On 21/04/15 11:22, Jan Beulich wrote:
Despite both pictures saying micro-seconds at the respective axis (and hence the problem not being _as bad_) - did the data collection reveal where these IRQ disable regions are, so we could look into eliminating them? (ISTR there being some open coded IRQ-disable, lock, unlock, IRQ-restore/enable sequences, but I'm not sure whether they got eliminated already.) Jan
The location of the locks was not collected, (Other then if it was from a spinlock or not) but I'm certainly considering doing this, as another graph of duration against cumulative time, shows the majority of time spent with interrupts disabled, is during code that keeps it disabled for long contiguous periods, all using raw (not spinlock) Interrupt disable. (A few occurrences, but very long when they happen) It would certainly be interesting to know where these come from.
-jenny _______________________________________________ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
