aboutsummaryrefslogtreecommitdiffstats
Commit message (Collapse)AuthorAge
* Fix locking bug in GSN-EDF.RTSS07Bjoern B. Brandenburg2007-05-18
|
* Fix partition handling of PSN-EDF.Bjoern B. Brandenburg2007-05-18
|
* Make PI semaphores under PSN-EDF non-preemptable.Bjoern B. Brandenburg2007-05-17
|
* Let PI semaphores inherit only inside a partition.Bjoern B. Brandenburg2007-05-17
| | | | | - Track hp.cpu_task per partition - inherit priority only from local tasks
* Prepare struct pi_semaphore for per-CPU PI protocol as required by FMLP.Bjoern B. Brandenburg2007-05-17
|
* Restructure PI semaphores so that they are easier to trace.Bjoern B. Brandenburg2007-05-17
|
* Add timestamps to semaphore functionsBjoern B. Brandenburg2007-05-17
|
* Don't disable interrupts in semaphore syscalls.Bjoern B. Brandenburg2007-05-17
| | | | | We don't need to protect against interrupts in the semaphore functions since semaphores are never touched from interrupt context.
* Fix compiler warnings.Bjoern B. Brandenburg2007-05-17
|
* Make the inclusion of sched_trace_XXX() and TRACE() a configurable option.Bjoern B. Brandenburg2007-05-17
| | | | This allows us to remove the debugging code from benchmarking kernels.
* Fix a compiler warning.Bjoern B. Brandenburg2007-05-17
|
* Insert trigger for enter_np()/exit_np() tracing.Bjoern B. Brandenburg2007-05-16
|
* Make sched_trace conditional.Bjoern B. Brandenburg2007-05-16
| | | | TODO: Make it a .config option.
* Place trigger for scheduling overhead benchmark.Bjoern B. Brandenburg2007-05-16
|
* Add time stamps for schedule(), the timer interrupt, and the pluginBjoern B. Brandenburg2007-05-16
| | | | scheduler_tick() and schedule() callbacks.
* Change feather-trace device to support enabling and disabling of events.Bjoern B. Brandenburg2007-05-16
|
* Merged Feather-Trace.Bjoern B. Brandenburg2007-05-15
|
* Activate ceiling blocking and fix a stupid list corruption bug.Bjoern B. Brandenburg2007-05-15
|
* Reworked SRP implementation.Bjoern Brandenburg2007-05-15
|
* PSN-EDF priority inheritance implementation.Bjoern B. Brandenburg2007-05-15
|
* Move edf_set_hp_task() to litmus_sem.c as it needs to have knowledge ofBjoern B. Brandenburg2007-05-15
| | | | what is stored in the wait queue's ->private field.
* PI implementation for sched_gsn_edf.cBjoern B. Brandenburg2007-05-15
|
* Fixed several bugs in the semaphore code concerning racing wake ups + miscBjoern B. Brandenburg2007-05-15
| | | | fixex.
* Fix try_to_wake_up() for RT tasks if it races with schedule.Bjoern B. Brandenburg2007-05-15
| | | | | We need to set the task state to TASK_RUNNING if it races, otherwise the task might get lost once it gets de-scheduled.
* First part of the PI semaphore implementation.Bjoern B. Brandenburg2007-05-14
| | | | | | | | | | | - move struct pi_semaphore to place where it makes more sense (not included everywhere) - change semantics of scheduler plugin callbacks and add comments - remove old unneeded code - compile fixes Note: The plugins don't actually work yet, since the semantics of the callbacks have changed. That will be fixed in the next patch.
* Move common code to edf_common.c.Bjoern B. Brandenburg2007-05-14
| | | | | Finding the highest-priority task in a pi_semaphore is same in all EDF-based plugins.
