The allocator mutex is recursive and already used for the task creation
in rtems_timer_initiate_server(). Just use this mutex instead of the
once mutex to serialize the initialization. This avoids a dependency on
condition variables which are not used here.
The zero size allocations had no consistent behaviour in RTEMS. For
example, malloc( 0 ) returned NULL and posix_memalign( &p, align, 0 )
returned in p a unique pointer (or NULL if no memory is available). In
POSIX, zero size memory allocations are implementation-defined
behaviour. The implementation has two options:
https://pubs.opengroup.org/onlinepubs/9699919799/functions/malloc.htmlhttps://pubs.opengroup.org/onlinepubs/9699919799/functions/posix_memalign.html
Linux and FreeBSD return a unique pointer for zero size memory
allocations. Use this approach for RTEMS as well throughout the memory
allocation directives
Close#4390.
The __builtin_unreachable() cannot be used with current GCC versions to
tell the compiler that a function does not return to the caller, see:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99151
Add a no return variant of _Thread_Dispatch_direct() to avoid generation
of dead code.
If only one constraint is present, then use a single sentence, otherwise
use a list for the constraints. Format the constraints so that the line
length limit is maintained.
The 'sc' variable was originally storing the return value of mkdir().
This was causing an issue, so it was changed to make it to where we
ignored the return value with (void). The 'sc' variable was left in
by mistake.
The "powerpc/shared/console" code has the start-up console value fixed
at 9600 baud. This changes the hard-wired constant "9600" in the code
to the configuration setting "BSP_CONSOLE_BAUD" and adds configuration
support in both the "waf" and the legacy configuration systems.
Note that the VME BSPs beatnik, mvme3100, and mve5100 can be improved
by adding a "mvmexxxx" BSP family. This configuration change, as well
as future configuration changes, could then be made in a "grp.yml" file.
In case the processor set is not large enough to contain the processor
affinity set of the task return RTEMS_INVALID_SIZE instead of
RTEMS_INVALID_NUMBER. This is more in line with other directives since
the issue is related to the size of an object.
Close#4393.
Refactored the i2c driver to parse register values from the device
tree rather than hardcoding them. But still the clocks have to
initialized manually.
This patch fixes issue #4338 by changing _TOD_Validate()
to only accept years till 2105. This requires another patch
to change the documentation of rtems_clock_set() and other
affected API functions (indicating the end date is 2105 not 2514).
I tried to support till year 2514 but it turned out that
this needs changing the Timer Manager too. That in turn
would mean to change _TOD_Seconds_since_epoch( void )
from 32 to 64 bit. Sebastian pointed out that a naive extension
leads to trouble with 32 bit processors. He deemed a safe
re-implementation too costly performance wise considering
that year 2106 is far away and current binaries using RTEMS
Classic API are unlikely to be in use by 2106.
The constant TOD_SECONDS_AT_2100_03_01_00_00 in
cpukit/rtems/src/clocktodtoseconds.c happens to be wrong by
1 hour. When setting the date 2100-Feb-28 23:59:59 and then
reading the date again you will find yourself in 2100-Feb-27.
Update #4338
Set the Main Stack Pointer (MSP) to the ISR stack area end just in case
we run using the Process Stack Pointer (PSP). This helps if
applications are started by a boot loader.
Sebastian Huber
Vijay Kumar Banerjee
Ryan Long
Peter Dufault
Robin Mueller
G S Niteesh Babu
Frank Kühndel