CMake linker generator and MPU_ALIGN

[bjorniuppsala]

I am working to make CMake linker generator able to work together with CONFIG_USERSPACE.

In that there is an on problem that is a bit more complicated, and I would very much appreciate some input on how to solve.
The problem is how to transfer and combine information from the original cmake-invocation and the analysis of intermediary linking steps.

Most of this is simple, #85241 I have added two ways to define input to the linker generator:

• An easy way to include generated files from e.g. gen_app_partitions.py (via zephyr_linker_include_generated() )
• @variable@ in the parameters to zephyr_linker_*() to handle indiviual values

As an example these two are used to handle KOBJECT_DATA_SZ like so:
In the main CMake machinery (CMakelists.txt & cmake/linker_scripts/arm/linker.cmake)

zephyr_linker_include_generated(HEADER linker-kobject-prebuild-data.h) # Generated by gen_???.py
zephyr_linker_section(NAME kobject_data ...) 
zephyr_linker_section_configure(SECTION kobject_data INPUT ".kobject_data.data*" ALIGN @KOBJECT_DATA_ALIGN,undef:4@)

When the linker generator runs it picks up the #defines from the .h file, and uses them to to replace the @var@ (The "undef:" thing allows us to have the same definition in the linker passes where the information is not yet available).

In the ld-scripts it is handled with an #include "linker-kobject-prebuild-data.h and running the c-preprocessor.

The thing that complicates matters is that e.g. MPU_ALIGN requires more expressive power than straight variable substitution.

For arm/cortex_m MPU_ALIGN is defined like (from include/zephyr/arch/arm/cortex_m/scripts/linker.ld):

#if !defined(CONFIG_CUSTOM_SECTION_ALIGN) && defined(CONFIG_MPU_REQUIRES_POWER_OF_TWO_ALIGNMENT)
#define MPU_ALIGN(region_size) \
    . = ALIGN(_region_min_align); \
    . = ALIGN( 1 << LOG2CEIL(region_size))
#else
#define MPU_ALIGN(region_size) \
    . = ALIGN(_region_min_align)
#endif

For other targets the content varies, but the observation that we need input from both the main cmake machinery (the def of MPU_ALIGN) and the analysis steps (expression for the region_size) needs to be combined by the linker generator. For the ld-script skeletons this is accomplished by using the c-preprocessor. This is not feasible for the linker generator since it generates at least three varying file-formats (ld-script, scatter files and IAR icf files, all of which would require different handling).

My idea to handle this is to allow "recursive @-variable expansion". Meaning that when we would keep evaluating @foo@ as long as its value contains other (defined) @-variables. Together with allowing to define new variables as parameters inside the @-expression: @FOO,region_size=end-start@ would solve this problem.

Edit:
With my proposal, the above defintion of MPU_ALIGN moves to cmake/linker_script/arm/linker.cmake

if((NOT DEFINED CONFIG_CUSTOM_SECTION_ALIGN) AND DEFINED  CONFIG_MPU_REQUIRES_POWER_OF_TWO_ALIGNMENT)
  zephyr_linker_include_var(VAR MPU_ALIGN VALUE "MAX(${region_min_align} , 1 << LOG2CEIL(@region_size@) )")
else()
  zephyr_linker_include_var(VAR MPU_ALIGN VALUE "${region_min_align}")
endif()

And when using it:

zephyr_linker_section(NAME .foo_data ALIGN "@MPU_ALIGN,region_size=_foo_data_end - _foo_data_start@)

Is this a reasonable approach ? The @-expression "language" gets somewhat more complicated, somewhere we still need a cmake-file to define the body of MPU_ALIGN. This may be done either directly in the main cmake-machinery, or by including a defintion at linker generator time.

One possible alternative might be to allow any cmake expression inside of the @'s. This comes with complications of how to handle return values from cmake-functions though.

[bjorniuppsala]

@tejlmand ?

[bjorniuppsala]

There is now a working implementation of this in bjorniuppsala/iar-zephyr-fixes@953b7bb

[bjorniuppsala]

This is now a PR #89391