Skip to content

README.md

Latest commit

 

History

History
104 lines (74 loc) · 4.51 KB

README.md

File metadata and controls

104 lines (74 loc) · 4.51 KB

IPC4 Architecture

This directory holds the handlers and topology parsing logic for Inter-Processor Communication Version 4. IPC4 introduces a significantly denser, compound-command structure heavily based around the concept of "pipelines" and dynamic "modules" rather than static DSP stream roles.

Overview

Unlike older iterations (IPC3) which trigger single components via scalar commands, IPC4 uses compound structures. A single host interrupt might contain batch operations like building an entire processing chain, setting module parameters sequentially, and triggering a start across multiple interconnected blocks simultaneously.

Message Handling and Dispatch

IPC4 messages are received via the generic IPC handler entry point ipc_cmd(). ipc_cmd() determines the IPC target and dispatches the message appropriately. Global messages are dispatched to ipc4_process_glb_message(), while module-specific messages are routed directly to ipc4_process_module_message().

graph TD
    Mailbox[IPC Mailbox Interrupt] --> CoreIPC[ipc_cmd]

    CoreIPC --> TypeSel{Decode IPC Message Target}
    TypeSel -->|FW_GEN_MSG| Glb[ipc4_process_glb_message]
    TypeSel -->|MODULE_MSG| Mod[ipc4_process_module_message]

    subgraph Global Handler
        Glb --> NewPipe[ipc4_new_pipeline]
        Glb --> DelPipe[ipc4_delete_pipeline]
        Glb --> MemMap[ipc4_process_chain_dma]
        Glb --> SetPipe[ipc4_set_pipeline_state]
    end

    subgraph Module Handler
        Mod --> InitMod[ipc4_init_module_instance]
        Mod --> SetMod[ipc4_set_module_params]
        Mod --> GetMod[ipc4_get_module_params]
        Mod --> Bind[ipc4_bind]
        Mod --> Unbind[ipc4_unbind]
    end
Loading

Processing Flows

Pipeline State Management (ipc4_set_pipeline_state)

The core driver of graph execution in IPC4 is ipc4_set_pipeline_state(). This accepts a multi-stage request (e.g., START, PAUSE, RESET) and coordinates triggering the internal pipelines.

  1. State Translation: It maps the incoming IPC4 state request to an internal SOF state (e.g., IPC4_PIPELINE_STATE_RUNNING -> COMP_TRIGGER_START).
  2. Graph Traversal: It fetches the pipeline object associated with the command and begins preparing it (ipc4_pipeline_prepare).
  3. Trigger Execution: It executes ipc4_pipeline_trigger(), recursively changing states across the internal graphs and alerting either the LL scheduler or DP threads.
sequenceDiagram
    participant Host
    participant IPC4Set as ipc4_set_pipeline_state
    participant PPLPrep as ipc4_pipeline_prepare
    participant PPLTrig as ipc4_pipeline_trigger
    participant Comp as Graph Components

    Host->>IPC4Set: IPC4_PIPELINE_STATE_RUNNING
    activate IPC4Set

    IPC4Set->>PPLPrep: Maps to COMP_TRIGGER_START
    PPLPrep->>Comp: Applies PCM params & formatting
    Comp-->>PPLPrep: Components ready

    IPC4Set->>PPLTrig: execute trigger
    PPLTrig->>Comp: pipeline_trigger(COMP_TRIGGER_START)
    Comp-->>PPLTrig: Success

    IPC4Set-->>Host: Reply: ipc4_send_reply()
    deactivate IPC4Set
Loading

Module Instantiation and Binding (ipc4_bind)

In IPC4, modules (components) are bound together dynamically rather than constructed statically by the firmware at boot time.

  1. Instantiation: ipc4_init_module_instance() allocates the module via the DSP heap arrays based on UUIDs.
  2. Binding: ipc4_bind() takes two module IDs and dynamically connects their sink and source pins using intermediate comp_buffer objects.
sequenceDiagram
    participant Host
    participant IPC4Bind as ipc4_bind
    participant SrcMod as Source Module
    participant SinkMod as Sink Module
    participant Buff as Connection Buffer

    Host->>IPC4Bind: Bind Src(ID) -> Sink(ID)
    activate IPC4Bind

    IPC4Bind->>SrcMod: Locate by ID
    IPC4Bind->>SinkMod: Locate by ID

    IPC4Bind->>Buff: buffer_new() (Create Intermediate Storage)

    IPC4Bind->>SrcMod: Bind source pin to Buff (via comp_bind/comp_buffer_connect)
    IPC4Bind->>SinkMod: Bind sink pin to Buff (via comp_bind/comp_buffer_connect)

    IPC4Bind-->>Host: Reply: Linked
    deactivate IPC4Bind
Loading

Compound Messages (ipc_wait_for_compound_msg)

To accelerate initialization, IPC4 enables Compound commands. A host can send multiple IPC messages chained back-to-back using a single mailbox trigger flag before waiting for ACKs.

ipc_compound_pre_start and ipc_compound_post_start manage this batch execution safely without overflowing the Zephyr work queues or breaking hardware configurations during intermediate states.