[Bug] Intermittent spikes in FFB

[matthiasvegh]

Category

Force Feedback

Description

I've been noticing this for a good while now, but recent updates t Assetto Corsa Rally have made it pretty straightforward to reproduce.
Assetto Corsa Rally's FFB has never been great, but I've not seen any mention on the forums about spikey values.

At the start of a stage (Vallee de Munster Descente) the FFB is pretty weak and the steering feels light (0-0.1Nm most of the time, 0.4-0.6Nm sometimes). In and of itself, this is alright for a rally car. But as the stage progresses, after about 45s I start getting 1NM stabs to the right which initially kind of make sense given what the car is doing (although feel a bit overblown) but about a minute later I get 3-4 Nm jabs to the right that ping about once every second. After that, the car straightens out and the stabs continue, but fade down from 4Nm down to 1Nm and then subside.

Initiall, I thought it was a bug and did a git bisect between (6fcf774) and the current master, but I'm getting it there as well now, so I assume that ACR 0.5.0 did something to make things worse.

I've played with FFB filters, turning TF audio down from 30 to 0 (no difference for me) and turning of "advance force feedback gain" (i.e., trueforce) in game as well, none made a difference.

If I had to guess, I'd say some effect somehow is applied cumulatively in that if a medium force is applied for a moment and then is applied again it somehow amplifies. But I can't seem to reproduce that with ffmvforce so that's probably a dead-end.

Steps to Reproduce

1. Load the driver
2. Launch Assetto Corsa Rally, pick the Citroen Xsara and Vallee de Munster Descente on a time trial.
3. Observe the first big FFB moment in the tight 2-left at the top of the hill
4. Observe that shortly before the hairpin at around 1m50s 3-4Nm jabs to the right happen that do not have any basis in what is happening in game

Kernel Version

6.14

Distribution

Debian Trixie

Installation Method

DKMS

dmesg Output

Current sysfs Values (if relevant)

Game/Application

Assetto Corsa Rally 0.5.0 (Proton)

USB capture (recommended for FFB / wheel-state bugs)

No response

Checklist

• I installed the driver via setup.sh (or dkms-update.sh)
• I verified the driver loaded (lsmod | grep hid_logitech_dd)
• I checked dmesg for errors

Additional Context

No response

[matthiasvegh]

Since I started capturing, the issue seems to be a bit wayward. I have a 150MB capture that contains the issue, but not one smaller yet.
If the script could be told to save the last 60s instead of the next 60s, that'd be quite handy. I'd trim the pcap myself, but then I'd find it hard to reason about which values are broken.

I'll keep trying in the meantime, if I get a short capture with the bug I'll post it.

[mescon]

Thanks for keeping at it, and sorry it went wayward on you.

The "keep the last N seconds" mode you asked for just landed on master (PR #32). Pull it and capture with the ring buffer:

git pull
sudo ./tools/linux_game_capture.sh ffbspike ring:90

It records continuously into a small rolling buffer and keeps roughly the last 90 seconds. Play until you feel a spike, then press Ctrl-C right after it happens and it stitches the surviving window into a single pcap. Bump ring:90 higher if you want a longer tail (minimum is 10).

That should give us a short, focused capture instead of the 150 MB one.

If the spike stays elusive and the only capture that caught it is still the big one, that is fine too. Please don't trim it yourself. If you can drop the full 150 MB pcap somewhere I can pull it from (it is over GitHub's attachment limit), I will pick out the broken values from the protocol side. Either the short ring capture or the full one works.

Thanks again for the legwork on this one.

[mescon]

Gentle nudge now that v0.16.2 is out: the ring-buffer capture mode you asked for ships in the release (and the settings apps grew a live input monitor that makes the spikes easy to eyeball while you reproduce). If Assetto Corsa Rally still shows the spikes on 0.16.2, a ring capture of one spike moment is exactly what this needs - and if the recent releases happen to have changed the behavior, that is useful to know too.