/rtl8812au

Linux kernel driver for USB WiFi dongles based on Realtek rtl8812au chipset

Primary LanguageC

rtl8812au

Linux kernel driver for USB WiFi dongles based on Realtek rtl8812au chipset

Contents

Description

Linux kernel driver for USB WiFi dongles based on Realtek rtl8812au chipset.

This is a fork of gordboy/rtl8812au, which is based on 5.2.20 realtek driver version:

gordboy repo already has VHT patches and is updated to work with latest kernels.

This repo is for testing various performance-related patches/tweaks on top of that.

For ArchLinuxARM PKGBUILD of module from this repo for ODROID-C2 (aarch64, mainline 4.13.x+ kernel), see:

For more orthodox forks of this driver, which should probably be picked instead of this one for any kind of general use, see following links:

Note on support / development / patches

I'm not supporting or developing this driver here, just collected a bunch of ad-hoc patches from different sources for my own purposes.

Please do not ask me to add or merge additional featues or fix something here - I'm not a kernel developer and don't even have hardware to test this locally.

Maybe just fork the repo and apply whatever you need there instead, or complain about anything that doesn't work right to hardware vendor or realtek (though note that they haven't released this code publicly - see source links at the top of this README).

Upstreaming status

Coming from the future (2018+), it's possible that you might find this driver to be superseded by one of these two modules in mainline kernels:

  • rtl8xxxu - as of early 2018, 881xau ac dongles are not supported there.

    More info on history, rationale and internals of both rtl8xxxu and realtek vendor drivers:

  • rtlwifi - more official one, supported by realtek people.

    Already supports some USB dongles (rtl8192cu), and since Aug 2017 support for rtl882xxe (PCI-E, some in staging) and common 802.11ac stuff (e.g. phydm) have been merged there too.

    Though Nov 2017 thread on linux-wireless ML shows that there are no plans for porting 8812au there, and that it is highly unlikely to happen:

    To my knowledge, there are no plans to convert the buggy, and badly written, code for the RTL8812AU into a form that would be suitable for wireless. Even getting it into a form that would be acceptable for staging would be a formidable task. -- Larry Finger

    https://www.spinics.net/lists/linux-wireless/msg167866.html

Performance

Should very heavily depend on card/dongle used, but between two ALFA Network AWUS036AC USB dongles plugged into RPi3/ODROID-C2 (USB 2.0) that I've tested using WiFi Infrastructure mode (AP + STA) with VHT (802.11ac) and 80MHz channels (in 5GHz band), throughput max with iperf3 seem to be about 220mbps or ~26 MiB/s (half-duplex).

Note that such result is for one specific pair of AP/STA dongles, running in a specific environment and configuration, so does not mean that other devices supported by this driver will have same limitations, and here just to show that driver itself can at least support that much (easily - extra cpu load is negligible, no bugs).

Relevant hostapd.conf options used for AP dongle and test:

hw_mode=a
ieee80211n=1
require_ht=1
ieee80211ac=1
require_vht=1

channel=100
vht_oper_chwidth=1
vht_oper_centr_freq_seg0_idx=106
ht_capab=[HT40+][SHORT-GI-40]
vht_capab=[SHORT-GI-80][MAX-MPDU-11454][MAX-A-MPDU-LEN-EXP7]

Note that some of LDPC/STBC/Beamforming options should be enabled for AP/STA implicitly, if dongle/driver support these, and can be checked by grepping "RTW: Current STA" in dmesg (with debug logging enabled) after connection (vs flags in rtw_vht.c).

Same as for test results, these aren't necessarily supported by all dongles, and some dongles might allow e.g. 160MHz or 80+80 channel widths (different vht_oper_chwidth values), 4x4 MIMO/beamforming, and such, some of which driver seem to have the code for.

For a bit more info on AP/STA mode configuration, see following links:

Very similar performance between 5.1.5 (old-5.1.5 branch) and 5.2.20 (this one) driver versions, while 5.2.9 is way worse.

Misc Notes

  • Driver initially had bunch of regdb overrides, maybe to operate with no/broken regdb, which are removed here, to avoid any extra restrictions from these.

