Re: [PATCH v3 00/12] wcn36xx: Enable downstream consistent Wake on Lan

[Benjamin Li <benl@squareup.com>]

On 3/19/21 9:15 AM, Bryan O'Donoghue wrote:
> V3:
> - Ifdefs an IPV6 specific function which will not compile unless CONFIG_IPV6
>   is swithced on - Kernel test robot
> 
> V2:
> - Drops work-around for scan in progress
>   This is not required on a tip of tree kernel - Kalle
> - Churn out "this patch" form commit logs - Kalle
> - Adds comment about packing structures to commit log patch #8 - Kalle
> - Moves enablement of magic-packet and GTK rekey to last patch
>   This means that the functionality is only switched on when the full
>   series has been applied - Kalle
> 
> https://www.spinics.net/lists/linux-wireless/msg206927.html
> 
> V1:
> This series enables Wake on WLAN for the wcn36xx. Downstream implements two
> competing methods for WoWLAN, only one of which is used in practice.
> 
> 1. The downstream prima driver and firmware have a dedicated wowlan SMD
>    command set, however for whatever reason, the downstream driver doesn't
>    actually use that code path for suspend/resume.
>    Similarly the downstream kernel will mask out the SMD interrupt when
>    entering into suspend, so its clear the firmware assisted wowlan command
>    is not used at all.
>    In short, the code is implemented but not used.
> 
> 2. The alternative version of wake on wlan implemented downstream does
>    A. An RXP filter
>    B. Frimware ipv4 ARP offload
>    C. Frimware ipv6 namesapece offload
>    D. GTK replay counter save/update on suspend/resume
> 
> This patchset replicates method #2 since it is the only method that
> actually works. When replicating the functionality from downstream to
> upsteam I found the RXP filter step to be redundant, so I skipped that.
> 
> Verified on Android/4.19 with qcom out-of-tree low-power patch on wcn3680.
> Verified on Debian/kvalo/master on wcn3680 and wcn3620.
> 
> The Debian/upstream test doesn't include the low-power patches from qcom so
> the time it takes the system to wake up when pinged is longer.
> 
> Test method:
> 
> 1. Enable magic packet wakeup on target
>    iw phy0 wowlan enable magic-packet
>    iw phy0 wowlan show
> 
> 2. Flush arp tables on host
>    sudo ip -s -s neigh flush all
>    sudo ip -6 -s -s neigh flush all
> 
> 3. Suspend device
>    echo 1 > /sys/kernel/debug/clk/debug_suspend
>    echo mem > /sys/power/state ; cat /sys/kernel/debug/rpm_stats
>    echo 0 > /sys/module/wcn36xx/parameters/debug_mask
>    dmesg
> 
> 4. Ping either ipv4 or ipv6
>    ping 192.168.0.85
>    ping -6 fe80::6455:44ff:fe33:2211%wlo1

Tested-by: Benjamin Li <benl@squareup.com>

Independently re-tested on Square Terminal with WCN3680B using these steps
on a recently rebased tree (last month).

Used the Linaro DB410c Debian image as userspace & called nmcli to enable
WoWLAN instead of iw directly.

sudo nmcli dev wifi connect <SSID> password "<password>"
sudo nmcli connection modify <SSID> connection.autoconnect yes
# (^ need to set MAC address in device tree for autoconnect at boot)
sudo nmcli connection modify <SSID> wifi.wake-on-wlan magic
sudo service network-manager restart

- arping returns the WLAN MAC address without waking up the device from
  suspend.
- ping wakes up the device from suspend (it does take ~8-16s without
  LPM patches).
- Device stayed connected after leaving device in suspend for longer than
  my network's GTK rekeying interval (1hr) and then waking from suspend.

> 
> In both cases using Wireshark we see the target system receive an ARP or
> NS6 lookup and respond, followed by the host sending a ping or ping6 and
> the target system coming out of suspend.
> 
> Similarly watching GTK keying on the AP when the device is in suspend, we
> see the firmware successfully rekey and when we resume, the host picking up
> the rekey material with persistence of the WiFi link during suspend/resume.
> 
> commit: 5336fad96e8f ("wcn36xx: Enable firmware offloaded keepalive")
> ensures that the link is kept alive in suspend, this additional set gives
> the ability to 
> 
> - Resolve ARP/NS6 lookups without waking the system
> - Rekeying in suspend
> - Suspending and resuming based on received datagram
> 
> https://git.linaro.org/people/bryan.odonoghue/kernel.git/log/?h=ath.master-wcn36xx-fw-offload-suspend-resume
> 
> Bryan O'Donoghue (12):
>   wcn36xx: Return result of set_power_params in suspend
>   wcn36xx: Run suspend for the first ieee80211_vif
>   wcn36xx: Add ipv4 ARP offload support in suspend
>   wcn36xx: Do not flush indication queue on suspend/resume
>   wcn36xx: Add ipv6 address tracking
>   wcn36xx: Add ipv6 namespace offload in suspend
>   wcn36xx: Add set_rekey_data callback
>   wcn36xx: Add GTK offload to WoWLAN path
>   wcn36xx: Add GTK offload info to WoWLAN resume
>   wcn36xx: Add Host suspend indication support
>   wcn36xx: Add host resume request support
>   wcn36xx: Enable WOWLAN flags
> 
>  drivers/net/wireless/ath/wcn36xx/hal.h     |  20 +-
>  drivers/net/wireless/ath/wcn36xx/main.c    | 118 ++++++++-
>  drivers/net/wireless/ath/wcn36xx/smd.c     | 265 +++++++++++++++++++++
>  drivers/net/wireless/ath/wcn36xx/smd.h     |  17 ++
>  drivers/net/wireless/ath/wcn36xx/wcn36xx.h |  14 ++
>  5 files changed, 420 insertions(+), 14 deletions(-)
>