On Wed Feb 18 13:01:37 2026 +0000, Elizabeth Figura wrote:
I really, really don't think we should be adding memory barriers without understanding the problem they're solving. At worst you've just made the problem slightly less frequent, as apparently here. I agree. The barrier is only a workaround and I don’t want to push it as a real fix without understanding the root cause.
Unfortunately I haven’t been able to reliably diagnose where the USER lock is being held. Traditional debugging/backtraces are difficult here due to the mixed Wine PE + Android ELF + JVM environment (most stack traces are unusable). Do you have any advice on a good way to track down where the USER lock is acquired/held in this setup, or any preferred instrumentation approach you’d recommend? -- https://gitlab.winehq.org/wine/wine/-/merge_requests/9874#note_129878