support "cold" data and code

.github/workflows/zephyr.yml

@@ -73,6 +73,11 @@ jobs:
           path: ./workspace/sof
           filter: 'tree:0'
 
+      - name: free space
+        run: |
+          sudo rm -rf /usr/share/dotnet
+          sudo rm -rf /opt/ghc
+
       - name: west clones
         run: pip3 install west && cd workspace/sof/ && west init -l &&
                west update --narrow --fetch-opt=--filter=tree:0

Kconfig.sof

@@ -122,6 +122,14 @@ config COMPILER_INLINE_FUNCTION_OPTION
         help
           When enabled, -fno-inline-function option is not passed to compiler
 
+config COLD_STORE_EXECUTE_DRAM
+        bool "Execute and use cold data and code in DRAM"
+        help
+	  Non-performance critical data and code can be kept in DRAM to be
+	  accessed and executed there without copying to SRAM. Select this
+	  option to enable this feature to save SRAM and to speed up SRAM
+	  copying of performance-critical data and code.
+
 rsource "src/Kconfig"
 
 # See zephyr/modules/Kconfig

app/boards/intel_adsp_ace15_mtpm.conf

@@ -46,6 +46,7 @@ CONFIG_SOF_TELEMETRY_IO_PERFORMANCE_MEASUREMENTS=y
 CONFIG_SOF_TELEMETRY_PERFORMANCE_MEASUREMENTS=y
 CONFIG_ZEPHYR_NATIVE_DRIVERS=y
 CONFIG_ZEPHYR_DP_SCHEDULER=y
+CONFIG_COLD_STORE_EXECUTE_DRAM=y
 
 # SOF / loadable modules
 CONFIG_INTEL_MODULES=y

posix/include/sof/lib/memory.h

@@ -10,4 +10,12 @@
 
 #include <platform/lib/memory.h>
 
+#ifndef __cold
+#define __cold
+#endif
+
+#ifndef __cold_rodata
+#define __cold_rodata
+#endif
+
 #endif /* __SOF_LIB_MEMORY_H__ */

scripts/llext_link_helper.py

@@ -136,10 +136,10 @@ def main():
 	# run at arbitrary memory locations. One of the use-cases is running
 	# parts of the module directly in DRAM - sacrificing performance but
 	# saving scarce SRAM. We achieve this by placing non-performance
-	# critical functions in a .text.dram ELF section. When compiling and
-	# linking such functions, an additional .literal.dram section is
-	# automatically created. Note, that for some reason the compiler also
-	# marks that section as executable.
+	# critical functions in a .cold ELF section. When compiling and linking
+	# such functions, an additional .cold.literal section is automatically
+	# created. Note, that for some reason the compiler also marks that
+	# section as executable.
 	# This script links those sections at address 0. We could hard-code
 	# section names, but so far we choose to only link .text the "original"
 	# way and all other executable sections we link at 0.

src/audio/mfcc/mfcc.c

@@ -38,7 +38,7 @@ SOF_DEFINE_REG_UUID(mfcc);
 
 DECLARE_TR_CTX(mfcc_tr, SOF_UUID(mfcc_uuid), LOG_LEVEL_INFO);
 
-const struct mfcc_func_map mfcc_fm[] = {
+__cold_rodata const struct mfcc_func_map mfcc_fm[] = {
 #if CONFIG_FORMAT_S16LE
 	{SOF_IPC_FRAME_S16_LE,  mfcc_s16_default},
 #endif /* CONFIG_FORMAT_S16LE */

src/audio/mixin_mixout/mixin_mixout_generic.c

@@ -283,7 +283,7 @@ static void mix_s32_gain(struct cir_buf_ptr *sink, int32_t start_sample, int32_t
 }
 #endif	/* CONFIG_FORMAT_S32LE */
 
-const struct mix_func_map mix_func_map[] = {
+__cold_rodata const struct mix_func_map mix_func_map[] = {
 #if CONFIG_FORMAT_S16LE
 	{ SOF_IPC_FRAME_S16_LE, mix_s16, mix_s16_gain },
 #endif

src/audio/mixin_mixout/mixin_mixout_hifi3.c

@@ -607,7 +607,7 @@ static void mix_s32_gain(struct cir_buf_ptr *sink, int32_t start_sample, int32_t
 
