| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The current implementation of task migration in RTEMS has some
implications with respect to the interrupt latency. It is crucial to
preserve the system invariant that a task can execute on at most one
processor in the system at a time. This is accomplished with a boolean
indicator in the task context. The processor architecture specific
low-level task context switch code will mark that a task context is no
longer executing and waits that the heir context stopped execution
before it restores the heir context and resumes execution of the heir
task. So there is one point in time in which a processor is without a
task. This is essential to avoid cyclic dependencies in case multiple
tasks migrate at once. Otherwise some supervising entity is necessary to
prevent life-locks. Such a global supervisor would lead to scalability
problems so this approach is not used. Currently the thread dispatch is
performed with interrupts disabled. So in case the heir task is
currently executing on another processor then this prolongs the time of
disabled interrupts since one processor has to wait for another
processor to make progress.
It is difficult to avoid this issue with the interrupt latency since
interrupts normally store the context of the interrupted task on its
stack. In case a task is marked as not executing we must not use its
task stack to store such an interrupt context. We cannot use the heir
stack before it stopped execution on another processor. So if we enable
interrupts during this transition we have to provide an alternative task
independent stack for this time frame. This issue needs further
investigation.
|
|
|
|
|
| |
A default handler is not necessary. The test message sender must ensure
that a handler is installed.
|
| |
|
|
|
|
|
| |
This handler can be used to test the inter-processor interrupt
implementation.
|
|
|
|
|
| |
Avoid the SMP_FATAL_SCHEDULER_WITHOUT_PROCESSORS fatal error and make it
a run-time error in rtems_scheduler_ident() and _Scheduler_Get_by_id().
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Use "cpu" for an arbitrary Per_CPU_Control variable.
Use "cpu_self" for the Per_CPU_Control of the current processor.
Use "cpu_index" for an arbitrary processor index.
Use "cpu_index_self" for the processor index of the current processor.
Use "cpu_count" for the processor count obtained via
_SMP_Get_processor_count().
Use "cpu_max" for the processor maximum obtained by
rtems_configuration_get_maximum_processors().
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Clustered/partitioned scheduling helps to control the worst-case
latencies in the system. The goal is to reduce the amount of shared
state in the system and thus prevention of lock contention. Modern
multi-processor systems tend to have several layers of data and
instruction caches. With clustered/partitioned scheduling it is
possible to honour the cache topology of a system and thus avoid
expensive cache synchronization traffic.
We have clustered scheduling in case the set of processors of a system
is partitioned into non-empty pairwise-disjoint subsets. These subsets
are called clusters. Clusters with a cardinality of one are partitions.
Each cluster is owned by exactly one scheduler instance.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Add and use _CPU_SMP_Start_processor(). Add and use
_CPU_SMP_Finalize_initialization(). This makes most
_CPU_SMP_Initialize() functions a bit simpler since we can calculate the
minimum value of the count of processors requested by the application
configuration and the count of physically or virtually available
processors in the high-level code.
The CPU port has now the ability to signal a processor start failure.
With the support for clustered/partitioned scheduling the presence of
particular processors can be configured to be optional or mandatory.
There will be a fatal error only in case mandatory processors are not
present.
The CPU port may use a timeout to monitor the start of a processor.
|
| |
|
|
|
|
|
|
|
|
| |
Add _Debug_Is_thread_dispatching_allowed(). This makes it possible to
assert the opposite.
Use _ISR_Disable_without_giant()/_ISR_Enable_without_giant() to avoid
misleading secondary assertion failures.
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
Add a local context structure to the SMP lock API for acquire and
release pairs. This context can be used to store the ISR level and
profiling information. It may be later used to enable more
sophisticated lock algorithms, e.g. MCS locks.
There is only one lock that cannot be used with a local context. This
is the per-CPU lock since here we would have to transfer the local
context through a context switch which is very complicated.
|
|
|
|
| |
New test smptests/smpfatal03.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Rename _SMP_Request_other_cores_to_perform_first_context_switch() into
_SMP_Request_start_multitasking() since this requests now a multitasking
start on all configured and available processors. The name corresponds
_Thread_Start_multitasking() and
_SMP_Start_multitasking_on_secondary_processor() actions issued in
response to this request. Move in source file to right place.
Rename PER_CPU_STATE_READY_TO_BEGIN_MULTITASKING into
PER_CPU_STATE_READY_TO_START_MULTITASKING.
Rename PER_CPU_STATE_BEGIN_MULTITASKING into
PER_CPU_STATE_REQUEST_START_MULTITASKING.
Rename _SMP_Request_other_cores_to_shutdown() into
_SMP_Request_shutdown().
