zLightz

The files of my drivers version: Android X Armbian ssv6051-android-dtb.zip

Hi, I have two rk322x TV boxes (RK3229 with SSV6051 SDIO wifi) where wlan0 is created but never comes up on Armbian. Both boxes had working wifi on their original Android firmware. I instrumented the driver to trace this and would appreciate input from anyone familiar with ssv6051. SYSTEM Board rk322x-box (Generic RK322x TV Box, RK3229) Armbian 26.11.0-trunk.25 trixie, BRANCH=current Kernel 6.18.46-current-rockchip Wifi chip SDIO 0x3030:0x3030, "SouthSiliconValleyInc. / Cabrio" Driver ssv6051 (in-tree, from Armbian's wifi-driver-ssv6051.patch) Firmware /lib/firmware/ssv6051-sw.bin and ssv6051-wifi.cfg both present SYMPTOM The SDIO card enumerates and wlan0 appears, but bringing it up fails: # ip link set wlan0 up RTNETLINK answers: Operation not permitted dmesg: ssv6051 mmc1:0001:1: device id: 3030:3030 SSV WLAN driver ssv6200: Chip identity from efuse: 00000000 SSV WLAN driver ssv6200: Using SSV6051Q setting SSV WLAN driver ssv6200: Unexpected register value WARNING: CPU: 0 PID: 282 at drivers/net/wireless/ssv6051/smac/init.c:1044 ssv6xxx_dev_probe+0x1540/0x1870 [ssv6051] ... SSV WLAN driver ssv6200: chip id: RSV6200A0-201311, tag: 2014012420010960 SSV WLAN driver ssv6200: Failed to allocate packet buffer of 904 bytes in 0 type. SSV WLAN driver ssv6200: opps allocate pbuf error WARNING: CPU: 2 PID: 2652 at drivers/net/wireless/ssv6051/smac/init.c:608 ssv6xxx_init_mac+0xc28/0xc30 [ssv6051] SSV WLAN driver ssv6200: Failed to initialize mac, ret=1 The two WARNINGs are the two places in the driver that verify a register write by reading it back: init.c:1044 (ADR_TRX_DUMMY_REGISTER) and init.c:608 (pbuf alloc). WHAT I DID I rebuilt ssv6051 out-of-tree against 6.18.46 headers with logging added to ssv6xxx_sdio_write_reg() and ssv6xxx_pbuf_alloc(), so every register write shows its address, value, the ioport it went through, and the return code. FINDING 1 - the SDIO transport reports no problem at all - ioport_reg = 0x10020, read correctly via cmd52 in ssv6xxx_sdio_read_parameter() - all 303 register writes during probe return ret=0 from sdio_memcpy_toio() - /sys/kernel/debug/mmc1/err_stats is all zeros, including during the failure: no command timeout, no command CRC, no data timeout, no data CRC, no controller timeout, no unexpected IRQ - the driver's own counter reports "sdio cmd52 fail: 0" - reads are stable and repeatable: the ROM ID reads back "RSV6200A0-201311" identically on every attempt FINDING 2 - but no register write ever persists, in any block Test run from inside the driver (not through debugfs), writing 0x12345678 to one register in each hardware block, immediately after the MAC reset in ssv6xxx_init_mac(): address block original wrote read back result 0xce010040 RF 0x00550800 0x12345678 0x00550800 not written 0xce000050 PHY 0x00000000 0x12345678 0x00000000 not written 0xc0001d10 SYS 0x00000000 0x12345678 0x00000000 not written 0xc6002508 MAC 0x00000000 0x12345678 0x00000000 not written 0xc1000034 HCI 0x00000008 0x12345678 0x00000008 not written 0xcd010008 ID_MNG 0x00000000 0x12345678 0x00000000 not written 0xcf000004 MMU 0x00000000 0x12345678 0x00000000 not written Same picture for the driver's real init_mac() writes. Every one of these registers still holds the reset default defined in ssv6200_reg.h: register address driver wrote chip contains reset default ADR_CONTROL 0xc1000000 0x12000006 0x02700008 0x02700008 TRX_ID_THRESHOLD 0xcd010020 0x01ee3c13 0x01ee3c3c 0x01ee3c3c ID_LEN_THRESHOLD1 0xcd010038 0x000e6731 0x000f0641 0x000f0641 TXQ4_MTX_Q_AIFSN 0xc6002504 0xffff2101 0x0000a502 0x0000a502 MMU_CTRL 0xcf000000 0x00ff6042 0x00006042 - So the chip answers reads perfectly and acknowledges writes