Hello, Roland McGrath, le Sat 11 Apr 2009 18:47:13 -0700, a écrit : > vm_read should return a normal out-of-line page there--it's just an RPC > stub. So yes, there should be a mapping there. Unless there is a race > with a deallocation in another thread or exec, or something, that seems > like a potential microkernel bug (unless it's a mig stub bug or something).
For the record, I've eventually found the issue: exec calls finish() several times, but it wasn't idempotent, and that would double-free the map buffer. With bad luck, an allocation would happen in between. Samuel
