It wasn't possible to keep the CS line low between multiple message
descriptors in one transfer. This patch reworks the driver so that it is
possible.
Update #4180
In _Thread_queue_Flush_critical(), update the priority of the thread
queue owner only if necessary. The scheduler update priority operation
could be expensive.
In order to ensure FIFO fairness across schedulers, the thread queue
surrender operation must be used to dequeue a thread from the thread
queue. The thread queue extract operation is intended for timeouts.
Add _Thread_queue_Resume() which may be used to make extracted or
surrendered threads ready again.
Remove the now unused _Thread_queue_Extract_critical() function.
Close#4509.
The priority queues in clustered scheduling configurations use a per
scheduler priority queue rotation to ensure FIFO fairness across
schedulers. This mechanism is implemented in the thread queue surrender
operation. Unfortunately some semaphore and message queue directives
used wrongly the thread queue extract operation. Fix this through the
use of _Thread_queue_Surrender().
Update #4358.
The behaviour of the futex operations is defined by Linux:
https://man7.org/linux/man-pages/man2/futex.2.html
Use EAGIN instead of EWOULDBLOCK to be in line with the Linux man page.
These error numbers have the same value in Newlib. Using the same error
numbers helps to avoid confusion.
When you look at the history of the Linux man page you see that they
replaced EWOULDBLOCK with EAGAIN over time. At the time of the RTEMS
futex implementation they used EWOULDBLOCK.
- Support all possible descriptors in a select call. Borrowed
from Christain and his mDNS change in LibBSD
- If select (or poll) fails pause for a bit rather than
locking up in a hard loop
The Pi firmware added a wfe(wait for event), the cores 1-3 wait
for the start address being written to the mailbox register, followed
by a SEV poke to the mailbox that acts as a wfe wake-up event.
The NULL pointer check for the executing thread was introduced by
commit:
commit be3c257286
Author: Sebastian Huber <sebastian.huber@embedded-brains.de>
Date: Thu Jun 5 11:17:26 2014 +0200
score: Avoid NULL pointer access
Check that the executing thread is not NULL in _Scheduler_Tick(). It
may be NULL in case the processor has an optional scheduler assigned and
the system was not able to start the processor.
However, it is no longer necessary since now the clock interrupt is
distributed to the online processors.
In SMP configurations, on 64-bit architectures use plain atomic
operations to set/get the priority value of a scheduler node. On 32-bit
architectures use an ISR lock. Using a sequence lock has no real
benefit since it uses atomic read-modify-write operations for both the
read and the write lock. Simply use a ticket lock instead so that only
one SMP synchronization primitive is used for everything.
Use the new Priority_Group_order enum instead of a boolean to indicated if a
priority should be inserted as the first or last node into its priority group.
This makes the code more expressive. It is also a bit more efficient since a
branch in _Scheduler_Node_set_priority() is avoided and a simple bitwise or
operation can be used.
The rtems_partition_return_buffer() wrongly accepted which were exactly
at the buffer area end. Use the buffer area limit address for the range
checking.
Close#4490.
the timer_create() method can use CLOCK_MONOTONIC
but there was no test for this.
Also it implements the functionality to
create a CLOCK_MONOTONIC timer and gettime() .
Closes#3888
Cloning under Cygwin turned off executable permission on these
files. This shows them as modified even though they have not
explicitly been touched. Executable permission should not have
been on for these files so this is just a minor clean up.
Change license to BSD-2-Clause according to file histories and
documentation re-licensing agreement.
Place the group into the I/O Manager group. Add all source files to the
group.
Update #3899.
Update #3993.
Update #4482.
Move _CPU_Fatal_halt() declaration to <rtems/score/cpuimpl.h> and make sure it
is a proper declaration of a function which does not return. Fix the type of
the error code. If necessary, add the implementation to cpu.c. Implementing
_CPU_Fatal_halt() as a function makes it possible to wrap this function for
example to fully test _Terminate().
The per-CPU states which control the SMP system initialization were added quite
early during the SMP support development. Replace this initial implementation
with a simplified one. There is no longer a global SMP lock required which
serialized the state changes of all processors. The new implementation better
integrates with the per-CPU jobs.
Remove _CPU_SMP_Processor_event_broadcast() and
_CPU_SMP_Processor_event_receive(). These functions are hard to use since they
are subject to the lost wake up problem.
Use the same function to output the '\r\n' combination produced by
rtems_putc(). Fix the format.
Change licence according to file history.
Update #3053.
Users have access to the fatal error source and code though the fatal error
extension. The user-specific fatal error handling should be done in statically
initialized fatal error handlers. The _Internal_errors_What_happened was
updated after the fatal error extension. In addition, there was no API to get
the information stored in _Internal_errors_What_happened. In SMP
configurations, this object could contain a mix of different fatal errors.
Remove this object to save some bytes of storage.
Do not return a status code in bsp_interrupt_facility_initialize() since this
leads to unreachable code in bsp_interrupt_initialize(). Use RTEMS_DEBUG
assertions in bsp_interrupt_facility_initialize() if necessary.
Christian Mauderer
Sebastian Huber
Chris Johns
pranav
Eshan dhawan
Zacchaeus Leung
Joel Sherrill