* Fix gsnedf_sleep_next_period() locking.Bjoern B. Brandenburg2007-05-14
|
* Fix GSN-EDF sleep_next_periodBjoern B. Brandenburg2007-05-14
| | | | | The standard EDF implementation does not work for GSN-EDF since it does not perform the task preparation in schedule().
* Reimplemented LITMUS standard FIFO semaphores + misc fixesBjoern B. Brandenburg2007-05-14
| | | | | | | This version is much more straightforward than the last one and actually enforces FIFO order in all cases. Further, it removes the need for the sem_wake_up() calls by providing a custom wake up function for the wait queue.
* Move specialized PI semaphore code into litmus_sem.c + small fixBjoern B. Brandenburg2007-05-14
| | | | | | | - Let's keep all the code belonging to LITMUS in one place, this makes maintenance and porting much easier. - add_wait_queue_exlclusive_locked() sets the exclusive flag anyway
* Start of LITMUS semaphore, PI, and SRP rewrite.Bjoern B. Brandenburg2007-05-14
| | | | | Rearrange the code to get out of the architecture dependent areas of the kernel and make it live in a central file, kernel/litmus_sem.c.
* Remove old, unneeded ASM cruft.Bjoern B. Brandenburg2007-05-14
|
* Initial implementation of inherit and return priority for PSN-EDF.Bjoern B. Brandenburg2007-05-13
|
* Streamline PI implementation a little.Bjoern B. Brandenburg2007-05-13
|
* Kick out commented ASM code.Bjoern B. Brandenburg2007-05-13
|
* Added psnedf_handle_priority_change() implementationBjoern B. Brandenburg2007-05-13
|
* NP section backport and debug fixesBjoern B. Brandenburg2007-05-13
| | | | | | - port NP section logic from PSN-EDF to GSN-EDF - remove old debugging stuff from GSN-EDF - add some tracing info
* Add macros for convenient task tracing.Bjoern B. Brandenburg2007-05-13
|
* Add a whole bunch of TRACEs to facilitate debugging.Bjoern B. Brandenburg2007-05-13
|
* TRACE RT task arrivals and departures.Bjoern B. Brandenburg2007-05-13
|
* Improve debugging output and supportBjoern B. Brandenburg2007-05-13
| | | | | | - TRACE jobs being queued - TRACE enter_np/exit_np - fix double assignment of F in sysrq handlers
* Pay attention to (preempt_count() & PREEMPT_ACTIVE) hack.Bjoern B. Brandenburg2007-05-13
| | | | | | | | | | This change should fix the long standing problem that certain IO intensive work loads would "crash" the kernel. What really happened was that real-time tasks caught an interrupt right before they would have called schedule as a part of a suspension anyway. As they were already in a different state then TASK_RUNNING, they were not requeued and got lost. If they were holding important locks (such as a TTY lock or the BKL), then "bad things" happened...
* Use volatile access and relax_cpu() in queue lock.Bjoern B. Brandenburg2007-05-13
|
* PSN-EDF fixes.Bjoern B. Brandenburg2007-05-10
| | | | | | - fix recursive lock acquisition of cpu state lock, found by deplock - properly guard cpu state during mode change - cosmetic fix in initialization
* Pull in John's SRP fixes.Bjoern B. Brandenburg2007-05-10
|\ | | | | | | Merge with git+ssh://cvs/cvs/proj/litmus/repo/litmus
| * Modified SRP semaphores so that waiting for prio to exceed SPC worksJohn M. Calandrino2007-05-10
| | | | | | | | correctly, mostly by adding waitqueues and rewriting wait_until_exceed_spc.
* | Add PSN-EDF.Bjoern B. Brandenburg2007-05-10
| |
* | Minor fixes.Bjoern B. Brandenburg2007-05-10
| |
* | Break out __release_pending() from try_release_pending()Bjoern B. Brandenburg2007-05-10
| |
* | Some random code cleanup.Bjoern B. Brandenburg2007-05-10
| | | | | | | | | | | | - IRQs are disabled in _schedule - use TRACE - coding style