aboutsummaryrefslogtreecommitdiffstats
path: root/litmus
Commit message (Collapse)AuthorAge
* Bugfix: Fix FMLP priority inheritance under GSN-EDFHEADmasterBjoern B. Brandenburg2009-10-05
| | | | | | | | | | | | | | | This fixes a regression introduced by the RTSS'09 improvements. Just calling unlink() on any task is not safe: it might reside in a release heap. This was done in the case that a new blocker increases a lock holder's priority. To avoid crashing in this case, the FMLP code was changed to be more careful when propagating an inherited priority. The code now detects three different scenarios and uses heap_decrease() to update the position of the lock holder's heap node. This fixes a simple test case.
* binomial heap: support decrease() operationBjoern B. Brandenburg2009-10-05
| | | | | | | Used to adjust position of heap node after the key of the node was decreased (= made more important). Cannot be used if key was increased (= made less important).
* remove concurrent heap from mainlineBjoern B. Brandenburg2009-09-10
|
* cleanup trailing whitespaceBjoern B. Brandenburg2009-09-10
|
* Remove non-mainline G-EDF pluginsBjoern B. Brandenburg2009-09-10
| | | | | This removes the plugins created for RTSS'09 that we don't want to track in mainline Linux.
* EDF plugins: don't forget to preempt new tasks on release-masterBjoern B. Brandenburg2009-05-04
| | | | | This fixes a bug that lead to task stacks being used s simultaneously on two CPUs.
* litmus core: cleanupBjoern B. Brandenburg2009-05-04
| | | | Line breaks, ammend comments, etc.
* G-EDF: always mark scheduled tasks as being so (bugfix)Bjoern B. Brandenburg2009-05-04
| | | | | | If schedule() was invoked twice in a retry, then the scheduled task might have not been "picked" properly because it was still scheduled.
* rt domain: remove unused debugging codeBjoern B. Brandenburg2009-05-04
|
* rt domain: evil hack to avoid recursive hrtimer lock acquisitionBjoern B. Brandenburg2009-05-04
| | | | | | | | This is a really dirty hack, but works for now since our real-time tasks do not use hrtimer_wakeup(). The need for this will go away when we port since newer versions of the Linux kernel allow timers to be activated directly in the scheduler.
* Merge branch 'master' of ↵Bjoern B. Brandenburg2009-05-03
|\ | | | | | | ssh://cvs.cs.unc.edu/cvs/proj/litmus/repo/papers/RTSS09/litmus2008 into rtss09
| * GQ-EDF: remove link completely (bugfix)Bjoern B. Brandenburg2009-05-03
| | | | | | | | Fixes fa940ad71184dec3d156b7dd673ee38067e7c0f6.
* | rt domain: fix rare deadlock between timer base lock and release lockBjoern B. Brandenburg2009-05-03
|/ | | | | | If a timer happens to go off at the wrong time on the wrong CPU, the old code could deadlock. This avoided in the new version by dropping the release_lock before using the hrtimer API.
* GQ-EDF: don't blindly overwrite entry->linked (bugfix)Bjoern B. Brandenburg2009-05-03
| | | | | Check if an already-linked task must be re-inserted into the ready queue.
* rt domain: make sure release heap is not in use before re-initBjoern B. Brandenburg2009-05-03
| | | | | | | | This fixes a bug observed under G-EDF: A task with extremely low wcet budget could get scheduled and re-added to its own release heap while the heap was still in use in the same timer callback that released the task. Havoc, predictably, ensued.
* rt domain: avoid use-after-free and re-init-while-in-use bugsBjoern B. Brandenburg2009-05-03
| | | | | | Don't just blindly overwrite the timers if they could be still in use. The the use-after-free bug was observed under Qemu.
* GQ-EDF: don't put running tasks in to released queue (bugfix)Bjoern B. Brandenburg2009-05-02
|
* litmus core: report ready tasks to user spaceBjoern B. Brandenburg2009-05-02
|
* fix layout of Litmus config menuBjoern B. Brandenburg2009-05-02
|
* litmus core: add dump-trace-buffer magic sysrqBjoern B. Brandenburg2009-05-02
| | | | Pump stuff from the TRACE() buffer to printk() on demand.
* rt domain: don't arm the same release heap timer twice (bugfix)Bjoern B. Brandenburg2009-05-02
| | | | | If the timeout was realy close and the timing unfortunate, the timer callback got invoked twice, with oopsing consequences.
