megi's PinePhone Development Log RSS

+972

2020–12–02: Fixing fallout from the new WiFi driver

While testing caching and latency of call indication messages, I've noticed that the updated WiFi driver broke suspend/resume cycle, and also that it increased the resume times significantly.

Basically the driver has a compile option to either do the whole of WiFi resume synchronously during resume, or to punt it to the separate thread and don't block the resume process. The latter option only works with Android wakelock API. Though the only thing that the driver needs to be able to do this is to block further suspend to RAM until the asynchronous resume job finishes. That was fairly easy to achieve with mainline API, so now WiFi driver will do its 850ms of resume tasks without blocking the resume process as a whole, which makes userspace responsive just 400ms after resume is initiated, instead of after 1.2s.

Also WoWLAN support code path made the driver return -ENOSYS (-38) from the suspend callback, which aborted the suspend to RAM. I disabled WoWLAN in this patch https://megous.com/git/linux/commit/?h=wifi-5.10&id=6959135b030f7b3bb4db4c42824ab1743436889a I don't think anyone uses WoWLAN support, so that seems like an acceptable solution for now.

One thing that I noticed about the WiFi on Pinephone with the updated driver and newly enabled power management is that it can take some time to ping through to the phone from my PC, after there was no network activity for some long period of time. This is the same behavior that the broadcom wifi on Pinebook Pro exhibits, and it's probably just some property of improved power management.

The fixes are in this branch: https://megous.com/git/linux/log/?h=wifi-5.10