[DRAFT/TESTING] userspace LL/audio test PR (latest ver: V43)

app/boards/intel_adsp_ace30_ptl.conf

@@ -21,8 +21,8 @@ CONFIG_KCPS_DYNAMIC_CLOCK_CONTROL=n
 CONFIG_PROBE=y
 CONFIG_PROBE_DMA_MAX=2
 CONFIG_SOF_TELEMETRY=y
-CONFIG_SOF_TELEMETRY_IO_PERFORMANCE_MEASUREMENTS=y
-CONFIG_SOF_TELEMETRY_PERFORMANCE_MEASUREMENTS=y
+CONFIG_SOF_TELEMETRY_IO_PERFORMANCE_MEASUREMENTS=n
+CONFIG_SOF_TELEMETRY_PERFORMANCE_MEASUREMENTS=n
 CONFIG_COLD_STORE_EXECUTE_DRAM=y
 
 # SOF / loadable modules
@@ -82,4 +82,24 @@ CONFIG_SOF_STACK_SIZE=8192
 CONFIG_SOF_USERSPACE_PROXY=y
 CONFIG_MAX_THREAD_BYTES=3
 
-CONFIG_MAX_DOMAIN_PARTITIONS=32
+CONFIG_MAX_DOMAIN_PARTITIONS=64
+
+# Userspace LL (was app/overlays/ptl/ll_userspace_overlay.conf)
+# Run Low-Latency audio pipelines in user-space threads by default.
+CONFIG_SOF_USERSPACE_LL=y
+CONFIG_SOF_USERSPACE_INTERFACE_DMA=y
+CONFIG_DAI_USERSPACE=y
+
+# Settings currently required to enable user-space LL. The cold-store,
+# telemetry, loadable-module and misc feature disables above/here are not
+# yet user-space compatible (see the former overlay for rationale).
+CONFIG_COLD_STORE_EXECUTE_DEBUG=n
+CONFIG_SOF_BOOT_TEST_ALLOWED=n
+CONFIG_CROSS_CORE_STREAM=n
+CONFIG_INTEL_ADSP_MIC_PRIVACY=n
+CONFIG_XRUN_NOTIFICATIONS_ENABLE=n
+# Extend the shared LL user-space heap to 0.5MB. This
+# is current maximum for PTL builds with virtual heap
+# enabled and uses the vmh allocation intended for KBP.
+# TODO: needs some better solution to allocate
+CONFIG_SOF_ZEPHYR_SYS_USER_HEAP_SIZE=0x80000

app/boards/intel_adsp_ace30_wcl.conf

@@ -21,8 +21,8 @@ CONFIG_KCPS_DYNAMIC_CLOCK_CONTROL=n
 CONFIG_PROBE=y
 CONFIG_PROBE_DMA_MAX=2
 CONFIG_SOF_TELEMETRY=y
-CONFIG_SOF_TELEMETRY_IO_PERFORMANCE_MEASUREMENTS=y
-CONFIG_SOF_TELEMETRY_PERFORMANCE_MEASUREMENTS=y
+CONFIG_SOF_TELEMETRY_IO_PERFORMANCE_MEASUREMENTS=n
+CONFIG_SOF_TELEMETRY_PERFORMANCE_MEASUREMENTS=n
 CONFIG_COLD_STORE_EXECUTE_DRAM=y
 
