Inter-Processor Call

备注

The IPC is an Inter-Processor Call and NOT Inter-Process Communication as found on other operating systems.

Overview

Due to the dual core nature of the ESP32-S3, there are instances where a certain callback must be run in the context of a particular CPU such as:

  • When allocating an ISR to an interrupt source of a particular CPU (applies to freeing a particular CPU’s interrupt source as well).

  • On particular chips (such as the ESP32), accessing memory that is exclusive to a particular CPU (such as RTC Fast Memory).

  • Reading the registers/state of another CPU.

The IPC (Inter-Processor Call) feature allows a particular CPU (the calling CPU) to trigger the execution of a callback function on another CPU (the target CPU). The IPC feature allows execution of a callback function on the target CPU in either a task context, or a High Priority Interrupt context (see High-Level Interrupts for more details). Depending on the context that the callback function is executed in, different restrictions apply to the implementation of the callback function.

IPC in Task Context

The IPC feature implements callback execution in a task context by creating an IPC task for each CPU during application startup. When the calling CPU needs to execute a callback on the target CPU, the callback will execute in the context of the target CPU’s IPC task.

When using IPCs in a task context, users need to consider the following:

  • IPC callbacks should ideally be simple and short. An IPC callback should avoid attempting to block or yield.

  • The IPC tasks are created at the highest possible priority (i.e., configMAX_PRIORITIES - 1) thus the callback should also run at that priority as a result. However, CONFIG_ESP_IPC_USES_CALLERS_PRIORITY is enabled by default which will temporarily lower the priority of the target CPU’s IPC task to the calling CPU before executing the callback.

  • Depending on the complexity of the callback, users may need to configure the stack size of the IPC task via CONFIG_ESP_IPC_TASK_STACK_SIZE.

  • The IPC feature is internally protected by a mutex. Therefore, simultaneous IPC calls from two or more calling CPUs will be handled on a first come first serve basis.

API Usage

Task Context IPC callbacks have the following restrictions:

  • The callback must be of type void func(void *arg)

  • The callback should avoid attempting to block or yield as this will result in the target CPU’s IPC task blocking or yielding.

  • The callback must avoid changing any aspect of the IPC task (e.g., by calling vTaskPrioritySet(NULL, x)).

The IPC feature offers the API listed below to execute a callback in a task context on a target CPU. The API allows the calling CPU to block until the callback’s execution has completed, or return immediately once the callback’s execution has started.

  • esp_ipc_call() will trigger an IPC call on the target CPU. This function will block until the target CPU’s IPC task begins execution of the callback.

  • esp_ipc_call_blocking() will trigger an IPC on the target CPU. This function will block until the target CPU’s IPC task completes execution of the callback.

IPC in ISR Context

In some cases, we need to quickly obtain the state of another CPU such as in a core dump, GDB stub, various unit tests, and DPORT workaround. For such scenarios, the IPC feature supports execution of callbacks in a High Priority Interrupt context. The IPC feature implements the High Priority Interrupt context by reserving a High Priority Interrupt on each CPU for IPC usage. When a calling CPU needs to execute a callback on the target CPU, the callback will execute in the context of the High Priority Interrupt of the target CPU.

When using IPCs in High Priority Interrupt context, users need to consider the following:

  • Since the callback is executed in a High Priority Interrupt context, the callback must be written entirely in assembly. See the API Usage below for more details regarding writing assembly callbacks.

  • The priority of the reserved High Priority Interrupt is dependent on the CONFIG_ESP_SYSTEM_CHECK_INT_LEVEL option

  • When the callback executes:

API Usage

High Priority Interrupt IPC callbacks have the following restrictions:

  • The callback must be of type void func(void *arg) but implemented entirely in assembly

  • The callback is invoked via the CALLX0 instruction with register windowing disabled, thus the callback:
    • Must not call any register window related instructions (e.g., entry and retw).

    • Must not call other C functions as register windowing is disabled

  • The callback should be placed in IRAM at a 4-byte aligned address

  • (On invocation of/after returning from) the callback, the registers a2, a3, a4 are (saved/restored) automatically thus can be used in the callback. The callback should ONLY use those registers.
    • a2 will contain the void *arg of the callback

    • a3/a4 are free to use as scratch registers

The IPC feature offers the API listed below to execute a callback in a High Priority Interrupt context.

  • esp_ipc_isr_asm_call() will trigger an IPC call on the target CPU. This function will busy-wait until the target CPU begins execution of the callback.

  • esp_ipc_isr_asm_call_blocking() will trigger an IPC call on the target CPU. This function will busy-wait until the target CPU completes execution of the callback.

The following code-blocks demonstrates a High Priority Interrupt IPC callback written in assembly that simply reads the target CPU’s cycle count.

/* esp_test_ipc_isr_get_cycle_count_other_cpu(void *arg) */
// this function reads CCOUNT of the target CPU and stores it in arg.
// use only a2, a3 and a4 regs here.
.section    .iram1, "ax"
.align      4
.global     esp_test_ipc_isr_get_cycle_count_other_cpu
.type       esp_test_ipc_isr_get_cycle_count_other_cpu, @function
// Args:
// a2 - void* arg
esp_test_ipc_isr_get_cycle_count_other_cpu:
rsr.ccount a3
s32i    a3, a2, 0
ret
unit32_t cycle_count;
esp_ipc_isr_asm_call_blocking(esp_test_ipc_isr_get_cycle_count_other_cpu, (void *)cycle_count);

备注

The number of scratch registers available for use is sufficient for most simple use cases. But if your callback requires more scratch registers, void *arg can point to a buffer that is used as a register save area. The callback can then save and restore more registers. See the system/ipc/ipc_isr.

