By: Paul A. Clayton (paaronclayton.delete@this.gmail.com), July 16, 2015 6:23 pm
Room: Moderated Discussions
Michael S (already5chosen.delete@this.yahoo.com) on July 16, 2015 4:31 pm wrote:
> Consider interrupt followed by task preemption.
> And when preempted task is finally resumed, possibly a months later (you hibernated
> and left for holidays) it can be running on different core/socket.
Under extreme conditions, the out-of-order work could be ignored (this assumes the copy does not overwrite data needed for safe restarting, including that page mappings are not strangely duplicated). While crossing core/socket introduces issues, identifying the thread even without changing cores seems problematic. Page table base address and copy parameters may not be enough to uniquely identify a copy.
> Consider interrupt followed by task preemption.
> And when preempted task is finally resumed, possibly a months later (you hibernated
> and left for holidays) it can be running on different core/socket.
Under extreme conditions, the out-of-order work could be ignored (this assumes the copy does not overwrite data needed for safe restarting, including that page mappings are not strangely duplicated). While crossing core/socket introduces issues, identifying the thread even without changing cores seems problematic. Page table base address and copy parameters may not be enough to uniquely identify a copy.