 #endif	/* CONFIG_FORMAT_S32LE */
 
-const struct mix_func_map mix_func_map[] = {
+__cold_rodata const struct mix_func_map mix_func_map[] = {
 #if CONFIG_FORMAT_S16LE
 	{ SOF_IPC_FRAME_S16_LE, mix_s16, mix_s16_gain },
 #endif

src/audio/mixin_mixout/mixin_mixout_hifi5.c

@@ -269,7 +269,7 @@ static void mix_s32(struct cir_buf_ptr *sink, int32_t start_sample, int32_t mixe
 #endif	/* CONFIG_FORMAT_S32LE */
 
 /* TODO: implement mixing functions with gain support!*/
-const struct mix_func_map mix_func_map[] = {
+__cold_rodata const struct mix_func_map mix_func_map[] = {
 #if CONFIG_FORMAT_S16LE
 	{ SOF_IPC_FRAME_S16_LE, mix_s16, mix_s16 },
 #endif

src/audio/volume/volume.c

@@ -193,8 +193,12 @@ static uint32_t vol_zc_get_s32(const struct audio_stream *source,
 
 #endif /* CONFIG_FORMAT_S32LE */
 
-/** \brief Map of formats with dedicated zc functions. */
-static const struct comp_zc_func_map zc_func_map[] = {
+/**
+ * \brief Map of formats with dedicated zc functions.
+ *
+ * This is only used during @c .prepare() so it isn't performance-critical.
+ */
+__cold_rodata static const struct comp_zc_func_map zc_func_map[] = {
 #if CONFIG_FORMAT_S16LE
 	{ SOF_IPC_FRAME_S16_LE, vol_zc_get_s16 },
 #endif /* CONFIG_FORMAT_S16LE */

src/include/module/module/llext.h

@@ -34,10 +34,4 @@ static const struct sof_module_api_build_info buildinfo __section(".mod_buildinf
 	.api_version_number.full = SOF_MODULE_API_CURRENT_VERSION, \
 }
 
-#if CONFIG_LLEXT_TYPE_ELF_RELOCATABLE && defined(LL_EXTENSION_BUILD)
-#define __cold __section(".text.dram")
-#else
-#define __cold
-#endif
-
 #endif

src/ipc/ipc4/handler.c

@@ -891,7 +891,7 @@ static int ipc4_process_glb_message(struct ipc4_message_request *ipc4)
  * delete module <-------> free component
  */
 
-static int ipc4_init_module_instance(struct ipc4_message_request *ipc4)
+__cold static int ipc4_init_module_instance(struct ipc4_message_request *ipc4)
 {
 	struct ipc4_module_init_instance module_init;
 	struct comp_dev *dev;

src/ipc/ipc4/helper.c

@@ -102,7 +102,7 @@ static inline char *ipc4_get_comp_new_data(void)
 }
 #endif
 
-struct comp_dev *comp_new_ipc4(struct ipc4_module_init_instance *module_init)
+__cold struct comp_dev *comp_new_ipc4(struct ipc4_module_init_instance *module_init)
 {
 	struct comp_ipc_config ipc_config;
 	const struct comp_driver *drv;
@@ -966,9 +966,9 @@ const struct comp_driver *ipc4_get_comp_drv(uint32_t module_id)
 	uint32_t lib_idx = LIB_MANAGER_GET_LIB_ID(module_id);
 
 	if (lib_idx == 0) {
-		/* module_id 0 is used for base fw which is in entry 1 */
+		/* module_id 0 is used for base fw which is in entry 1 or 2 */
 		if (!module_id)
-			entry_index = 1;
+			entry_index = 1 + IS_ENABLED(CONFIG_COLD_STORE_EXECUTE_DRAM);
 		else
 			entry_index = module_id;
 

src/platform/ace30/include/platform/lib/memory.h

@@ -58,6 +58,12 @@
  */
 #define HEAPMEM_SIZE 0xD0000
 
+#if CONFIG_COLD_STORE_EXECUTE_DRAM && \
+	(CONFIG_LLEXT_TYPE_ELF_RELOCATABLE || !defined(LL_EXTENSION_BUILD))
+#define __cold __section(".cold")
+#define __cold_rodata __section(".coldrodata")
+#endif
+
 #endif /* __PLATFORM_LIB_MEMORY_H__ */
 
