* Two simple fixes to suspension issues seen in GCPerfSim:
- Insert allow_fgc() call in background_mark_simple - this fixes the cases where there are a ton of GC handles referencing simple objects not containing pointers.
- Change PING_JIT_TIMEOUT constant from 10 milliseconds to 1 millisecond. This fixes the case where return address hijacking doesn't work quickly, because the hijacked thread doesn't return (e.g. because it's in a loop doing further calls). In this case we have to retry the hijack, and changing the timeout constant makes this happen more quickly.
Commit migrated from https://github.com/dotnet/coreclr/commit/
fab7aa20dc4c5856bb4759bbbde0bb88ffd439a2
background_mark_simple1 (o THREAD_NUMBER_ARG);
}
}
+ allow_fgc();
}
}
// Every PING_JIT_TIMEOUT ms, check to see if a thread in JITted code has wandered
// into some fully interruptible code (or should have a different hijack to improve
// our chances of snagging it at a safe spot).
-#define PING_JIT_TIMEOUT 10
+#define PING_JIT_TIMEOUT 1
// When we find a thread in a spot that's not safe to abort -- how long to wait before
// we try again.