  • Module parameter rtw_tx_pwr_idx_override (1-63, 0 - default/disabled) allows to override "tx power index" value for all channels/rates, regardless of all other parameters (regdb, CONFIG_TXPWR_BY_RATE_EN, CONFIG_TXPWR_LIMIT_EN, CONFIG_USB2_EXTERNAL_POWER, etc), if set to non-0.

    CONFIG_DBG_TXPOWER can be used to check default/overidden values for that (enabled with DEBUG>1, "TXPWR:" prefix).

  • CONFIG_USB2_EXTERNAL_POWER is default-enabled in Makefile, can be set to "n" to force some txpower limit, don't remember which one.

  • DFS is default-disabled in Makefile.

Debugging

First of all, enable DEBUG= with make or dkms.conf, e.g. make DEBUG=2.

Values for DEBUG= are 0-3 for progressively more info, see Makefile and include/autoconf.h for details or to tweak which specific bits should be enabled.

With DEBUG>0, kmsg logging verbosity can be controlled by rtw_drv_log_level= module parameter: 0=none, 2=error, 3=warn, 4=info (default), 5=debug, 6=max.

For more information about what your specific dongle supports and is configured for, use /proc interface (with DEBUG>0) that this driver provides under /proc/net/rtl8812au/.

Some useful info nodes there (replace "wlan0" below with your interface name):

  • /proc/net/rtl8812au/ver_info - loaded driver version.

  • /proc/net/rtl8812au/drv_cfg - build-time driver configuration info.

  • /proc/net/rtl8812au/log_level - kmsg (dmesg) logging control.

  • /proc/net/rtl8812au/wlan0/phy_cap - phy capabilities (VHT, STBC, Beamforming, etc), as supported by hardware/driver and configuration, including resulting mask of them.

  • /proc/net/rtl8812au/wlan0/{ap,sta}_info

  • Counters / stats / metrics:

    • /proc/net/rtl8812au/wlan0/rx_signal
    • /proc/net/rtl8812au/wlan0/trx_info
    • /proc/net/rtl8812au/wlan0/trx_info_debug
    • /proc/net/rtl8812au/wlan0/rx_stat - counts of frames transmitted at specific mimo modes and rates (e.g. VHTSS2MCS4 = VHT + 2 Spatial Streams + MCS4 coding set, see include/hal_com.h and http://mcsindex.com/ ), gets reset after each poll.
    • /proc/net/rtl8812au/wlan0/{rx,tx,int}_logs - lots of counters.

  • /proc/net/rtl8812au/wlan0/odm/cmd

    Read/write "console" for phydm_cmd() (phydm_debug.c, PHY DM = PHY-layer Dynamic Management) interface in the driver, which can be used to debug its low-level operation.

    Expects "command arg1 arg2 ... argX" (commas also work) string to be written there, ending in newline, which goes to phydm_cmd() and output can be read via subsequent read on that node ("GET, nothing to print" indicates no output/command).

    phydm-cmd.py script (python3/readline) in this repo can be used to work with this interface interactively.

    Some phydm_cmd() commands/examples:

    • -h - list supported commands, except -h itself and demo.

    • demo 10 0x3a z abcde - dummy command to print arguments, parsed as different types.

    • dbg - controls debug logging for different phydm components.

      • dbg 100 - dump different phydm debug logging components, and whether logging for each one is enabled ("V") or not (".").

        Should look something like this:

        ================================
[Debug Message] PhyDM Selection
================================
00. (( . ))DIG
01. (( . ))RA_MASK
02. (( V ))DYNAMIC_TXPWR
03. (( . ))FA_CNT
04. (( . ))RSSI_MONITOR
05. (( . ))CCKPD
06. (( . ))ANT_DIV
07. (( . ))SMT_ANT
...

      • dbg 101 - disable phydm debug logging (all components).

      • dbg 4 1 - enable phydm debug logging for RSSI_MONITOR component - 04 in the "dbg 100" list (see above, number parsed as decimal) - 1=enable, 2=disable.