备注

For more examples of High Priority Interrupt IPC callbacks, see components/esp_ipc/src/esp_ipc_isr/esp_ipc_isr_routines.S and :components/esp_system/test/test_ipc_isr.S

The High Priority Interrupt IPC API also provides the following convenience functions that can stall/resume the target CPU. These API utilize the High Priority Interrupt IPC, but supply their own internal callbacks:

API Reference

Functions

esp_err_t esp_ipc_call(uint32_t cpu_id, esp_ipc_func_t func, void *arg)

Execute a callback on a given CPU.

Execute a given callback on a particular CPU. The callback must be of type “esp_ipc_func_t” and will be invoked in the context of the target CPU’s IPC task.

  • This function will block the target CPU’s IPC task has begun execution of the callback

  • If another IPC call is ongoing, this function will block until the ongoing IPC call completes

  • The stack size of the IPC task can be configured via the CONFIG_ESP_IPC_TASK_STACK_SIZE option

备注

In single-core mode, returns ESP_ERR_INVALID_ARG for cpu_id 1.

参数
  • cpu_id[in] CPU where the given function should be executed (0 or 1)

  • func[in] Pointer to a function of type void func(void* arg) to be executed

  • arg[in] Arbitrary argument of type void* to be passed into the function

返回

  • ESP_ERR_INVALID_ARG if cpu_id is invalid

  • ESP_ERR_INVALID_STATE if the FreeRTOS scheduler is not running

  • ESP_OK otherwise

esp_err_t esp_ipc_call_blocking(uint32_t cpu_id, esp_ipc_func_t func, void *arg)

Execute a callback on a given CPU until and block until it completes.

This function is identical to esp_ipc_call() except that this function will block until the execution of the callback completes.

备注

In single-core mode, returns ESP_ERR_INVALID_ARG for cpu_id 1.

参数
  • cpu_id[in] CPU where the given function should be executed (0 or 1)

  • func[in] Pointer to a function of type void func(void* arg) to be executed

  • arg[in] Arbitrary argument of type void* to be passed into the function

返回

  • ESP_ERR_INVALID_ARG if cpu_id is invalid

  • ESP_ERR_INVALID_STATE if the FreeRTOS scheduler is not running

  • ESP_OK otherwise

Type Definitions

typedef void (*esp_ipc_func_t)(void *arg)

IPC Callback.

A callback of this type should be provided as an argument when calling esp_ipc_call() or esp_ipc_call_blocking().

Functions

void esp_ipc_isr_asm_call(esp_ipc_isr_func_t func, void *arg)

Execute an assembly callback on the other CPU.

Execute a given callback on the other CPU in the context of a High Priority Interrupt.

  • This function will busy-wait in a critical section until the other CPU has started execution of the callback

  • The callback must be written in assembly, is invoked using a CALLX0 instruction, and has a2, a3, a4 as scratch registers. See docs for more details

备注

This function is not available in single-core mode.

参数
  • func[in] Pointer to a function of type void func(void* arg) to be executed

  • arg[in] Arbitrary argument of type void* to be passed into the function

void esp_ipc_isr_asm_call_blocking(esp_ipc_isr_func_t func, void *arg)

Execute an assembly callback on the other CPU and busy-wait until it completes.

This function is identical to esp_ipc_isr_asm_call() except that this function will busy-wait until the execution of the callback completes.

备注

This function is not available in single-core mode.

参数
  • func[in] Pointer to a function of type void func(void* arg) to be executed

  • arg[in] Arbitrary argument of type void* to be passed into the function

void esp_ipc_isr_stall_other_cpu(void)

Stall the other CPU.

This function will stall the other CPU. The other CPU is stalled by busy-waiting in the context of a High Priority Interrupt. The other CPU will not be resumed until esp_ipc_isr_release_other_cpu() is called.

  • This function is internally implemented using IPC ISR

  • This function is used for DPORT workaround.

  • If the stall feature is paused using esp_ipc_isr_stall_pause(), this function will have no effect

备注

This function is not available in single-core mode.

void esp_ipc_isr_release_other_cpu(void)

Release the other CPU.

This function will release the other CPU that was previously stalled from calling esp_ipc_isr_stall_other_cpu()

  • This function is used for DPORT workaround.

  • If the stall feature is paused using esp_ipc_isr_stall_pause(), this function will have no effect

备注

This function is not available in single-core mode.

void esp_ipc_isr_stall_pause(void)

Puase the CPU stall feature.

This function will pause the CPU stall feature. Once paused, calls to esp_ipc_isr_stall_other_cpu() and esp_ipc_isr_release_other_cpu() will have no effect. If a IPC ISR call is already in progress, this function will busy-wait until the call completes before pausing the CPU stall feature.

void esp_ipc_isr_stall_abort(void)

Abort a CPU stall.

This function will abort any stalling routine of the other CPU due to a pervious call to esp_ipc_isr_stall_other_cpu(). This function aborts the stall in a non-recoverable manner, thus should only be called in case of a panic().

  • This function is used in panic handling code

void esp_ipc_isr_stall_resume(void)

Resume the CPU stall feature.

This function will resume the CPU stall feature that was previously paused by calling esp_ipc_isr_stall_pause(). Once resumed, calls to esp_ipc_isr_stall_other_cpu() and esp_ipc_isr_release_other_cpu() will have effect again.

Type Definitions

typedef void (*esp_ipc_isr_func_t)(void *arg)

IPC ISR Callback.

A callback of this type should be provided as an argument when calling esp_ipc_isr_asm_call() or esp_ipc_isr_asm_call_blocking().