Add a per-CPU state lock to protect all changes. This was necessary to
offer a controlled shutdown of the system (atomic read/writes alone are
not sufficient for this kind of synchronization).
Add documentation for Per_CPU_State.
Delete debug output.
New tests smptests/smpfatal01 and smptests/smpfatal02.
|
|
|
|
|
| |
Inline _SMP_Inter_processor_interrupt_handler() to avoid function call
overhead. Remove debug output.
|
|
|
|
|
|
| |
Wait for per-CPU changes into PER_CPU_STATE_READY_TO_BEGIN_MULTITASKING
later. There is no need to delay the initialization of the main
processor at this point.
|
| |
|
| |
|
|
|
|
|
|
| |
Rename rtems_smp_process_interrupt() into
_SMP_Inter_processor_interrupt_handler(). Delete unused header file
<rtems/bspsmp.h>.
|
|
|
|
|
|
| |
Rename rtems_smp_secondary_cpu_initialize() into
_SMP_Start_multitasking_on_secondary_processor(). Move declaration to
<rtems/score/smpimpl.h>.
|
|
|
|
|
| |
Collect SMP implementation specific parts in the
<rtems/score/smpimpl.h> header file.
|
|
|
|
|
|
| |
Since the per-CPU SMP lock must be acquired and released to send the
message a single interrupt broadcast operations offers no benefits. If
synchronization is required, then a SMP barrier must be used anyway.
|
|
|
|
|
| |
Rename bsp_smp_initialize() into _CPU_SMP_Initialize() since every CPU
port must supply this function.
|
|
|
|
|
| |
Rename RTEMS_BSP_SMP_SHUTDOWN in SMP_MESSAGE_SHUTDOWN since SMP messages
have nothing to do with the BSP. Use UINT32_C() instead of casts.
|
|
|
|
|
|
| |
Use rtems_fatal() instead of _CPU_Fatal_halt() to shutdown processors in
SMP configurations since this allows intervention of BSP or application
specific fatal extensions.
|
|
|
|
|
|
| |
Move _SMP_Request_other_cores_to_shutdown() invocation from
rtems_shutdown_executive() to _Internal_error_Occurred() to allow a
proper shutdown on SMP configurations even in the error case.
|
|
|
|
|
|
|
|
|
|
|
|
| |
Do not return to BSP context in the exit() shutdown path. This makes it
possible to re-use the initialization stack. It can be used for the
interrupt stack for example. On targets with a small RAM this is a
considerable benefit.
This change eliminates also some special cases and simplifies the code.
Delete _Thread_Set_global_exit_status(),
_Thread_Get_global_exit_status() and _Thread_Stop_multitasking().
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Add and use _ISR_Disable_without_giant() and
_ISR_Enable_without_giant() if RTEMS_SMP is defined.
On single processor systems the ISR disable/enable was the big hammer
which ensured system-wide mutual exclusion. On SMP configurations this
no longer works since other processors do not care about disabled
interrupts on this processor and continue to execute freely.
On SMP in addition to ISR disable/enable an SMP lock must be used.
Currently we have only the Giant lock so we can check easily that ISR
disable/enable is used only in the right context.
|
| |
|
| |
|
|
|
|
|
|
|
| |
Rename _Per_CPU_Lock_acquire() to _Per_CPU_ISR_disable_and_acquire().
Rename _Per_CPU_Lock_release() to _Per_CPU_Release_and_ISR_enable().
Add _Per_CPU_Acquire() and _Per_CPU_Release().
|
|
|
|
|
|
|
|
|
|
| |
Add context parameter to _Thread_Start_multitasking() and use this
function in rtems_smp_secondary_cpu_initialize(). This avoids
duplication of code.
Fix missing floating point context initialization in
rtems_smp_secondary_cpu_initialize(). Now performed via
_Thread_Start_multitasking().
|
|
|
|
|
| |
Use an event triggered unicast to inform remote processors about a
necessary thread dispatch instead.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Delete _ISR_Enable_on_this_core(), _ISR_Flash_on_this_core(),
_ISR_SMP_Disable(), _ISR_SMP_Enable(), _ISR_SMP_Flash().
The ISR disable/enable interface has no parameter to pass a specific
object. Thus it is only possible to implement a single global lock
object with this interface. Using the ISR disable/enable as the giant
lock on SMP configurations is not feasible.
Potentially blocking resource obtain sequences protected by the thread
dispatch disable level are subdivided into smaller ISR disabled critical
sections. This works since on single processor configurations there is
only one thread of execution that can block. On SMP this is different
(image a mutex obtained concurrently by different threads on different
processors).
The thread dispatch disable level is currently used as the giant lock.
