Hello
[1.] One line summary of the problem:
Race between upload and playback in ff-memless
[2.] Full description of the problem/report:
If update of already playing force effect is quickly followed by effect stop in ff-memless based driver, the hw force might be not stopped.
What I think is going on: When ml_ff_upload processes started effect, it will clear the FF_EFFECT_PLAYING bit and schedule timer. If ml_ff_playback is called to stop the effect before the timer runs, it will see cleared bit and directly clear the FF_EFFECT_STARTED bit instead of doing the full abort so the hw force is not updated unless there is another effect of the same type.
[3.] Keywords (i.e., modules, networking, kernel):
input, force feedback, ff-memless
[4.] Kernel information
[5.] Most recent kernel version which did not have the bug:
Looking at the source code it seems to be present in all versions of ff-memless since its introduction.
[7.] A small shell script or example program which triggers the
problem (if possible)
https://pastebin.com/yaQTXgk6
With regards
Jiri Dvorak