 # SOF / loadable modules
@@ -69,3 +69,36 @@ CONFIG_PM_DEVICE_RUNTIME_ASYNC=n
 CONFIG_LOG_BACKEND_ADSP=n
 CONFIG_LOG_FLUSH_SLEEP_US=5000
 CONFIG_WINSTREAM_CONSOLE=n
+
+# Userspace base (mirrored from intel_adsp_ace30_ptl.conf)
+# Required so that user-space LL (below) can actually be enabled, since
+# CONFIG_SOF_USERSPACE_LL depends on CONFIG_USERSPACE.
+CONFIG_USERSPACE=y
+CONFIG_DYNAMIC_THREAD=y
+CONFIG_DYNAMIC_THREAD_ALLOC=y
+CONFIG_DYNAMIC_THREAD_PREFER_ALLOC=y
+CONFIG_SOF_STACK_SIZE=8192
+CONFIG_SOF_USERSPACE_PROXY=y
+CONFIG_MAX_THREAD_BYTES=3
+CONFIG_MAX_DOMAIN_PARTITIONS=64
+CONFIG_XTENSA_MMU_NUM_L2_TABLES=128
+
+# Userspace LL (was app/overlays/ptl/ll_userspace_overlay.conf)
+# Run Low-Latency audio pipelines in user-space threads by default.
+CONFIG_SOF_USERSPACE_LL=y
+CONFIG_SOF_USERSPACE_INTERFACE_DMA=y
+CONFIG_DAI_USERSPACE=y
+
+# Settings currently required to enable user-space LL. The cold-store,
+# telemetry, loadable-module and misc feature disables above/here are not
+# yet user-space compatible (see the former overlay for rationale).
+CONFIG_COLD_STORE_EXECUTE_DEBUG=n
+CONFIG_SOF_BOOT_TEST_ALLOWED=n
+CONFIG_CROSS_CORE_STREAM=n
+CONFIG_INTEL_ADSP_MIC_PRIVACY=n
+CONFIG_XRUN_NOTIFICATIONS_ENABLE=n
+# Extend the shared LL user-space heap to 0.5MB. This
+# is current maximum for WCL builds with virtual heap
+# enabled and uses the vmh allocation intended for KBP.
+# TODO: needs some better solution to allocate
+CONFIG_SOF_ZEPHYR_SYS_USER_HEAP_SIZE=0x80000

app/overlays/ptl/ll_userspace_overlay.conf

@@ -37,3 +37,6 @@ CONFIG_XRUN_NOTIFICATIONS_ENABLE=n
 # enabled and uses the vmh allocation intended for KBP.
 # TODO: needs some better solution to allocate
 CONFIG_SOF_ZEPHYR_SYS_USER_HEAP_SIZE=0x80000
+
+# sof-ptl-nocodec.tplg requires more than 32 partitions for DP
+CONFIG_MAX_DOMAIN_PARTITIONS=64

src/audio/module_adapter/module_adapter.c

@@ -77,11 +77,12 @@ static struct vregion *module_adapter_dp_heap_new(const struct comp_ipc_config *
 static
 struct processing_module *module_adapter_mem_alloc(const struct comp_driver *drv,
 						   const struct comp_ipc_config *config,
-						   const struct module_ext_init_data *ext_init)
+						   const struct module_ext_init_data *ext_init,
+						   struct mod_alloc_ctx *ppl_alloc)
 {
-	struct k_heap *mod_heap;
-	struct vregion *mod_vreg;
+	struct k_heap *mod_heap = NULL;
 	struct processing_module *mod;
+	struct mod_alloc_ctx *alloc;
 	struct comp_dev *dev;
 	/*
 	 * For DP shared modules the struct processing_module object must be
@@ -92,49 +93,53 @@ struct processing_module *module_adapter_mem_alloc(const struct comp_driver *drv
 	 */
 	uint32_t flags = config->proc_domain == COMP_PROCESSING_DOMAIN_DP ?
 		SOF_MEM_FLAG_USER | SOF_MEM_FLAG_COHERENT : SOF_MEM_FLAG_USER;
+	bool use_ppl_alloc = ppl_alloc &&
+			     config->proc_domain == COMP_PROCESSING_DOMAIN_LL;
 
