On both 3.0-rt and 3.2-rt I ran a stress test on CPU hotplug and they both failed miserably. These bugs also show themselves on suspend. After spending 5 long days debugging this, and finding lots of bugs I have a patch series that makes the CPU hot plugging rather stable. The first few patches fix other broken RT features that were discovered while debugging hotplug. Patch 7 is a rework of how RT handles taking down a CPU. There were several corner cases that the original approach failed on, and this rework seems to cover them now. Patch 8 and 9 revert the rework that was done on workqueue for RT. Yes, letting workqueues play games with THREAD_BOUND is bad, but the fixes introduced several more races that were very hard to fix. Every fix seemed to produce a new race, and it was a bad game of cat and mouse running around a goose hunt whacking moles! Since the rework of the CPU down code no longer had issues with workqueue messing with THREAD_BOUND, reverting the patches that did the rework of workqueues fixed the bugs that I was hitting. Lets fix CPU hotplug today, and leave the redesign of workqueues for another day. Comments, -- Steve -- To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
