Skip to content

Hardware notes ​

Facts about the RG35XX SP that cost time to find. Feature notes has them in full.

The device ​

PartDetails
SoCAllwinner H700: 4× Cortex-A53 up to 1512 MHz, Mali-G31 GPU
RAM1 GB; LPDDR3 at 1.2 V on the test unit, LPDDR4 on others
Display640x480 at 59.94 Hz with the NV3052C panel, 59.52 Hz with the second panel (card1; Panfrost is card0)
PMICX-Powers AXP717
Wi-Fi / BluetoothRealtek RTL8821CS: Wi-Fi on SDIO (mmc1, rtw88), Bluetooth on UART1 (H5)
StorageTF1 (mmc0, boot) and TF2 (mmc2)
USBOne USB-C port, device mode only (MUSB)
OtherMini-HDMI, lid switch, vibration motor, speaker and headphone jack

Memory ​

Don't trust the model name. The test unit has LPDDR3 at 1.2 V, while mainline U-Boot and the device tree assume LPDDR4 at 1.1 V. Read a working boot0 (8 KiB into a stock or muOS card) instead:

OffsetValue
0x38DRAM clock
0x3cType: 7 LPDDR3, 8 LPDDR4
0x40 / 0x44 / 0x48DX_ODT / DX_DRI / CA_DRI
0x4cODT_EN
0xa0 / 0xa4 / 0xa8 / 0xacTPR6 / TPR10 / TPR11 / TPR12

The SP comes with LPDDR4 or LPDDR3, and the SPL finds out which (U-Boot patch 0001): it tries LPDDR4 at 1.1V (CONFIG_AXP_DCDC3_VOLT and the CONFIG_DRAM_SUNXI_* values in board/rg35xxsp/uboot.fragment), and if that fails, LPDDR3 at 1.2V (the CONFIG_DRAM_SUNXI_LPDDR3_* values). Kernel patch 0006 gives reg_dcdc3 a range of 1.1V to 1.2V, so Linux keeps the voltage the SPL set. If neither type works, the SPL stops, which shows as a 1 Hz green blink.

USB ​

  • EHCI0 and OHCI0 share PHY 0 with the MUSB device controller and switch the port to host mode, so patch 0007 disables them. Host mode needs that reverted plus a dual-role setup like ROCKNIX's 0151 and 0152, which also drive 5 V on the port: review them before use while connected to a PC.
  • The controller has 4 IN and 4 OUT endpoints. Serial plus Ethernet use 4 IN and 2 OUT, so another function needs one dropped: USB STORAGE swaps serial for the drive. /usr/sbin/usb-gadget sets the gadget up through configfs.
  • MUSB flushes RX packets that arrive before a request is queued (mainline commit 3fc137386c46), so mass storage lost commands. Patch 0014 reads them instead.

Power ​

  • Plugging in USB power starts the device (VBUS rising, set in the AXP717's efuse). REG 0x20 says what started it: bit 0 the power button, bit 2 USB. Test shutdown with the cable unplugged.
  • Kernel power-off writes the wrong AXP717 register, but falls through to PSCI, and TF-A turns the device off.
  • The charger runs at 1024 mA; CHARGE TO writes the battery's voltage_max (4.0, 4.1 or 4.2 V). The driver's current_now has an unknown offset, so it isn't used.

Inputs and outputs ​

  • X and Y are swapped on this unit against the shared device tree; patch 0008 swaps them.
  • The vibration motor is on PH2, PWM channel 2 (patch 0015). It adds an input device, so /dev/input/event* numbers move: find devices by name.
  • TF2 is mmc2 (patch 0018), with a 3.3 V switch on PE4 that must stay on. /usr/sbin/cards finds each slot's card by its controller, not its number.
  • The SP has one of two panels, on the same wires. Kernel patch 0020 resets the panel when the driver starts and reads its ID: an NV3052C reports 30 52 01 (its setup then replaces that), and any other answer gets the second panel's setup and timings (720 x 560 pixels per frame, 59.52 Hz). dmesg and bootlog.txt say which: panel: NV3052C, ID 305201. panel_newvision_nv3052c.panel=nv3052c or =v2 on the kernel's command line overrides it.
  • HDMI mirrors fb0 at 640x480 via video=HDMI-A-1:640x480@60. Its TMDS clock is 24.75 MHz against the 25.175 MHz mode: the first suspect if a TV shows nothing.

Audio ​

  • amixer percentages are raw steps. If the DAC switch is off, playback fails with error -5.
  • Headphones are silent while the Speaker path is on.
  • The driver waits 700 ms for the amp when the card opens after about 5 s idle.
  • The amp stays on 5 s after a sound; S20audio stop turns it off before power-off to avoid a crackle.

Wi-Fi and Bluetooth ​

  • Without keep-power-in-suspend (patch 0013) rtw88 aborts every suspend.
  • Sleeping while connected hangs the device, so sleep sends DISCONNECT first and RECONNECT after.
  • Changing rtw88's MAC while Wi-Fi runs stalled the link 2 times in 7, so MAC changes restart Wi-Fi.
  • Bluetooth shares the chip: Wi-Fi fails to probe while the Bluetooth half is held in reset, and rtw88's deep power save stalled Wi-Fi, so rtw88_core.disable_lps_deep=Y is on the kernel command line (the driver reads it only when the interface comes up). Patch 0017 keeps the Bluetooth driver from sleeping in probe.
  • LE encryption started right after connecting never completes with some controllers; BlueZ patch 0001 raises security on demand. btmon is in the image.
  • This wpa_supplicant has no syslog: -s just prints usage.

Other ​

  • The GPU needs CONFIG_SUN50I_H6_PRCM_PPU, or Panfrost fails silently. It runs at a fixed 432 MHz.
  • The CPU governor raises the speed slowly from idle, so after the in-game menu the CPU is held at 1008 MHz for two seconds, or games stutter.
  • Kernel modules are off. Don't turn CONFIG_MODULES off over a config with =m entries: they all become built in.
  • The linux-sunxi wiki returns 403 to automated fetches. Use U-Boot's doc/board/allwinner/sunxi.rst, the ROCKNIX tree and the source instead.

Tetsu OS is released under the MIT license. Emulator cores keep their own licences.