ttl=64 time=3818 ms       
64 bytes from 192.168.1.224: icmp_seq=5 ttl=64 time=2805 ms
64 bytes from 192.168.1.224: icmp_seq=6 ttl=64 time=1791 ms
64 bytes from 192.168.1.224: icmp_seq=8 ttl=64 time=3875 ms
64 bytes from 192.168.1.224: icmp_seq=9 ttl=64 time=2851 ms
64 bytes from 192.168.1.224: icmp_seq=10 ttl=64 time=1838 ms
64 bytes from 192.168.1.224: icmp_seq=11 ttl=64 time=841 ms

We've seen this starting with 0.22.0 already. 0.21.1 works fine:

4 bytes from 192.168.1.224: icmp_seq=100 ttl=64 time=2.45 ms
64 bytes from 192.168.1.224: icmp_seq=101 ttl=64 time=10.8 ms
64 bytes from 192.168.1.224: icmp_seq=102 ttl=64 time=9.07 ms
64 bytes from 192.168.1.224: icmp_seq=103 ttl=64 time=8.51 ms
64 bytes from 192.168.1.224: icmp_seq=104 ttl=64 time=16.2 ms
64 bytes from 192.168.1.224: icmp_seq=105 ttl=64 time=6.67 ms
64 bytes from 192.168.1.224: icmp_seq=106 ttl=64 time=4.16 ms
64 bytes from 192.168.1.224: icmp_seq=107 ttl=64 time=14.1 ms
64 bytes from 192.168.1.224: icmp_seq=108 ttl=64 time=1.31 ms
64 bytes from 192.168.1.224: icmp_seq=109 ttl=64 time=18.3 ms
64 bytes from 192.168.1.224: icmp_seq=110 ttl=64 time=9.64 ms
64 bytes from 192.168.1.224: icmp_seq=111 ttl=64 time=8.80 ms

I tried bisecting, which leads to this commit:

fbc57542a8f4b71e30f0dcea4045c508ce753139 is the first bad commit                                                                                               
commit fbc57542a8f4b71e30f0dcea4045c508ce753139 (HEAD)                                                                                                         
Author: Dániel Buga <[email protected]>                                                                                                                       
Date:   Tue Nov 12 11:36:25 2024 +0100                                                                                                                         
                                                                                                                                                               
    Remove pins from `Io` (#2508)                                              
     
    * Split pins off of Io
     
    * Remove the GPIO peripheral
     
    * p.GPIO

(... but that doesn't make sense to me)

To Reproduce

The wifi_embassy_dhcp easily reproduces this. While it always manages to connect and get the nice crab drawing, pinging the node shows the weird timings:

SSID="foo" PASSWORD="bar" cargo xtask run-example examples esp32c6 wifi_embassy_dhcp

Expected behavior

Environment

• Target device: [e.g. ESP32-S3]

Chip type:         esp32c6 (revision v0.0)
Crystal frequency: 40 MHz
Flash size:        8MB
Features:          WiFi 6, BT 5
MAC address:       40:4c:ca:51:39:b4

(I have esp32, esp32-s3 and esp32-c3 that seem to work fine.)

• Crate name and version: [e.g. esp-hal 0.20.0]

see above

Let me know how I can help debugging this!

[bugadani]

Can you please see if wifi_controller.set_power_saving(PowerSaveMode::None) changes anything? I guess in #2446 we changed the default from None to Minimum, which may increase latency

[kaspar030]

Can you please see if wifi_controller.set_power_saving(PowerSaveMode::None) changes anything? I guess in #2446 we changed the default from None to Minimum, which may increase latency

That fixes the issue, thanks! 👍

Pings look like this again:

❯ ping 192.168.1.224
PING 192.168.1.224 (192.168.1.224) 56(84) bytes of data.
64 bytes from 192.168.1.224: icmp_seq=1 ttl=64 time=21.5 ms
64 bytes from 192.168.1.224: icmp_seq=2 ttl=64 time=20.0 ms
64 bytes from 192.168.1.224: icmp_seq=3 ttl=64 time=18.8 ms
64 bytes from 192.168.1.224: icmp_seq=4 ttl=64 time=6.91 ms

I'll not close this as maybe the default is suboptimal for the c6?