The support of PROPERTY_LINKER_SCRIPT_DEFINES has been broken since
the transition to CMake in 12f8f76165.
The intention was probably to allow users / projects to adjust
PROPERTY_LINKER_SCRIPT_DEFINES by setting a CMake cache variable.
The implementation tests for the CMake variable (local scope or cache)
PROPERTY_LINKER_SCRIPT_DEFINES, but it never uses such CMake variable.
Instead it uses a CMake global property named
PROPERTY_LINKER_SCRIPT_DEFINES. CMake variables and CMake global
properties are two very different things, and therefore the current
implementation has never worked. The fact that no one has never noticed
this flaw, means that the feature has never been used as intended.
Simplify the code by removing the use of the global CMake property and
instead set the value of the property on the linker script
pre-processing invocation.
Signed-off-by: Torsten Rasmussen <Torsten.Rasmussen@nordicsemi.no>
Unlike GNU ld, lld's gp relaxation is disabled by default and must be
explicitly enabled via `--relax-gp`. Pass this flag to enable gp relaxation
for lld when both linker relaxations and gp usage for RISC-V are enabled.
Signed-off-by: Jonathon Penix <jpenix@quicinc.com>
This patch adds Kconfig options to select either GNU libgcc or LLVM
compiler-rt. The 'rtlib' flag is provided in a config file, so this
patch introduces 'clang_libgcc.cfg' and 'clang_compiler_rt.cfg' which
enable appropriate library. The file is selected by concatenating
the 'clang_' prefix with library name.
Signed-off-by: Patryk Duda <pdk@semihalf.com>
Until now, linker-tool-gcc.h was used when LLD linker was chosen.
This causes linking issues because for GNU LD we use ALIGN_WITH_INPUT
attribute which is not available in LLVM LLD.
When using GNU LD we have to use ALIGN_WITH_INPUT to make sure that the
difference between VMA and LMA remains the same between output sections
that have different memory regions for VMA and LMA (RAM and FLASH).
With ALIGN_WITH_INPUT it's safe to do the memcpy of sections
that needs to be copied from flash to RAM in one function call:
(from z_data_copy() in kernel/xip.c)
```
z_early_memcpy(&__data_region_start, &__data_region_load_start,
__data_region_end - __data_region_start);
```
By default, LLVM LLD aligns both VMA and LMA to the same value, but
when --omagic (-N) option is provided then only the first output section
of given region has aligned LMA and the difference between VMA addresses
(0 is this is the first section) is added.
As a result the difference between LMA and VMA is constant for every
section, so this emulates ALIGN_WITH_INPUT option present in GNU LD
(required by XIP systems).
The --omagic flag is defined in cmake/linker/lld/target_baremetal.cmake
Example:
```
MEMORY {
ROM : ORIGIN = 0x1000, LENGTH = 1K
RAM : ORIGIN = 0x11000, LENGTH = 1K
}
SECTIONS {
.text 0x1000 : {
*(.text*)
} >ROM
.data.rel.ro : {
*(.data.rel.ro)
} >RAM AT>ROM
.data : {
*(.data*)
} >RAM AT>ROM
}
```
```
echo '.globl _start; _start: nop; .byte 1;'\
'.data.rel.ro; .balign 16; .byte 0;'\
'.data; .balign 32; .byte 0;' | \
llvm-mc -filetype=obj -triple=arm - -o test.o
armv7m-cros-eabi-ld.lld --sort-section=alignment -N -T script.ld \
test.o -o lld_out
```
```
Idx Name Size VMA LMA File off Algn
0 .text 00000005 00001000 00001000 00000094 2**2
1 .data.rel.ro 00000001 00011000 00001010 000000a0 2**4
2 .data 00000001 00011020 00001030 000000c0 2**5
```
In this example the first section has lower alignment than the following
section, but with -N option the difference between VMA and LMA is the
same for .data.rel.ro and .data sections.
For comparison, using BFD linker with --omagic option results in the
following:
```
Idx Name Size VMA LMA File off Algn
0 .text 00000005 00001000 00001000 00000094 2**2
1 .data.rel.ro 00000001 00011000 00001005 000000a0 2**4
2 .data 00000001 00011020 00001006 000000c0 2**5
```
with ALIGN_WITH_INPUT added, GNU LD adds the difference between VMA to
LMA, but doesn't align LMA of .data.rel.ro section:
```
Idx Name Size VMA LMA File off Algn
0 .text 00000005 00001000 00001000 00000074 2**2
1 .data.rel.ro 00000001 00011000 00001005 00000080 2**4
2 .data 00000001 00011020 00001025 000000a0 2**5
```
Signed-off-by: Patryk Duda <pdk@semihalf.com>
GNU ld and LLVM lld both complain under C++:
error: section: init_array is not contiguous with other relro sections
So do not create RELRO program header.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
This adds a new linker property specifically for passing
"-no-pie" to linker. Older binutils' LD (<= 2.36) do not
support this flag and will behave erratically if set. It
would parse "-no-pie" separately as "-n" and "-o-pie",
which would result in the output file being "-pie"
instead of "zephyr*.elf". Moreover, LLVM lld does not
support -no-pie but --no-pie (note the extra hyphen).
By having no-pie as a linker property, we can pass
correct no-pie flag to these linkers (or none at all).
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
Some distros may provide config files for clang to change its
default behavior. We need to override that, or else developers
may be using different defaults and we will have confusing
bug reports in the future.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
In some architectures the linker performs global optimization relaxing
address modes and changing intructions in the output object file. This
is a problem when userspace is enabled since it assumes that addresses
won't change after certain build stage. In no supported architectures
this option is ignored.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
See also
https://github.com/zephyrproject-rtos/zephyr/pull/38903
This is required when building tests for native_posix on ubuntu 22.04 using
clang-14 from the normal deb repository.
Signed-off-by: Christian Taedcke <christian.taedcke@lemonbeat.com>