There is not need to complicate things with this unused interface.
|
|
|
|
|
|
| |
Add and use _Per_CPU_Get_by_index() and _Per_CPU_Get_index(). Add
_Per_CPU_Send_interrupt(). This avoids direct access of
_Per_CPU_Information.
|
|
|
|
|
|
|
|
| |
Move implementation specific parts of thread.h and thread.inl into new
header file threadimpl.h. The thread.h contains now only the
application visible API.
Remove superfluous header file includes from various files.
|
|
|
|
| |
Delete bsp_smp_interrupt_cpu().
|
|
|
|
|
|
|
|
|
|
| |
Add and use _SMP_Get_current_processor() and
rtems_smp_get_current_processor().
Delete bsp_smp_interrupt_cpu().
Change type of current processor index from int to uint32_t to match
_SMP_Processor_count type.
|
|
|
|
|
| |
The thread dispatch is a side-effect of interrupt processing, thus there
is no need to send an explicit message.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The new Simple SMP Scheduler allocates a processor for the processor
count highest priority ready threads. The thread priority and position
in the ready chain are the only information to determine the scheduling
decision. Threads with an allocated processor are in the scheduled
chain. After initialization the scheduled chain has exactly processor
count nodes. Each processor has exactly one allocated thread after
initialization. All enqueue and extract operations may exchange threads
with the scheduled chain. One thread will be added and another will be
removed. The scheduled and ready chain is ordered according to the
thread priority order. The chain insert operations are O(count of ready
threads), thus this scheduler is unsuitable for most real-time
applications.
The thread preempt mode will be ignored.
|
|
|
|
|
|
|
|
|
| |
Rename in rtems_smp_get_processor_count(). Always provide
<rtems/score/smp.h> and <rtems/rtems/smp.h>. Add
_SMP_Get_processor_count(). This function will be a compile time
constant defined to be one on uni-processor configurations. This allows
iterations over all processors without overhead on uni-processor
configurations.
|
|
|
|
| |
Add and use _Per_CPU_Lock_release().
|
| |
|
|
|
|
|
|
|
| |
Move thread dispatch declarations and inline functions to new header
<rtems/score/threaddispatch.h> to make it independent of the
Thread_Control structure. This avoids a cyclic dependency in case
thread dispatch functions are used for the object implementation.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Move the SMP lock implementation to the CPU port. An optimal SMP lock
implementation is highly architecture dependent. For example the memory
models may be fundamentally different.
The new SMP lock API has a flaw. It does not provide the ability to use
a local context for acquire and release pairs. Such a context is
necessary to implement for example the Mellor-Crummey and Scott (MCS)
locks. The SMP lock is currently used in _Thread_Disable_dispatch() and
_Thread_Enable_dispatch() and makes them to a giant lock acquire and
release. Since these functions do not pass state information via a
local context there is currently no use case for such a feature.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Delete bsp_smp_wait_for(). Other parts of the system work without
timeout, e.g. the spinlocks. Using a timeout here does not make the
system more robust.
Delete bsp_smp_cpu_state and replace it with Per_CPU_State. The
Per_CPU_State follows the Score naming conventions. Add
_Per_CPU_Change_state() and _Per_CPU_Wait_for_state() functions to
change and observe states.
Use Per_CPU_State in Per_CPU_Control instead of the anonymous integer.
Add _CPU_Processor_event_broadcast() and _CPU_Processor_event_receive()
functions provided by the CPU port. Use these functions in
_Per_CPU_Change_state() and _Per_CPU_Wait_for_state().
Add prototype for _SMP_Send_message().
Delete RTEMS_BSP_SMP_FIRST_TASK message. The first context switch is
now performed in rtems_smp_secondary_cpu_initialize(). Issuing the
first context switch in the context of the inter-processor interrupt is
not possible on systems with a modern interrupt controller. Such an
interrupt controler usually requires a handshake protocol with interrupt
acknowledge and end of interrupt signals. A direct context switch in an
interrupt handler circumvents the interrupt processing epilogue and may
leave the system in an inconsistent state.
Release lock in rtems_smp_process_interrupt() even if no message was
delivered. This prevents deadlock of the system.
Simplify and format _SMP_Send_message(),
_SMP_Request_other_cores_to_perform_first_context_switch(),
_SMP_Request_other_cores_to_dispatch() and
_SMP_Request_other_cores_to_shutdown().
|
|
|
|
|
|
|
| |
Do not call bsp_smp_secondary_cpu_initialize() in
rtems_smp_secondary_cpu_initialize(). This allows more flexibilty in
the BSP low-level code. Specify context requirements for a call to
rtems_smp_secondary_cpu_initialize().
|
| |
|