Andi Kleen wrote: > Zachary Amsden <zach at vmware.com> writes: > >> That is a really nasty problem. You need a synchronization primitive >> which guarantees a flat stack, so you can't do it in the interrupt >> handler as I have tried to do. I'll bang my head on it awhile. In >> the meantime, were there ever any solutions to the syscall patching >> problem that might lend me a clue as to what to do (or not to do, or >> impossible?). >> > > stop_machine_run() solves the problem I think. It is currently not > exported though. I don't think there's anything in there that couldn't > be reimplemented in a module, but then we could also just export it > if there's a useful user. > Well, I don't think anything is sufficient for a preemptible kernel. I think that's just plain not going to work. You could have a kernel thread that got preempted in a paravirt-op patch point, and making all the patch points non-preempt is probably a non-starter (either +12 bytes each or no native inlining). Finding out after the fact that you have a kernel thread that was preempted in a patch point is very hard work, but it is possible. The fixing it up is where you need to take liberties with reality. stop_machine_run() is almost what I want, but even that is not sufficient. You also need to disable NMIs and debug traps, which is pretty hairy, but doable. The problem with stop_machine_run() is that I don't just want the kernel to halt running on remote CPUs, I want the kernel on all CPUs to actually do something simultaneously - the entry into paravirt mode requires a hypervisor call on each CPU, and stop_machine() doesn't provide a facility to fire a callback on each CPU from the stopmachine state. Since this code is so rather, um, custom, I was going to reimplement stop_machine in the module. Zach