      These will be logged to kmsg/dmesg, same as other debug stuff from driver.

    • h2c 0x00 0x01 ... 0x07 - send H2C command 0x00 to firmware with specified parameters.

      H2C = Host-To-Chip (?) command from CMD/EVENT mechanism to make firmware do stuff on its own (offload) or change parameters, reporting back via C2H event. See e.g. RTL8712_D0_1_Programming_Guide_20090601.pdf for info on such concepts.

      Check h2c_cmd enum in include/hal_com_h2c.h for list of commands, or h2c_cmd struct in rtl8xxxu.h under linux sources (which is probably more descriptive), or similar stuff in rtlwifi module.

    • tune_para - easy way to add runtime tweaks to any values in phydm struct (which can then be used/applied pretty much anywhere in the driver), see phydm_parameter_adjust().

    • set_txagc - sets TX power index (0-63) for specific antenna (RF path, 0-3 for A-D) and MCS (0xff - all of them), e.g. set_txagc 1 0 0xff 0x3f to set power index for antenna-A to 63 (max) for all rates.

      set_txagc 0 - disable all such overrides.
      Not sure what it has to do with AGC, as it usually applies to RX only.
      Seem to be for specific non-8812a chips only - 8822b / 8197f / 8821c, oh well.
      Has get_txagc counterpart, which prints TXAGC values on supported chips.

      This module version has similar tweak as rtw_tx_pwr_idx_override= module option, which sets specified power index for all rf paths and rates.

    • ra - Rate Adaptivity/Adjustment info and tuning - how driver/firmware adjusts rates, depending on errors and radio conditions.

      • ra -h - help on arguments there.

      • ra 1 ... - show (ra 1 100) or set PCR threshold offset (affects retries?), e.g. ra 1 0 10 for -10, ra 1 1 20 for +20.

        Range: -127 - +127, always 0 unless set manually.
        Sent in ODM_H2C_RSSI_REPORT to fw every ~2 seconds.

      • ra 2 ... - enable/disable fixed rate in fw for specific macid.

        For example ra 2 1 0 2 63 for enabling fixed rate for macid=0 (first/only peer, see various log msgs like rtw_alloc_macid for these), bw=2 for 80M (calculated as 20<<2=80, 0=20M, 1=40M and so on), rate=63 for VHTSS2MCS9 (see ./phydm-cmd.py -x or hal_com.h).

        See also dbg 31 1 (ODM_COMP_API) for logging from phydm_fw_fix_rate and such.

      • ra 3 ... (only with CONFIG_RA_DBG_CMD) - dump or tweak per-rate dm_ra_table parameters via odm_RA_debug.

      See also:

      • RA/RA_MASK (01, 09) in dbg command for logging of RA-related stuff.

      • Related build-time options - CONFIG_DBG_RA, CONFIG_RA_DBG_CMD, CONFIG_RA_FW_DBG_CODE, CONFIG_RA_DYNAMIC_RTY_LIMIT, CONFIG_RA_DYNAMIC_RATE_ID - in autoconf.h and phydm_rainfo.c.

        CONFIG_RA_FW_DBG_CODE in particular to catch c2h (chip-to-host?) debug info from fw and dump them if DBG_FW_TRACE is enabled (dbg 22 - FW_DEBUG_TRACE logging).

      • fw_dbg command to enable firmware debug components via h2c, of which RA (00) is probably the only one available on consumer chips (see fw_dbg 100 for full list). fw_dbg 101 disables all fw debug components.

      • phydm_c2h_ra_report_handler and odm_c2h_ra_para_report_handler in phydm_rainfo.c - relatively recent code ("2017.04.20 Dino, the 3rd PHYDM reform"), though not sure if these ever trigger on end-user chips.

    • ... and there's much more of them, see -h output.

Links

Repositories that seem to be most active (as of Jan 2018) wrt info on this driver, i.e. places to watch for new issues, commits, pull requests and forks (in no particular order):

Not all (or any?) of these forks are linked under "Forks" tab on github.

Be sure to check different branches in these, as there are several different upstream sources (code dumps) for this driver, which these are usually based on.

More general links: