en:users:wi-fi-alliance-certification
Differences
This shows you the differences between two versions of the page.
| Both sides previous revision Previous revision Next revision | Previous revision Next revision Both sides next revision | ||
|
en:users:wi-fi-alliance-certification [2021/08/02 02:55] Ming Shuo Chiu [Table] |
en:users:wi-fi-alliance-certification [2021/10/18 09:22] Reto Schneider [Table] Patch #8 wil not be upstreamed |
||
|---|---|---|---|
| Line 9: | Line 9: | ||
| ===== rtl8xxxu on RTL8188CUS on GARDENA smart gateway ===== | ===== rtl8xxxu on RTL8188CUS on GARDENA smart gateway ===== | ||
| + | RealTek provides a GPLed Linux driver named "8192cu" for its RTL8188CUS hardware. While it has worked [[https://api.cert.wi-fi.org/api/certificate/download/public?variantId=14856 | well enough to pass the Wi-Fi Alliance certification]], the last release [[https://github.com/rettichschnidi/realtek-linux/tree/master/RTL8188C_8192C/USB/v4.0 | is from 2017]], its code does not work well with modern kernels (FullMAC, and notoriously unstable) and does no longer receive any security updates. Luckily, it has been superseded by TWO mainlined kernel drivers: | ||
| + | * rtl8192cu: Based on the RealTek driver, shoehorned into mainline Linux. Suffers from various stability issues and code is close to unreadable. | ||
| + | * rtl8xxxu: Written from scratch, [[en::users::drivers::rtl819x | will replace rtl8192cu]]. It has been shown to to be stable during long-term testing. Code is readable. Not (yet) feature equivalent to the 8192cu or even the rtl8192cu. | ||
| + | |||
| + | Both, rtl8192cu and rtl8xxxu are worse than 8192cu when it comes to features and performance. | ||
| + | |||
| + | Given the code quality and stability, it is the goal of GARDENA to develop the rtl8xxxu driver as far as needed to pass the Wi-Fi Alliance certification. | ||
| + | |||
| + | A big hurdle is [[https://blog.linuxplumbersconf.org/2016/ocw/system/presentations/4089/original/2016-11-02-rtl8xxxu-presentation.pdf | the lack of sources as well as documentation for the firmware]]. Therefore, digging in the (hard to read) sources of the vendor driver is needed. | ||
| ==== Hardware ==== | ==== Hardware ==== | ||
| * End product: ARMv5, Atmel SAM9G25 based GARDENA smart gateway ([[https://github.com/husqvarnagroup/smart-garden-gateway-public|sources]]) | * End product: ARMv5, Atmel SAM9G25 based GARDENA smart gateway ([[https://github.com/husqvarnagroup/smart-garden-gateway-public|sources]]) | ||
| Line 32: | Line 41: | ||
| == Chris == | == Chris == | ||
| - | * AP: ? | + | * AP: Dlink DIR-612 (2x2) SSID: CHT57196 with BSSID: 18:0f:76:c5:0a:ea |
| * STA: rtl8xxxu running on Edimax EW-7811UN (amd64) | * STA: rtl8xxxu running on Edimax EW-7811UN (amd64) | ||
| - | * Distance: ? | + | * Distance: fixed to 3m and 6m for experiment |
| - | * Environment: ? | + | * Environment: use less-noisy channel 9 in a 8x9 square meters area which only have 1 router. |
| === Baseline === | === Baseline === | ||
| Line 107: | Line 116: | ||
| </code> | </code> | ||
| - | Unsure if this is a common issue. | + | <del>Unsure if this is a common issue.</del> Seems hardware specific - USB dongle probably broken. |
| == RX IQK failed == | == RX IQK failed == | ||
| Line 140: | Line 149: | ||
| </code> | </code> | ||
| - | Unsure if this is a common issue. | + | <del>Unsure if this is a common issue.</del> Seems hardware specific - USB dongle probably broken. |
| === Patches === | === Patches === | ||
| - | | **Number** | **Description** | **Link** | **Testing results (Reto)** | **Upstream State** | | + | | **Number** | **Description** | **Link** | **Testing results (Reto)** | **Upstream State** | |
| - | | 1 | Handle BSS_CHANGED_TXPOWER/IEEE80211_CONF_CHANGE_POWER | https://github.com/husqvarnagroup/linux/commit/798796ff5070255a7cccc7529c272b629da79d7b | | | | + | | 1 | Handle BSS_CHANGED_TXPOWER/IEEE80211_CONF_CHANGE_POWER | https://github.com/husqvarnagroup/linux/commit/798796ff5070255a7cccc7529c272b629da79d7b | | | |
| - | | 2 | Handle for mac80211 get_txpower | https://github.com/husqvarnagroup/linux/commit/5df123ce78422b76603eedf799fc50f5e3303011 | | | | + | | 2 | Handle for mac80211 get_txpower | https://github.com/husqvarnagroup/linux/commit/5df123ce78422b76603eedf799fc50f5e3303011 | | | |
| - | | 3 | Enable RX STBC by default | https://github.com/husqvarnagroup/linux/commit/4efe4bf653a14b61ff9b3969050bc696976415eb | | | | + | | 3 | Enable RX STBC by default | https://github.com/husqvarnagroup/linux/commit/4efe4bf653a14b61ff9b3969050bc696976415eb | | | |
| - | | 4 | Feed antenna information for mac80211 | https://github.com/husqvarnagroup/linux/commit/070d0eca2dbe8086a215c33f6d3ee10cfe9a17cc | | | | + | | 4 | Feed antenna information for mac80211 | https://github.com/husqvarnagroup/linux/commit/070d0eca2dbe8086a215c33f6d3ee10cfe9a17cc | | | |
| - | | 5 | Fill up txrate info for all chips | https://github.com/husqvarnagroup/linux/commit/fa14d07da5566c8abebccd68473ddc17e908be45 | | | | + | | 5 | Fill up txrate info for all chips | https://github.com/husqvarnagroup/linux/commit/fa14d07da5566c8abebccd68473ddc17e908be45 | | | |
| - | | 6 | Fix the reported rx signal strength | https://github.com/husqvarnagroup/linux/commit/95f19fff95a62ab67ecc50d75e8c59669f936bd2 | | | | + | | 6 | Fix the reported rx signal strength | https://github.com/husqvarnagroup/linux/commit/95f19fff95a62ab67ecc50d75e8c59669f936bd2 | | | |
| - | | 7 | Fix the handling of TX A-MPDU aggregation | https://lore.kernel.org/linux-wireless/20210630160151.28227-1-chris.chiu@canonical.com/#t | GARDENA gateway: Improves TX throughtput, but worsens retry percentage; Reverse issue for RX. | WIP | | + | | 7 | Fix the handling of TX A-MPDU aggregation | https://lore.kernel.org/linux-wireless/20210804151325.86600-1-chris.chiu@canonical.com/ | GARDENA gateway: Improves TX throughput, but worsens retry percentage; Reverse issue for RX. | Merged | |
| - | | 8 | Improve the retransmission rate with HW_RTS enable | https://github.com/husqvarnagroup/linux/commit/31a00cbf0245877e9a98cdec6903758a928482e5 | GARDENA gateway: Neither troughput nor retry percentage improved | | | + | | 8 | Improve the retransmission rate with HW_RTS enable | https://github.com/husqvarnagroup/linux/commit/31a00cbf0245877e9a98cdec6903758a928482e5 | GARDENA gateway: Neither throughput nor retry percentage improved | | |
| - | | 9 | Set RTS rate to 24M for AMPDU | https://github.com/husqvarnagroup/linux/commit/b35338e600b003f830c4533c162a53c57d9d34d2 | GARDENA gateway: no positive effectes observable | | | + | | 9 | Set RTS rate to 24M for AMPDU | https://github.com/husqvarnagroup/linux/commit/b35338e600b003f830c4533c162a53c57d9d34d2 | GARDENA gateway: No positive effects observable | will not be upstreamed | |
| - | | 10 | Reduce number of USB interrupts | https://lore.kernel.org/linux-wireless/d32690a6-f679-c676-1461-10b47ae3428b@gmail.com/T/#t | Clearly improves throughput and retries | ACKed by maintainer | | + | | 10 | Reduce number of USB interrupts | https://lore.kernel.org/linux-wireless/d32690a6-f679-c676-1461-10b47ae3428b@gmail.com/ | GARDENA gateway: Clearly improves throughput and retries | Merged | |
| + | | 11 | Use lower tx rates for the ack packet | https://lore.kernel.org/linux-wireless/20211001040044.1028708-1-chris.chiu@canonical.com/ | GARDENA gateway: Neither throughput nor retry percentage improved | Merged | | ||
| ==== Driver Testing ==== | ==== Driver Testing ==== | ||
| Line 181: | Line 191: | ||
| | amd64, Edimax | Linksys | Linux v5.10.51, rtl8xxxu (vanilla) | 30.6 Mbits/s; 12% | 2m, two monitors in line of sight | Many "ghost MAC addresses" | https://files.reto-schneider.ch/rtl8xxxu/2021-07-23-Testing/2021-07-23-1-Linux-5.10.51-2021-07-22-vanilla-rtl8xxxu-edimax-08:be:ac:0b:14:c4-iperf-tx-tcp-12%25-retries-at-30.6MBps.pcapng.gz | | | amd64, Edimax | Linksys | Linux v5.10.51, rtl8xxxu (vanilla) | 30.6 Mbits/s; 12% | 2m, two monitors in line of sight | Many "ghost MAC addresses" | https://files.reto-schneider.ch/rtl8xxxu/2021-07-23-Testing/2021-07-23-1-Linux-5.10.51-2021-07-22-vanilla-rtl8xxxu-edimax-08:be:ac:0b:14:c4-iperf-tx-tcp-12%25-retries-at-30.6MBps.pcapng.gz | | ||
| | amd64, Edimax | Linksys | Linux v5.10.51, rtl8xxxu (vanilla) | 13.1 Mbits/s; 3% | 2m, two monitors in line of sight | Many "ghost MAC addresses"; unexplained performance/retry drop | https://files.reto-schneider.ch/rtl8xxxu/2021-07-23-Testing/2021-07-23-3-Linux-5.10.51-2021-07-22-vanilla-rtl8xxxu-edimax-08:be:ac:0b:14:c4-iperf-tx-tcp-3%25-retries-at-13.1MBps.pcapng.gz | | | amd64, Edimax | Linksys | Linux v5.10.51, rtl8xxxu (vanilla) | 13.1 Mbits/s; 3% | 2m, two monitors in line of sight | Many "ghost MAC addresses"; unexplained performance/retry drop | https://files.reto-schneider.ch/rtl8xxxu/2021-07-23-Testing/2021-07-23-3-Linux-5.10.51-2021-07-22-vanilla-rtl8xxxu-edimax-08:be:ac:0b:14:c4-iperf-tx-tcp-3%25-retries-at-13.1MBps.pcapng.gz | | ||
| - | | amd64, Edimax | DLink | Linux v5.13.1 rtl8xxxu | 23~27 Mbits/s; 12~14% | 3m, direct line of sight | performance drop for a few seconds | https://mega.nz/folder/LI0ATTAK#0E2J5DHksD1jQi1oJCOJWg | | ||
| - | | amd64, Edimax | DLink | Linux v5.13.1 rtl8xxxu | 7-14 Mbits/s; 16~20% | 6m, direct line of sight | performance drop to 0 sometimes | https://mega.nz/folder/vMdgADRD#XHljNHbzzlp63qqFsT-rkQ | | ||
| | amd64, Edimax | Linksys | Linux v5.10.51, 8192cu (tag rs/2021-07-23) | 4.02 Mbits/s; 17.2% | 2m, two monitors in line of sight | No "ghost MAC addresses"; Surprisingly bad performance | https://files.reto-schneider.ch/rtl8xxxu/2021-07-23-Testing/2021-07-23-5-Linux-5.10.51-2021-07-22-8192cu-edimax-08:be:ac:0b:14:c4-iperf-tx-tcp-16%25-retries-at-4.02MBps.pcapng.gz | | | amd64, Edimax | Linksys | Linux v5.10.51, 8192cu (tag rs/2021-07-23) | 4.02 Mbits/s; 17.2% | 2m, two monitors in line of sight | No "ghost MAC addresses"; Surprisingly bad performance | https://files.reto-schneider.ch/rtl8xxxu/2021-07-23-Testing/2021-07-23-5-Linux-5.10.51-2021-07-22-8192cu-edimax-08:be:ac:0b:14:c4-iperf-tx-tcp-16%25-retries-at-4.02MBps.pcapng.gz | | ||
| | amd64, Edimax | Linksys | Linux v5.10.51, 8192cu (tag rs/2021-07-23) | 3.91 Mbits/s; 18.8% | 2m, two monitors in line of sight | No "ghost MAC addresses"; Surprisingly bad performance | https://files.reto-schneider.ch/rtl8xxxu/2021-07-23-Testing/2021-07-23-6-Linux-5.10.51-2021-07-22-8192cu-edimax-08:be:ac:0b:14:c4-iperf-tx-tcp-18%25-retries-at-3.91MBps.pcapng.gz | | | amd64, Edimax | Linksys | Linux v5.10.51, 8192cu (tag rs/2021-07-23) | 3.91 Mbits/s; 18.8% | 2m, two monitors in line of sight | No "ghost MAC addresses"; Surprisingly bad performance | https://files.reto-schneider.ch/rtl8xxxu/2021-07-23-Testing/2021-07-23-6-Linux-5.10.51-2021-07-22-8192cu-edimax-08:be:ac:0b:14:c4-iperf-tx-tcp-18%25-retries-at-3.91MBps.pcapng.gz | | ||
| | amd64, Edimax | Linksys | Linux v5.10.51, 8192cu (tag rs/2021-07-23) | 49.7 Mbits/s; 4.8% | 1m, direct line of sight | Some "ghost MAC addresses" | https://files.reto-schneider.ch/rtl8xxxu/2021-07-23-Testing/2021-07-23-8-Linux-5.10.51-2021-07-22-8192cu-edimax-08:be:ac:0b:14:c4-iperf-tx-tcp-4.8%25-retries-at-49.7MBps.pcapng.gz | | | amd64, Edimax | Linksys | Linux v5.10.51, 8192cu (tag rs/2021-07-23) | 49.7 Mbits/s; 4.8% | 1m, direct line of sight | Some "ghost MAC addresses" | https://files.reto-schneider.ch/rtl8xxxu/2021-07-23-Testing/2021-07-23-8-Linux-5.10.51-2021-07-22-8192cu-edimax-08:be:ac:0b:14:c4-iperf-tx-tcp-4.8%25-retries-at-49.7MBps.pcapng.gz | | ||
| | amd64, Edimax | Linksys | Linux v5.10.51, rtl8xxxu (vanilla) | 45.7 Mbits/s; 9.4% | 1m, direct line of sigh | No "ghost MAC addresses" | https://files.reto-schneider.ch/rtl8xxxu/2021-07-23-Testing/2021-07-23-10-Linux-5.10.51-2021-07-22-vanilla-rtl8xxxu-edimax-08:be:ac:0b:14:c4-iperf-tx-tcp-9.4%25-retries-at-45.7MBps.pcapng.gz | | | amd64, Edimax | Linksys | Linux v5.10.51, rtl8xxxu (vanilla) | 45.7 Mbits/s; 9.4% | 1m, direct line of sigh | No "ghost MAC addresses" | https://files.reto-schneider.ch/rtl8xxxu/2021-07-23-Testing/2021-07-23-10-Linux-5.10.51-2021-07-22-vanilla-rtl8xxxu-edimax-08:be:ac:0b:14:c4-iperf-tx-tcp-9.4%25-retries-at-45.7MBps.pcapng.gz | | ||
| - | | amd64, Edimax | DLink | Linux v5.12, 8192cu (https://github.com/mschiu77/rtl8188cus_vendor.git master) | 40~42 Mbits/s; 9% | 3m, direct line of sight | steady performance | https://mega.nz/folder/GAcwnDjY#Xf9lVMriWcPpIUACjuZiqg | | ||
| - | | amd64, Edimax | DLink | Linux v5.12, 8192cu ((https://github.com/mschiu77/rtl8188cus_vendor.git master) | 13-16 Mbits/s; 20~23% | 6m, direct line of sight | steady throughput w/o sudden drop | https://mega.nz/folder/HZ8iQRqY#2ss6WW9u6oxInJvT2R6iCw | | ||
| | ARMv5, GARDENA smart Gateway (00:1d:43:c0:19:8a) | Linksys | Linux v5.10.52, 8192cu (tag rs/2021-07-23) | 30.6 Mbits/s; 0.9% | 1m, direct line of sigh | Some "ghost MAC addresses"; This is the kind of performance wanted | https://files.reto-schneider.ch/rtl8xxxu/2021-07-24-Testing/2021-07-24-3-Linux-5.10.52-8192cu-gateway-00:1d:43:c0:19:8a-iperf-tx-tcp-0.9%-retries-at-30.6MBps.pcapng.gz | | | ARMv5, GARDENA smart Gateway (00:1d:43:c0:19:8a) | Linksys | Linux v5.10.52, 8192cu (tag rs/2021-07-23) | 30.6 Mbits/s; 0.9% | 1m, direct line of sigh | Some "ghost MAC addresses"; This is the kind of performance wanted | https://files.reto-schneider.ch/rtl8xxxu/2021-07-24-Testing/2021-07-24-3-Linux-5.10.52-8192cu-gateway-00:1d:43:c0:19:8a-iperf-tx-tcp-0.9%-retries-at-30.6MBps.pcapng.gz | | ||
| | ARMv5, GARDENA smart Gateway (00:1d:43:c0:19:8a) | Linksys | Linux v5.10.52, 8192cu (tag rs/2021-07-23) | 42.5 Mbits/s; 2.4% | 1m, direct line of sigh | Some "ghost MAC addresses"; This is the kind of performance wanted | https://files.reto-schneider.ch/rtl8xxxu/2021-07-24-Testing/2021-07-24-4-Linux-5.10.52-8192cu-gateway-00:1d:43:c0:19:8a-iperf-tx-tcp-2.4%-retries-at-42.5MBps.pcapng.gz | | | ARMv5, GARDENA smart Gateway (00:1d:43:c0:19:8a) | Linksys | Linux v5.10.52, 8192cu (tag rs/2021-07-23) | 42.5 Mbits/s; 2.4% | 1m, direct line of sigh | Some "ghost MAC addresses"; This is the kind of performance wanted | https://files.reto-schneider.ch/rtl8xxxu/2021-07-24-Testing/2021-07-24-4-Linux-5.10.52-8192cu-gateway-00:1d:43:c0:19:8a-iperf-tx-tcp-2.4%-retries-at-42.5MBps.pcapng.gz | | ||
| Line 213: | Line 219: | ||
| === Chris === | === Chris === | ||
| - | TBD | + | |
| + | == Setup == | ||
| + | - Blacklist all drivers (rlt8xxxu, 8192cu, rtl8192cu) | ||
| + | - Ensure no drivers are loaded | ||
| + | - Load driver to be tested (rlt8xxxu, 8192cu) with insmod | ||
| + | - Start 802.11 capture | ||
| + | - Once IP address got assigned, measure the performance using iperf3: | ||
| + | * TCP TX Throughput: <code>iperf3 -c 192.168.0.11 -t 30 -i 1</code> | ||
| + | * TCP RX Throughput: <code>iperf3 -c --reverse 192.168.0.11 -t 30 -i 1</code> | ||
| + | |||
| + | == Repos == | ||
| + | * Linux: https://github.com/mschiu77/linux/tree/chris/ampdu_action | ||
| + | * 8192cu: https://github.com/mschiu77/rtl8188cus_vendor | ||
| + | |||
| + | == Analysis == | ||
| + | * The throughput values are the ones reported by iperf3 (the lower value if server/client values differ) | ||
| + | * The retry values are taken from Wireshark in menu "Wireless" -> "WLAN Traffic" | ||
| + | |||
| + | |||
| + | == TCP TX Throughput == | ||
| + | |||
| + | | **DUT** | **AP** | **Driver** | **Throughput [Mbits/sec]; retries [%]** | **AP <-> DUT** | **Notes** | **PCAP** | | ||
| + | | amd64, Edimax | DLink | Linux v5.13.1 rtl8xxxu | 23~27 Mbits/s; 12~14% | 3m, direct line of sight | performance drop for a few seconds | https://mega.nz/folder/LI0ATTAK#0E2J5DHksD1jQi1oJCOJWg | | ||
| + | | amd64, Edimax | DLink | Linux v5.13.1 rtl8xxxu | 7-14 Mbits/s; 16~20% | 6m, direct line of sight | performance drop to 0 sometimes | https://mega.nz/folder/vMdgADRD#XHljNHbzzlp63qqFsT-rkQ | | ||
| + | | amd64, Edimax | DLink | Linux v5.12, 8192cu (https://github.com/mschiu77/rtl8188cus_vendor.git master) | 40~42 Mbits/s; 9% | 3m, direct line of sight | steady performance | https://mega.nz/folder/GAcwnDjY#Xf9lVMriWcPpIUACjuZiqg | | ||
| + | | amd64, Edimax | DLink | Linux v5.12, 8192cu ((https://github.com/mschiu77/rtl8188cus_vendor.git master) | 13-16 Mbits/s; 20~23% | 6m, direct line of sight | steady throughput w/o sudden drop | https://mega.nz/folder/HZ8iQRqY#2ss6WW9u6oxInJvT2R6iCw | | ||
| + | |||
| + | == TCP RX Throughput == | ||
| + | |||
| + | | **DUT** | **AP** | **Driver** | **Throughput [Mbits/sec]; retries [%]** | **AP <-> DUT** | **Notes** | **PCAP** | | ||
| + | | amd64, Edimax | DLink | Linux v5.13.1 rtl8xxxu | 36 Mbits/s; 12% | 3m, direct line of sight | performance is stable high | https://mega.nz/folder/3Zs0nDrK#i1fnW6Bp5E_jeC3dcPlWaw | | ||
| + | | amd64, Edimax | DLink | Linux v5.13.1 rtl8xxxu | 16 Mbits/s; 30% | 6m, direct line of sight | performance is stably low | https://mega.nz/folder/TcsQAZoI#VbQiPZF1B7EJK1_82x2kOg | | ||
| + | | amd64, Edimax | DLink | Linux v5.12, 8192cu | 35 Mbits/s; 10% | 3m, direct line of sight | performance is stable high | https://mega.nz/folder/KIsmCBRR#rN9X9sXrdLo7It8qj8PQgQ | | ||
| + | | amd64, Edimax | DLink | Linux v5.12, 8192cu | 20 Mbits/s; 11% | 6m, direct line of sight | performance and retry are better than rtl8xxxu | https://mega.nz/folder/2Q1WBJqS#6dU1pQ9OXbveETqlfQcZ2w | | ||
| + | |||
| + | ==== Driver Comparisation ==== | ||
| + | |||
| + | | **Criterion** | **rtl8xxxu** | **rtl8192cu** | **8192cu** | **Comment** | | ||
| + | | wlan.frag != 1 | rarely | often | never | Maybe a side-effect of failed transfers. Never sent by AP. | | ||
| + | | Block ACK Initiator | STA, then AP | AP, then AP | AP, then STA | | | ||
| + | | Block ACK Starting Sequence Control | oldest possible sequence # | | first non-acked sequence # | Should be benign, see https://gjermundraaen.com/2021/03/29/802-11-compressed-blockack-two-different-behaviors/ | | ||
| + | | A-MSDU | Permitted when ADDBA initiated by STA, not when by AP | alway permitted | never permitted | | | ||
| + | | Block ACK Timeout [1024 us] | 0x0 (disabled) | 0x1388 | 0x1388 | | | ||
| + | | wlan.qos.amsdupresent == 1 | some | | | | | ||
en/users/wi-fi-alliance-certification.txt · Last modified: 2021/11/10 00:43 by Reto Schneider
Page Tools
