When I try to set priorities in certain programs, such as init_priority(HIGH_PRIORITY), I've noticed that during linking, there's no guarantee that the programs will be compiled in the sequence I've specified based on priority. This has led to some runtime errors in my program.
I realized that in the ld file, when initializing dynamic arrays, there's no assurance of initializing init_array.* before init_array. This has resulted in runtime errors in the program. Consequently, I've rearranged the init_array.* in the ld file of NuttX to be placed before init_array and added a SORT operation to init_array.* to ensure accurate initialization based on priorities during linking.
replace *(.init_array .init_array.*) with KEEP(*(.init_array .init_array.*)).
The KEEP statement within a linker script will instruct the linker to keep the specified section, even if no symbols inside it are referenced. This statement is used within the SECTIONS section of the linker script. This becomes relevant when garbage collection is performed at link time, enabled by passing the --gc-sections switch to the linker. The KEEP statement instructs the linker to use the specified section as a root node when creating a dependency graph, looking for unused sections. Essentially forcing the section to be marked as used.
Signed-off-by: cuiziwei <cuiziwei@xiaomi.com>
## Summary
A lot of linker scripts were listed twice, once for unix, once for windows.
This PR cleans up the logic so they're only listed once.
## Impact
Any opportunity to use a single source of truth and reduce lines of code is a win!
## Testing
CI will test all build
This reverts commit 45672c269d.
Because:
* It's very confusing to have cc as LD.
* I don't see what "-nostartfiles -nodefaultlibs" in LDFLAGS are
supposed to do when we use LD directly. It would be simpler to
remove them from our LDFLAGS.
The patch
Make: use gcc as LD
introduced use of GCC wrapper as linker. LD variable references GCC
executable now. But when GCC wrapper s used to build relocatable
loadable objects (ELF executables and modules) then it causes
linking of toolchain default libc and other libraries even when -r
is usd. Another problem is that incorrect multiarch variant is selected
for libraries search and possibly even for LTO or C++ templates
instantiating and other glue code which causes fails during linking
if CFLAGS selects non/default miltiarch variant.
Corresponding CFLAGS are passed to LDMODULEFLAGS and LDELFFLAGS
as well as -nostartfiles -nodefaultlibs options.
Separate line is used to easily find and adjust lines if link
process is changed in future.
Signed-off-by: Pavel Pisa <ppisa@pikron.com>
Normally, statically allocated data goes in .bss, followed by the
initial stack, followed by HEAP. However, any data that is statically
allocated in SRAM4 with __attribute__ ((section (".sram4"))) will
clobber, and be clobbered by, the HEAP.
On STM32H7, BDMA can only access SRAM4. Therefore any BDMA buffers (or
any other data) placed in SRAM4 will expose this problem. In one case,
this manifested as a failure of NSH to start, because the SPI6 BDMA
buffers clobbered the /dev/console inode structs, which the OS
allocated earlier.
This PR ensures that only the rest of SRAM4, after any static data, is
added to the heap. This PR also allows SRAM4 to be completely excluded
from the heap by a new Kconfig, CONFIG_STM32H7_SRAM4EXCLUDE, similar
to what CONFIG_STM32H7_DTCMEXCLUDE does for the DTCM region.
Change required in linker scripts:
Every STM32H7 linker script must replace this:
.sram4 :
{
} > sram4
with this:
.sram4_reserve (NOLOAD) :
{
*(.sram4)
. = ALIGN(4);
_sram4_heap_start = ABSOLUTE(.);
} > sram4
or link will fail with: undefined reference to '_sram4_heap_start'.
The Release Notes should document this for users with out-of-tree
boards.
arch/arm/src/stm32h7/Kconfig:
* Add config STM32H7_SRAM4EXCLUDE to allow excluding all of SRAM4
from the HEAP.
arch/arm/src/stm32h7/stm32_allocateheap.c:
* Only when including SRAM4 in the heap, define HAVE_SRAM4,
SRAM4_START, SRAM4_END, and SRAM4_HEAP_START.
* Add "Private Data" section.
* Add extern for_sram4_heap_start, which must be defined in the
board's linker script.
* arm_addregion(): Only add SRAM4 to the heap when configured to
do so, i.e., unless CONFIG_STM32H7_SRAM4EXCLUDE is defined, and
only add the portion of SRAM4 that is past any static data.
boards/arm/stm32h7/nucleo-h743zi/scripts/flash.ld:
boards/arm/stm32h7/nucleo-h743zi/scripts/kernel.space.ld:
boards/arm/stm32h7/nucleo-h743zi2/scripts/flash.ld:
boards/arm/stm32h7/stm32h747i-disco/scripts/flash.ld:
boards/arm/stm32h7/stm32h747i-disco/scripts/kernel.space.ld:
* Update all in-tree STM32H7 board linker scripts as described in
"Change required in linker scripts" above.
Testing:
* Successfully built all of the following configurations:
nucleo-h743zi2:jumbo
nucleo-h743zi2:nsh
nucleo-h743zi:nxlines_oled
nucleo-h743zi:elf
nucleo-h743zi:otg_fs_host
nucleo-h743zi:nsh
nucleo-h743zi:netnsh
nucleo-h743zi:pwm
stm32h747i-disco:nsh
* Tested with custom board.
* nxstyle.
References:
[1] See the dev@nuttx.a.o mailing list discussion started 2021/03/25:
"How to ensure HEAP will not overlap static DMA buffer?"
https://lists.apache.org/thread.html/recf2bb9043f8c9f53c10917e2adb2ec64fe35dc5e6f9a695a7ac6ecc%40%3Cdev.nuttx.apache.org%3E
[2] See arm_addregion() in arch/arm/src/stm32h7/stm32_allocateheap.c
Thanks to Gregory Nutt and David Sidrane for suggestions and reviews.