at the SDIO layer, but the values never land. FINDING 3 - this is what makes pbuf allocation fail ssv6xxx_pbuf_alloc() writes the requested size to ADR_WR_ALC and reads it back, retrying up to 20 times: SSVDBG wr addr=0xcd010000 val=0x00000388 ioport_reg=0x10020 ret=0 (x20) SSVDBG pbuf size=904 type=0 alc=0x00000000 mmu=0x00006042 r04=0x00000213 ctl=0x02700008 alc reads back 0 on all 20 attempts, so the driver hits "opps allocate pbuf error" and takes the goto exit. Because that exit happens before SMAC_LOAD_FW(), the firmware is never downloaded: "attempt to load firmware" never appears in dmesg, and ssv_cmd reports "Firmware image version: 0". Worth noting: the MAC soft reset immediately before this (the ADR_BRG_SW_RST poll loop) does not report "MAC reset fail", so that particular write-then-poll appears to succeed. WHAT I HAVE ALREADY RULED OUT Kernel version Tested 6.6.16-current-rockchip (Feb 2024) on the same box. Byte-identical failure: same warnings, same addresses, same messages. So this is not a regression introduced between 6.6 and 6.18. Device tree I extracted the original Android DTB from the resource partition of a box that was never reflashed, decompiled it, and compared against Armbian's rk322x-box.dtb. Each difference was tested with a real reboot: - GPIO polarity. Android declares WIFI,poweren_gpio = <&gpio2 26 GPIO_ACTIVE_HIGH> while Armbian uses reset-gpios = <&gpio2 26 GPIO_ACTIVE_LOW> Same pin, opposite polarity. Rebuilding the DTB with ACTIVE_HIGH makes the card stop enumerating entirely ("mmc1: Failed to initialize a non-removable card"), so Armbian's polarity is the correct one. - SDIO pin drive strength. Android uses 4mA (pull-none on clk, pull-up on cmd/data); Armbian uses 2mA (pull-down on clk). Rebuilt the DTB with 4mA to match Android. No change. - SDIO clock. Forced max-frequency to 12.5 MHz and added post-power-on-delay-ms = 200 to the pwrseq node. No change. - io-domains. Both already map vccio2 to vccio_1v8. Identical. Power sequence Rockchip's own ssv6051-generic-wlan.c has the rockchip_wifi_power() and rockchip_wifi_set_carddetect() calls commented out inside ssv_wifi_power(); the function only does msleep(150). Armbian's generic-wlan does msleep(120). Functionally equivalent, so there is no missing power-on step. Driver build configuration CONFIG_PLATFORM_SDIO_OUTPUT_TIMING=3, CONFIG_PLATFORM_SDIO_BLOCK_SIZE=128 and CONFIG_SSV_CABRIO_E are identical between the Android build (readable in the debug_info of the Android .ko) and Armbian's platform-config.mak. ret_if_not_ready() Fires 0 times during probe and during the up. It only appears on module unload, so it is not blocking any of the writes. Driver-level config options Tried force_triger_reset, force_chip_identity, disabling HT/AMPDU/AP to reduce buffer demand, and switching volt_regulator between LDO and DCDC. No change. Hardware Two independent boxes behave identically, and both had working wifi under their original Android firmware, so I do not believe this is a dead chip. QUESTION Has anyone got SSV6051 working on rk322x with a mainline kernel? Reads work perfectly, cmd52/cmd53 report no errors, err_stats stays at zero, and writes are acknowledged at the SDIO layer, yet nothing lands in the chip. My remaining suspicion is something in the SDIO write path itself - sdio_memcpy_toio() to ioport_reg, or block-mode versus byte-mode transfer - that differs from what the 3.10 Android driver did. I could not confirm that without a logic analyzer on the bus. I am happy to run any test you suggest. I can also provide the full instrumented dmesg, the extracted Android DTB, and the Android ssv6051.ko (vermagic 3.10.104) if any of that is useful for comparison. Thanks android_1536.dtb rfkill-wlan.c sec0.bin