 #else

src/platform/lunarlake/include/platform/lib/memory.h

@@ -58,6 +58,12 @@
  */
 #define HEAPMEM_SIZE 0xF0000
 
+#if CONFIG_COLD_STORE_EXECUTE_DRAM && \
+	(CONFIG_LLEXT_TYPE_ELF_RELOCATABLE || !defined(LL_EXTENSION_BUILD))
+#define __cold __section(".cold")
+#define __cold_rodata __section(".coldrodata")
+#endif
+
 #endif /* __PLATFORM_LIB_MEMORY_H__ */
 
 #else

src/platform/meteorlake/include/platform/lib/memory.h

@@ -58,6 +58,12 @@
  */
 #define HEAPMEM_SIZE 0xF0000
 
+#if CONFIG_COLD_STORE_EXECUTE_DRAM && \
+	(CONFIG_LLEXT_TYPE_ELF_RELOCATABLE || !defined(LL_EXTENSION_BUILD))
+#define __cold __section(".cold")
+#define __cold_rodata __section(".coldrodata")
+#endif
+
 #endif /* __PLATFORM_LIB_MEMORY_H__ */
 
 #else

tools/rimage/config/lnl.toml.h

@@ -12,6 +12,20 @@
 
 	index = __COUNTER__
 
+#if CONFIG_COLD_STORE_EXECUTE_DRAM
+	[[module.entry]]
+	name = "COLD"
+	uuid = "D406D134-C3C1-402C-8AEC-6821C0C2B0E6"
+	affinity_mask = "3"
+	instance_count = "1"
+	domain_types = "0"
+	load_type = "0"
+	module_type = "0"
+	auto_start = "0"
+
+	index = __COUNTER__
+#endif
+
 	[[module.entry]]
 	name = "BASEFW"
 	uuid = "0E398C32-5ADE-BA4B-93B1-C50432280EE4"

tools/rimage/config/mtl.toml.h

@@ -12,6 +12,20 @@
 
 	index = __COUNTER__
 
+#if CONFIG_COLD_STORE_EXECUTE_DRAM
+	[[module.entry]]
+	name = "COLD"
+	uuid = "D406D134-C3C1-402C-8AEC-6821C0C2B0E6"
+	affinity_mask = "3"
+	instance_count = "1"
+	domain_types = "0"
+	load_type = "0"
+	module_type = "0"
+	auto_start = "0"
+
+	index = __COUNTER__
+#endif
+
 	[[module.entry]]
 	name = "BASEFW"
 	uuid = "0E398C32-5ADE-BA4B-93B1-C50432280EE4"

tools/rimage/config/ptl.toml.h

@@ -12,6 +12,20 @@ auto_start = "0"
 
 index = __COUNTER__
 
+#if CONFIG_COLD_STORE_EXECUTE_DRAM
+[[module.entry]]
+name = "COLD"
+uuid = "D406D134-C3C1-402C-8AEC-6821C0C2B0E6"
+affinity_mask = "3"
+instance_count = "1"
+domain_types = "0"
+load_type = "0"
+module_type = "0"
+auto_start = "0"
+
+index = __COUNTER__
+#endif
+
 [[module.entry]]
 name = "BASEFW"
 uuid = "0E398C32-5ADE-BA4B-93B1-C50432280EE4"

xtos/include/sof/lib/memory.h

@@ -10,4 +10,12 @@
 
 #include <platform/lib/memory.h>
 
+#ifndef __cold
+#define __cold
+#endif
+
+#ifndef __cold_rodata
+#define __cold_rodata
+#endif
+
 #endif /* __SOF_LIB_MEMORY_H__ */

zephyr/include/sof/lib/memory.h

@@ -8,4 +8,12 @@
 
 #include <platform/lib/memory.h>
 
+#ifndef __cold
+#define __cold
+#endif
+
+#ifndef __cold_rodata
+#define __cold_rodata
+#endif
+
 #endif /* __SOF_LIB_MEMORY_H__ */