Messages in this thread | | | Date | Fri, 20 Nov 2009 16:19:48 -0800 | From | "H. Peter Anvin" <> | Subject | Re: [RFC PATCH 2/6] jump label v3 - x86: Introduce generic jump patching without stop_machine |
| |
On 11/20/2009 04:06 PM, Masami Hiramatsu wrote: > > If we don't care about NMI, we can use stop_machine() (for > this reason, kprobe-jump-optimization can use stop_machine(), > because kprobes can't probe NMI code), but tracepoint has > to support NMI. >
Ingo pointed out that the NMI issue can be dealt with by having the NMI handler check a flag if code modification is in progress on the entry to the NMI handler. So yes, NMI can trap out of the IPI hold, but it will not go any further.
-hpa
| |