The SMP EDF scheduler supports one-to-one and one-to-all thread to
processor affinities. The one-to-one thread to processor affinity
introduces a constraint on the ordering of threads. The implementation
uses one ready queue for threads which have a one-to-all affinity and
one for each one-to-one affinity group. To order threads across the
ready queues, a generation number is used. However, the approach to
update the generation number each time a thread is inserted into a ready
queue was wrong. The generation number needs to be updated only in the
enqueue and enqueue scheduled operations where an insert priority is
available. The scheduled chain needs to take the generation number into
account.
An example scenario which shows the bug is this. Let T be a high
priority task affine to processor X. Let A be a lower priority task
affine to processor X. Let B be a lower priority task with no affinity
to a particular processor which executes on processor Y. Let B be in
the same priority group than A and after A. Let T set the affinity to
all processors. Now A (higher priority relative to B) should execute on
X and T (high priority) should execute on Y.
Close#4534.
This patch fixes an issue with the idle thread handling in the SMP
scheduler framework used for the MrsP locking protocol. The approach to
use a simple chain of unused idle threads is broken for schedulers which
support thread to processor affinity. The reason is that the thread to
processor affinity introduces another ordering indicator which may under
certain conditions lead to a reordering of idle threads in the scheduled
chain. This reordering is not propagated to the chain of unused idle
threads. This could lead to use an idle thread for a sticky scheduler
node which is already in use. This locks up the system in infinite
loops in the thread context switch procedure.
To fix this, the SMP scheduler implementations must now provide
callbacks to get and release an unused idle thread.
Update #4531.
This patch fixes the following broken behaviour:
While a thread is scheduled on a helping scheduler, while it does not
own a MrsP semaphore, if it obtains a MrsP semaphore, then no
scheduler node using an idle thread and the ceiling priority of the
semaphore is unblocked for the home scheduler.
This could lead to priority inversion issues and is not in line
with the MrsP protocol.
Introduce two new scheduler operations which are only enabled if
RTEMS_SMP is defined. The operations are used to make the scheduler
node of the home scheduler sticky and to clean the sticky property.
This helps to keep the sticky handing out of the frequently used
priority update operation.
Close#4532.
If a node is moved from the scheduled chain to the ready queue, then we
know that it is the highest priority ready node. So, it can be
prepended to the ready queue without doing any comparisons.
Update #4531.
Rework the handling of the affine ready queue for the EDF SMP scheduler.
Do the queue handling in the node insert, move, and extract operations.
Remove the queue handling from _Scheduler_EDF_SMP_Allocate_processor().
Update #4531.
Insert nodes after moving the second node to reduce the items in the
data structure for the insert operation. This also avoids having two
nodes for the same processor inserted into the scheduled chain.
Update #4531.
Do not set the CPU of the idle thread in _Scheduler_Use_idle_thread(). This
helps to use _Scheduler_Try_to_schedule_node() under more general conditions in
the future, for example in case the owner and user of a node are not the same.
Update #4531.
If we block the executing thread and it is not the heir thread, then there is
no need to run the schedule operation. The scheduler already selected a new
heir.
These two OAR copyright headers are the only two in the codebase with
a format that differs from the typical OAR copyright header. This makes
all of the OAR copyright headers consistent.
Ensure when both RTEMS_DEBUG is specified and pointers are large that
enough space is allocated to accomodate the Per_CPU_Control structure.
This changes the calculation to be more compositional instead of trying
to list out every permutation of options possible.
These external functions rtems_scheduler_get_processor() and
rtems_scheduler_get_processor_maximum() which may be used by bindings
for languages other than C/C++.
Check all IO driver table configuration options which are used to initialize
_IO_Driver_address_table[]. Checks for the following settings were missing:
* CONFIGURE_BSP_PREREQUISITE_DRIVERS
* CONFIGURE_APPLICATION_PREREQUISITE_DRIVERS
* CONFIGURE_APPLICATION_NEEDS_WATCHDOG_DRIVER
* CONFIGURE_APPLICATION_EXTRA_DRIVERS
Update #3875.
Evaluate CONFIGURE_DISABLE_BSP_SETTINGS for each affected application
configuration option. This makes the code easier to review since the influence
of CONFIGURE_DISABLE_BSP_SETTINGS is locally visible in the code.
Update #3875.
Priority ceiling semaphores (in contrast to MrsP semaphores) have only
one priority ceiling which is defined for the home scheduler of the task
which created the semaphore. Always check during a semaphore seize that
the home scheduler of the calling task is equal to this scheduler.
This avoids a scheduler mismatch if the semphore is acquired after the
calling thread blocked on the semaphore.
This change is a slight performance optimization for systems with a slow
64-bit division.
The th->th_scale and th->th_large_delta values only depend on the
timecounter frequency and the th->th_adjustment. The timecounter
frequency of a timehand only changes when a new timecounter is activated
for the timehand. The th->th_adjustment is only changed by the NTP
second update. The NTP second update is not done for every call of
tc_windup().
Move the code block to recalculate the scaling factor and
the large delta of a timehand to the new helper function
recalculate_scaling_factor_and_large_delta().
Call recalculate_scaling_factor_and_large_delta() when a new timecounter
is activated and a NTP second update occurred.
MFC after: 1 week
Timecounter registration is dynamic, i.e., there is no requirement that
timecounters must be registered during single-threaded boot. Loadable
drivers may in principle register timecounters (which can be switched to
automatically). Timecounters cannot be unregistered, though this could
be implemented.
Registered timecounters belong to a global linked list. Add a mutex to
synchronize insertions and the traversals done by (mpsafe) sysctl
handlers. No functional change intended.
Reviewed by: imp, kib
MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D32511
FreeBSD Foundation sys/ copyrights
These ones were unambiguous cases where the Foundation was the only
listed copyright holder (in the associated license block).
Sponsored by: The FreeBSD Foundation
In FreeBSD, the current time is computed from uptime + boottime. Uptime
is a continuous, smooth function that's monotonically increasing. To
effect changes to the current time, boottime is adjusted. boottime is
mutable and shouldn't be cached against future need. Document the
current implementation, with the caveat that we may stop stepping
boottime on resume in the future and will step uptime instead (noted in
the commit message, but not in the code).
Sponsored by: Netflix
Reviewed by: phk, rpokala
Differential Revision: https://reviews.freebsd.org/D30116
