The rtems_fdt_subnode_offset and rtems_fdt_get_name try to binary
search the entries index by the DTB offset, but the index is sorted
by alphabetically by the name, so the search is broken. Replace these
searches by directly calling the libfdt functions. Since the DTB offset
is known, the performance hit for invoking the libfdt should not be too
much.
Closes#5606.
sp13 already constructs Q3 using rtems_message_queue_construct()
in init.c but task1.c never sends to, receives from, or deletes
the constructed queue.
Extend task1.c to send a message to Q3, receive it back, and
delete the queue after the variable-size loop completes. This
exercises the full lifecycle of a queue created with static
storage and storage_free set to NULL.
Verified on arm/realview_pbx_a9_qemu with QEMU 9.0.2,
RTEMS 7.0.0.a413333f2f04130e82e15addba402887222c345f, GCC 15.2.0.
RAM parity self-tests run on several RAM modules, which work fine for
the 3137 variants, but on 4357 variants some of those modules use
an ECC memory protection scheme.
This commit turns the relevent ones off for 4357 until proper ECC
self-tests are created.
Move the upstream FatFs sources from cpukit/libfs/src/fatfs to
contrib/fatfs.Keep the RTEMS integration layer (rtems-fatfs-*.c and
rtems-diskio.c) in libfs. Introduce a wrapper header under rtems/fatfs
and update internal sources to use it.
Update the build specification to reference the new location without
introducing global include paths.
Fixes#5433
The recent FDT changes included some refactoring that missed the second
parameter in a rtems_fdt_valid_handle_unprotected() call. This adds the
missing parameter and fixes the build.
The change to avoid deadlock could introduce a race condition on the
blobs chain and the handle. Refactor the rtems_fdt_valid_handle() to
create an internal unprotected version, and call it with the global
rtems_fdt_lock() held.
Reading non-compressed dtb files was broken because of the initial
reading of the first two bytes to check for compression. Reset the
file pointer after making that first read regardless of whether a
gzip header is found or not.
Splits functionality so that reading from the dtb file is distinct from
loading it into memory. This allows directly loading a dtb from memory
instead of from a file.
The number of external interrupts can be configured through the FDT.
Read it during PLIC initialization and use that to determine the max
valid vector number.
The previous addition of s-mode support avoided touching the PLIC
registers and only allowed limited use of software interrupts for SMP
IPIs. This enables the PLIC in s-mode mapping the registers with the MMU
and unifying some aspects of the m-mode and s-mode implementations.
Updates #3337
Update to tms570-sci.c (commit af77d28) puts a gate to check
requests for non 8-bit bytes and CREAD not set, we therefore
must expclitily set them now.
- Changed the interrupt attributes so extended vectors
report can_raise as false.
- Changed the bsp_interrupt_raise to return
RTEMS_UNSATISFIED for extended vectors 16-31.
- BSP-specific validation for gr712rc and gr740
were added to exercised the extended vectors, and
check that the right interrupt handler is called.
Updates #4846.
Adds invalidation of cache (using
rtems_cache_invalidate_multiple_instruction_lines which is
included in the wrapper function
rtems_cache_instruction_sync_after_code_change.
This is to ensure full syncing of the I-cache after copying
instructions (in .fast_text into RAM with
bsp_start_copy_sections_compact in bspstarthooks.c
The smpclock01 test relies on tight synchronization between cores and it
checks cross-core Watchdog.ticks. On some targets the cores can become
de-synchronized so that the tick counts are not reliable. Slowing the
tick rate down seems to help this problem.
Add post-2100 validation checks in spclock_err02
without increasing runtime and remove excessive prints.
The test verifies century and leap-year rules via RTEMS
clock APIs and gmtime()/mktime() for future years.
Fixes#5507
_BM(x) produces 64-bit values which are implicitly converted to int
in reloc_target_bitmask. For x = 32, this causes truncation from
4294967295 to -1 on 32-bit targets, triggering -Woverflow.
Used uint32_t and cast the result to avoid implicit narrowing.
Fixes#5557
Fix parity handling in zynq_uart_set_attributes() by clearing existing
parity bits before setting a new mode instead of always ORing PAR_NONE,
which caused incorrect or ignored parity; also fix inverted PARODD logic
and add ZYNQ_UART_MODE_PAR_MASK to manage the parity field correctly.
Fixes#5554
The rtems_test_spin_for_ticks( x ) executes one partial tick and the
remaining x - 1 as complete ticks. Some tests assumed it executed x
complete ticks. Fix those tests.
Fixes#5575.
Fix incorrect comparison between fdt_getprop() length (bytes)
and address cells count.
The check now uses ac * sizeof(uint32_t) to avoid
reading invalid data.
Also change FDT pointer type from const char * to const void *
to match libfdt API.
No functional changes.
CAN controllers need a running worker task to function. Check if
the task is actually started and return NULL (and free already allocated
resources) if it fails.
Also assert if cleanup functions fail.
Signed-off-by: Michal Lenc <michallenc@seznam.cz>
- leon3 interrupt vector 15 is non-maskable and should
not be reported as disable-capable.
- added leon3_interrupt_is_maskable function, and
used it to set can_disable and maybe_disable in
bsp_interrupt_get_attributes().
- added verification for non-maskable vectors in
bsp_interrupt_vector_disable(), which is rejected
with error RTEMS_UNSATISFIED.
Updates #4830.
Gedare Bloom
Matteo Concas
Moksh Panicker
Avi Weiss
Chris Johns
Shivam Deolankar
Jeremy Lorelli
Kinsey Moore
Samuel Viegas
Kinsey Moore
Mohamed Ayman
Michal Lenc
Samuel Viegas