Lab notes
Board test reports
On this page
Each run below records the exact firmware and kernel revisions, what was tested and what happened. All runs used an Orange Pi Zero over SDIO, with an Intel adapter (wlp4s0) as the peer. The first two are connectivity checks only.
2026-11-19: driver series for review
| Item | Revision |
|---|---|
| Firmware | 58f32d3; bootloader SHA-256 05ba3714e15477ed72b38ec31ad0792f731cd74728b032cb72ce5229c53f0b99 (as in the first run), runtime bd43771e0bcfea9c83e29f3e8f8776edccf46f6839e3c3c32bfdb5a6cfa80d19, sdd_xr819.bin 84d3fb3ca8e5d25a0c113a5063bccbeb5b53da230a0afa236b5b625f37db5161 |
| Kernel | Ten-patch series on Torvalds master 4405c18c95d2, tip bb557dc24a46, reported as 7.4.0-rc3 |
| Peer | The Intel adapter on channel 6; WPA2-PSK with CCMP throughout |
Throughput is one TCP stream from a Python sender, 15 to 60 seconds per run.
Station. Every association completed in 1.1 to 1.3 seconds, one in 6.9. Throughput was 17 to 26 Mbit/s from the peer to the board and 15 to 24 Mbit/s back.
Soak. For 10 minutes the peer and the board took turns sending for a minute each: 728 MB to the board, 640 MB from it. A concurrent ping got 1,182 of 1,214 replies, with round trips up to 2.1 seconds while the link was saturated. The link stayed associated and the kernel logged no errors.
Rebind. Unbind and bind while associated, then a new association: 4 cycles with the SDIO interrupt and 5 with a GPIO interrupt. All associated; 20 pings right after lost at most one, except once 6 of 20, which recovered to 47 of 50 a minute later. Before 655e896fe871, every unbind of an associated board logged CMD req (0x0009) stuck in firmware, killing BH: the driver freed its interrupt before mac80211 removed the interface, so the confirmations of the commands sent during removal were lost. Associations that failed after a rebind early in this run came from the test script, which left an extra wpa_supplicant on the interface; the retry needed in the previous run may have had the same cause.
Access point. hostapd started without the failed START confirmation of the previous run: 58f32d3 accepts the listen-mode START that the driver sends first. The peer completed the four-way handshake. Pings with a 1,200-byte payload: 46 of 50 to the board and 50 of 50 back in one run, 30 of 30 and 28 of 30 in another. Throughput was 11.1 Mbit/s from the board and 9.5 Mbit/s to it.
GPIO interrupt. A device-tree overlay giving the XR819 node compatible = "xradio,xr819" and an interrupt on PG10 made the driver use the GPIO interrupt instead of the SDIO one. Throughput was 25 Mbit/s to the board and 16 Mbit/s from it.
Earlier runs used builds that lacked the listen-mode change or carried five behaviour-neutral lint rewrites, which change the image hashes. The images listed above were retested afterwards for two rebinds, throughput in both directions and an access-point start.
2026-11-19: rebind and WPA2 retest
| Item | Revision |
|---|---|
| Firmware | f9e52aa; runtime SHA-256 8069466211146778955ecf17545dcf45259678b1726779b5eaa40ac75790c3ed, bootloader as in the first run |
| Kernel | Six-patch series on Torvalds master 4405c18c95d2, tip e4bcf24ebf56, then 476daf13af49 (not published; the same reset is part of 2bd278da55ca on the review branch) |
WPA2. With e4bcf24ebf56 the board exchanged WPA2-CCMP traffic with the peer as station and as access point, and unbound without an oops.
Rebind. Bind then failed with -EINVAL: the XR819 was still in queue mode, so the driver could not start another firmware download. 476daf13af49 resets only the XR819 SDIO card on probe. With it, three consecutive unbind/bind cycles passed, and a fourth passed after rebooting into the revised module. Both firmware images read back unchanged.
After rebind. The first station association did not complete within 35 seconds; the retry did. Pings with a 1,200-byte payload: 19 of 20 from board to peer, 20 of 20 back. As an access point, 10 of 10 each way. The kernel still logged a failed WSM confirmation for request 0x0017 (START).
2026-11-19: first boot with the open firmware
| Item | Revision |
|---|---|
| Firmware | f53c5d4; bootloader SHA-256 05ba3714e15477ed72b38ec31ad0792f731cd74728b032cb72ce5229c53f0b99, runtime 2c0226d2bca44e12fbcdc27fa9e57c70aaaba313ecd1c6d6924b2db1f578c21a |
| Kernel | 8b3b8d68f31d on Torvalds master 4405c18c95d2, built with SDIO and SPI |
| Checks | 267 Rust tests, 44 Python tests, ARM stack check and both image builds passed; a fresh checkout produced identical images |
The probe reported XR819 open Rust native.
Station. WPA2-PSK/CCMP completed against the peer. Each end received 9 of 10 pings.
Access point. hostapd completed the four-way handshake. The board lost its test IPv4 address on the interface mode change; after reassigning it, both ends received 20 of 20 pings with a 1,200-byte payload.
Unbind. SDIO unbind faulted in cw1200_debug_release: unregistering the wiphy had already removed its debugfs subtree. Kernel e4bcf24ebf56 drops the second removal; the retest above uses it.