-	if (config->proc_domain == COMP_PROCESSING_DOMAIN_DP && IS_ENABLED(CONFIG_SOF_VREGIONS) &&
-	    IS_ENABLED(CONFIG_USERSPACE) && !IS_ENABLED(CONFIG_SOF_USERSPACE_USE_DRIVER_HEAP)) {
-		mod_vreg = module_adapter_dp_heap_new(config, ext_init);
-		if (!mod_vreg) {
-			comp_cl_err(drv, "Failed to allocate DP module heap / vregion");
-			return NULL;
-		}
+	if (use_ppl_alloc) {
+		/* LL modules share the pipeline's alloc context */
+		alloc = ppl_alloc;
+		vregion_get(alloc->vreg);
+	} else {
+		struct vregion *mod_vreg = NULL;
+
+		if (IS_ENABLED(CONFIG_SOF_VREGIONS) && IS_ENABLED(CONFIG_USERSPACE) &&
+		    !IS_ENABLED(CONFIG_SOF_USERSPACE_USE_DRIVER_HEAP) &&
+		    config->proc_domain == COMP_PROCESSING_DOMAIN_DP) {
+			mod_vreg = module_adapter_dp_heap_new(config, ext_init);
+			if (!mod_vreg) {
+				comp_cl_err(drv, "Failed to allocate DP module heap / vregion");
+				return NULL;
+			}
 #ifdef CONFIG_SOF_USERSPACE_LL
-		mod_heap = sof_sys_user_heap_get();
-#else
-		mod_heap = NULL;
+			mod_heap = sof_sys_user_heap_get();
 #endif
-	} else {
+		} else {
 #ifdef CONFIG_SOF_USERSPACE_LL
-		mod_heap = sof_sys_user_heap_get();
-		comp_cl_dbg(drv, "using ll user heap for module");
+			mod_heap = sof_sys_user_heap_get();
+			comp_cl_dbg(drv, "using ll user heap for module");
 #else
-		mod_heap = drv->user_heap;
+			mod_heap = drv->user_heap;
 #endif
-		mod_vreg = NULL;
-	}
+		}
+		alloc = sof_heap_alloc(mod_heap, flags, sizeof(*alloc), 0);
+		if (!alloc) {
+			vregion_put(mod_vreg);
+			return NULL;
+		}
 
-	if (!mod_vreg)
-		mod = sof_heap_alloc(mod_heap, flags, sizeof(*mod), 0);
-	else if (flags & SOF_MEM_FLAG_COHERENT)
-		mod = vregion_alloc_coherent(mod_vreg, sizeof(*mod));
-	else
-		mod = vregion_alloc(mod_vreg, sizeof(*mod));
+		memset(alloc, 0, sizeof(*alloc));
+		alloc->heap = mod_heap;
+		alloc->vreg = mod_vreg;
+	}
 
+	mod = sof_ctx_alloc(alloc, flags, sizeof(*mod), 0);
 	if (!mod) {
 		comp_cl_err(drv, "failed to allocate memory for module");
 		goto emod;
 	}
 
-	struct mod_alloc_ctx *alloc = sof_heap_alloc(mod_heap, flags, sizeof(*alloc), 0);
-
-	if (!alloc)
-		goto ealloc;
-
 	memset(mod, 0, sizeof(*mod));
-	alloc->heap = mod_heap;
-	alloc->vreg = mod_vreg;
 	mod->priv.resources.alloc = alloc;
 	mod_resource_init(mod);
 
@@ -144,11 +149,7 @@ struct processing_module *module_adapter_mem_alloc(const struct comp_driver *drv
 	 * then it can be cached. Effectively it can be only cached in
 	 * single-core configurations.
 	 */
-	if (mod_vreg)
-		dev = vregion_alloc_coherent(mod_vreg, sizeof(*dev));
-	else
-		dev = sof_heap_alloc(mod_heap, SOF_MEM_FLAG_COHERENT, sizeof(*dev), 0);
-
+	dev = sof_ctx_alloc(alloc, SOF_MEM_FLAG_COHERENT, sizeof(*dev), 0);
 	if (!dev) {
 		comp_cl_err(drv, "failed to allocate memory for comp_dev");
 		goto edev;
@@ -163,41 +164,39 @@ struct processing_module *module_adapter_mem_alloc(const struct comp_driver *drv
 	return mod;
 
 edev:
-	sof_heap_free(mod_heap, alloc);
-ealloc:
-	if (mod_vreg)
-		vregion_free(mod_vreg, mod);
-	else
-		sof_heap_free(mod_heap, mod);
+	sof_ctx_free(alloc, mod);
 emod:
-	vregion_put(mod_vreg);
+	vregion_put(alloc->vreg);
+	if (!use_ppl_alloc)
+		sof_heap_free(mod_heap, alloc);
 
 	return NULL;
 }
 
 static void module_adapter_mem_free(struct processing_module *mod)
 {
 	struct mod_alloc_ctx *alloc = mod->priv.resources.alloc;
-	struct k_heap *mod_heap = alloc->heap;
+	bool ppl_alloc = mod->dev->ipc_config.proc_domain == COMP_PROCESSING_DOMAIN_LL &&
+			 mod->dev->pipeline && mod->dev->pipeline->alloc == alloc;
 
 	/*
 	 * In principle it shouldn't even be needed to free individual objects
 	 * on the module heap since we're freeing the heap itself too
 	 */
 #if CONFIG_IPC_MAJOR_4
-	sof_heap_free(mod_heap, mod->priv.cfg.input_pins);
+	sof_heap_free(alloc->heap, mod->priv.cfg.input_pins);
 #endif
-	if (alloc->vreg) {
-		struct vregion *mod_vreg = alloc->vreg;
-
-		vregion_free(mod_vreg, mod->dev);
-		vregion_free(mod_vreg, mod);
-		if (!vregion_put(mod_vreg))
-			sof_heap_free(alloc->heap, alloc);
+	sof_ctx_free(alloc, mod->dev);
+	sof_ctx_free(alloc, mod);
+
+	if (ppl_alloc) {
+		/* alloc belongs to pipeline, just release vregion reference */
+		vregion_put(alloc->vreg);
+	} else if (alloc->vreg) {
+		if (!vregion_put(alloc->vreg))
+			rfree(alloc);
 	} else {
-		sof_heap_free(mod_heap, mod->dev);
-		sof_heap_free(mod_heap, mod);
-		sof_heap_free(mod_heap, alloc);
+		rfree(alloc);
 	}
 }
 
@@ -248,8 +247,19 @@ struct comp_dev *module_adapter_new_ext(const struct comp_driver *drv,
 		NULL;
 #endif
 
-	struct processing_module *mod = module_adapter_mem_alloc(drv, config, ext_init);
+	struct mod_alloc_ctx *ppl_alloc = NULL;
+#if CONFIG_IPC_MAJOR_4
+	struct ipc_comp_dev *ipc_pipe;
+	struct ipc *ipc = ipc_get();
+
+	/* resolve the pipeline pointer early to pass its alloc to mem_alloc */
+	ipc_pipe = ipc_get_comp_by_ppl_id(ipc, COMP_TYPE_PIPELINE, config->pipeline_id,
+					  IPC_COMP_IGNORE_REMOTE);
+	if (ipc_pipe && ipc_pipe->pipeline)
+		ppl_alloc = ipc_pipe->pipeline->alloc;
+#endif
 
+	struct processing_module *mod = module_adapter_mem_alloc(drv, config, ext_init, ppl_alloc);
 	if (!mod)
 		return NULL;
 
@@ -273,6 +283,21 @@ struct comp_dev *module_adapter_new_ext(const struct comp_driver *drv,
 	dst->ext_data = &ext_data;
 #endif
 
+#if CONFIG_IPC_MAJOR_4
+	/*
+	 * Set the pipeline pointer if ipc_pipe is valid. Do this
+	 * early so that we can use module_adapter_mem_free() in error
+	 * handling.
+	 */
+	if (ipc_pipe) {
+		dev->pipeline = ipc_pipe->pipeline;
+
+		/* LL modules have the same period as the pipeline */
+		if (dev->ipc_config.proc_domain == COMP_PROCESSING_DOMAIN_LL)
+			dev->period = ipc_pipe->pipeline->period;
+	}
+#endif
+
 #if CONFIG_ZEPHYR_DP_SCHEDULER
 	/* create a task for DP processing */
 	if (config->proc_domain == COMP_PROCESSING_DOMAIN_DP) {
@@ -306,22 +331,6 @@ struct comp_dev *module_adapter_new_ext(const struct comp_driver *drv,
 	else
 		goto err;
 
-#if CONFIG_IPC_MAJOR_4
-	struct ipc_comp_dev *ipc_pipe;
-	struct ipc *ipc = ipc_get();
-
-	/* set the pipeline pointer if ipc_pipe is valid */
-	ipc_pipe = ipc_get_comp_by_ppl_id(ipc, COMP_TYPE_PIPELINE, config->pipeline_id,
-					  IPC_COMP_IGNORE_REMOTE);
-	if (ipc_pipe) {
-		dev->pipeline = ipc_pipe->pipeline;
-
-		/* LL modules have the same period as the pipeline */
-		if (dev->ipc_config.proc_domain == COMP_PROCESSING_DOMAIN_LL)
-			dev->period = ipc_pipe->pipeline->period;
-	}
-#endif
-
 	/* Init processing module */
 	ret = module_init(mod);
 	if (ret) {