* litmus core: un-inline binomial heap implementationBjoern B. Brandenburg2009-05-02
| | | | | | Those functions are pretty huge, and used in a couple of plugins. This should save a lot of space, make debugging with objdump much easier, and maybe it even has some I$ benefit.
* litmus core: fix sysrq registrationBjoern B. Brandenburg2009-05-02
| | | | The letter 'q' is already taken in 2.6.24.
* GSN-EDF: add release master supportBjoern B. Brandenburg2009-04-29
|
* release master support: fix new job codeBjoern B. Brandenburg2009-04-29
|
* add support for tracing IPI latencyBjoern B. Brandenburg2009-04-28
|
* GHQ-EDF: first shot at hierarchical queues implementationBjoern B. Brandenburg2009-04-27
| | | | seems to run in QEMU
* GQ-EDF: release master support.Bjoern B. Brandenburg2009-04-24
| | | | Basically same as G-EDF.
* G-EDF: release master support bugfixBjoern B. Brandenburg2009-04-24
| | | | Don't relink to release master.
* litmus config: make itpossible to compile without overhead tracingBjoern B. Brandenburg2009-04-24
| | | | Factor out ftdev support from overhead tracing.
* G-EDF: make use of release_master interfaceBjoern B. Brandenburg2009-04-23
| | | | | | The current release_master setting is copied when activating the plugin. Changing it during execution is currently not supported.
* edf common: fix debugging outputBjoern B. Brandenburg2009-04-23
|
* litmus proc: add release_master interfaceBjoern B. Brandenburg2009-04-23
| | | | | Let user space set which CPU should handle releases. It is up to plugins to make use of this information.
* fixup commit: missing rtdomain piecesBjoern B. Brandenburg2009-04-23
|
* rt domain: add support for release masterBjoern B. Brandenburg2009-04-22
| | | | | A release master is a CPU that takes all timer interrupts for release of a given rt_domain_t. By default off.
* litmus core: reduce lenth of interrups off in tick_no_rqlock()Bjoern B. Brandenburg2009-04-22
| | | | this was never strictly required, so we might as well avoid it
* litmus core: provide central NO_CPU definitionBjoern B. Brandenburg2009-04-22
| | | | this removes a blatant case of copy&paste reuse
* G-EDF: compiles & runs in QEMUBjoern B. Brandenburg2009-04-19
|
* concurrent heap: disable lockdepBjoern B. Brandenburg2009-04-19
| | | | | | Lockdep doesn't like the heap's locking pattern. We'd have to allocate a locking class for each cheap_node, which is a bit excessive memory-wise.
* GQ-EDF: avoid compile warning w/ CONFIG_SCHED_DEBUG_TRACEBjoern B. Brandenburg2009-04-19
| | | | Variable declarations go first.
* GSN-EDF: remove old and wrong commentBjoern B. Brandenburg2009-04-19
|
* Merge commit 'origin/master' into rtss09Bjoern B. Brandenburg2009-04-18
|\
| * Merge branch 'master' of ssh://cvs/cvs/proj/litmus/repo/litmus2008Bjoern B. Brandenburg2009-04-18
| |\
| * | litmus core: remove no longer used sorted lists codeBjoern B. Brandenburg2009-04-18
| | | | | | | | | | | | No plugin uses it anymore.
* | | litmus core: add concurrent heap impl.Bjoern B. Brandenburg2009-04-18
| | |
* | | add G-EDF, a version of GSN-EDF with synchronization support ripped outBjoern B. Brandenburg2009-04-15
| | | | | | | | | | | | a good base version for other schedulers
* | | GQ-EDF: merge heaps if debug not enabledBjoern B. Brandenburg2009-04-15
| | | | | | | | | | | | | | | | | | Doing it task-by-task really only makes sense if we are going to trace every merged task in the debug trace.
* | | litmus core: fix CPU time accountingBjoern B. Brandenburg2009-04-11
| | | | | | | | | | | | | | | Tell Linux how much time each RT task consumed. This fixes the behavior of CPU timers and tools like top.
* | | make CONFIG_SCHED_TASK_TRACE compile on sparc64Bjoern B. Brandenburg2009-04-11
| | | | | | | | | | | | It won't compile with missing prototypes.
* | | rt_domain: use per-event timersBjoern B. Brandenburg2009-04-09
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Don't multiplex one timer among all release heaps. The hrtimer subsystem can handle many timers and is heavily optimize; make use of this fact. This also greatly simplifies the actual callback, which should help to bring down release overheads. This also saves memory as we do not need to maintain a separate heap of (release) heaps.
generated by cgit v1.2.2 (git 2.25.0) at 2025-04-05 13:43:06 -0400