Messages in this thread | | | Date | Tue, 9 Nov 2004 17:05:44 +0100 | From | Ingo Molnar <> | Subject | [patch] Real-Time Preemption, -RT-2.6.10-rc1-mm3-V0.7.23 |
| |
i have released the -V0.7.23 Real-Time Preemption patch, which can be downloaded from the usual place:
http://redhat.com/~mingo/realtime-preempt/
this release includes fixes, development/debugging improvements and latency improvements and other speedups.
the biggest change is the reworked timing/tracing framework. Wakeup timing became a compile-time thing and can be selected independently of the preemption mode - i.e. it can now be used on near-vanilla !PREEMPT kernels too, providing good wakeup-latency comparison of the various preemption models.
irqs-off and preempt-off critical section timing/tracing can be selected if wakeup timing is disabled, the two options can be selected separately or together as well.
another improvement is that wakeup-timing now works correctly on SMP too - the tracer 'follows' the highest-priority task in the system as it gets bounced between CPUs and always traces the CPU where the task is pending.
other changes since -V0.7.22:
- semaphore livelock fix: feedback from Mark H. Johnson pinpointed a bug in the down_trylock() semaphore code: if preempted in the wrong moment a lower-prio task could cause a higher-prio task to livelock indefinitely. This fix could solve the 'keventd looping' problem reported by Mark.
the fix is to make the down_trylock() code a bit simpler, but this also introduces the potential for down_trylock() to get 'spurious' locking-rejects. Hopefully this wont be a big problem - we dont really guarantee that down_trylock() succeeds - but code using higher semaphore counts to track resources could be negatively impacted by this. We'll see.
- console assert fix: implemented a different type of fbcon RT-preemption handling variant, this could solve the assert reported by Amit Shah.
- debugging improvement: implemented a sequence counter for max latency traces. This has the advantage of solving the 'slow console on SMP problem': the latency-timing code used to get confused by another CPU printing a timing message to a slow console and thus delaying that other CPU. Now any latency that gets generated while a maximum is printed is skipped.
- further shrunk the non-debug size of struct rt_mutex by moving the save_state logic into the debug paths - size is now 4 machine-words.
- fixed CONFIG_HIGHMEM latencies: all atomic-kmap APIs are now wrapped seemlessly and in a preemptible way.
- implemented an IO-APIC register cache to speed up the IRQ-redirection latency hotpath. Also, made the POST flush a bit faster.
- disable KGDB if PREEMPT_RT - it's broken for now.
- move some runtime checks to under DEBUG_PREEMPT - this speeds up CRITICAL_PREEMPT_TIMING.
to create a -V0.7.23 tree from scratch, the patching order is:
http://kernel.org/pub/linux/kernel/v2.6/linux-2.6.9.tar.bz2 http://kernel.org/pub/linux/kernel/v2.6/testing/patch-2.6.10-rc1.bz2 http://kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.10-rc1/2.6.10-rc1-mm3/2.6.10-rc1-mm3.bz2 http://redhat.com/~mingo/realtime-preempt/realtime-preempt-2.6.10-rc1-mm3-V0.7.23
Ingo - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |