@aiChaoSONG are you able to update cmake to 3.20 for Zephyr build CI ?

[ 84%] Building C object CMakeFiles/rimage.dir/src/adsp_config.c.o
[ 92%] Building C object CMakeFiles/rimage.dir/tomlc99/toml.c.o
[100%] Linking C executable rimage
[100%] Built target rimage
/workdir/zephyrproject
+ west build --build-dir build-apl --board intel_adsp_cavs15 zephyr/samples/subsys/audio/sof
/workdir/zephyrproject
-- west build: generating a build system
CMake Error at CMakeLists.txt:3 (cmake_minimum_required):
  CMake 3.20.0 or higher is required.  You are running version 3.18.3

are you able to update cmake to 3.20 for Zephyr build CI ?

@lgirdwood For internal CI, CMake is updated 3.21.1, we are good to build and test now. For github actions, as no Ubuntu is shipped with a higher version CMake, you will see zephyr build failure in github actions before we can fix it, I will talk with Marc on this.

Before we can fix the github action, you can check sof-ci/jenkins/pr-fw-build as an alternative.

You can check all firmware build log in the Details.

SOFCI TEST

Looks this introduces regression on APL, let me take a look to it.

@lgirdwood For internal CI, CMake is updated 3.21.1, we are good to build and test now. For github actions, as no Ubuntu is shipped with a higher version CMake, you will see zephyr build failure in github actions before we can fix it, I will talk with Marc on this.

@aiChaoSONG ubuntu CI can be forced to use a newer version of cmake by adding a private ppa for cmake. @marc-hb may have other or better ways to achieve this.

@lgirdwood Test running with Zephyr, will report results today. UPDATE: tests pass cleanly on Zephyr (TGL).

Zephyr Github Actions fixed by @aiChaoSONG in #4668

well, there is a point that even though for APL single core, it is preferred to use uncached lli address, this requires change for our cAVS heap definition @lgirdwood

/* heap block memory map */
struct mm {
	/* system heap - used during init cannot be freed */
	struct mm_heap system[PLATFORM_HEAP_SYSTEM];
	/* system runtime heap - used for runtime system components */
	struct mm_heap system_runtime[PLATFORM_HEAP_SYSTEM_RUNTIME];
-- #if CONFIG_CORE_COUNT > 1
	/* object shared between different cores - used during init cannot be freed */
	struct mm_heap system_shared[PLATFORM_HEAP_SYSTEM_SHARED];
	/* object shared between different cores */
	struct mm_heap runtime_shared[PLATFORM_HEAP_RUNTIME_SHARED];
-- #endif
	/* general heap for components */
	struct mm_heap runtime[PLATFORM_HEAP_RUNTIME];
	/* general component buffer heap */
	struct mm_heap buffer[PLATFORM_HEAP_BUFFER];

	struct mm_info total;
	uint32_t heap_trace_updated;	/* updates that can be presented */
	spinlock_t lock;	/* all allocs and frees are atomic */
};

@lgirdwood let me fix this with the heap refinement together, so let's pick it in this order:

1. heap refinement.
2. enlarge buffer heap.
3. DW DMA fixes.

@keyonjie looks like we need this to be well commented in the code due to the complexity of the problem along side the 1,2 and 3 fixes. Ok do do this a single PR that way we will keep your 1,22 3 ordering ?

Let me close this, will submit others first.