feat(ota): 实现GD32 OTA包间背压机制，优化传输节奏控制
移除固定的包间延迟配置 `OTA_PACKET_DELAY_MS`，采用基于SPI Link v2 CREDIT背压的动态节奏控制，在SpiMaster中通过remoteCredit==0时wait/拒发 实现流量控制，提高OTA传输稳定性。 refactor(stack): 统一任务栈大小配置并添加CPU统计功能 - 在app_config.h中集中定义所有任务栈大小常量，基于实测水位优化分配 - 新增monitor/cpu_stats.cpp组件，提供1Hz FreeRTOS CPU占用采样 - 更新main.cpp中的任务创建使用新的栈大小常量 - 调整health_monitor和ota_relay等任务的栈分配 perf(ota): 改进ESP32和GD32 OTA分区预擦策略 - ESP32 OTA从边写边擦改为整分区预擦模式，避免写入时的扇区擦除阻塞 - GD32 OTA在prepare阶段进行整分区预擦，提升后续写入吞吐量 - 优化OTA协调器状态管理，修复FINALIZING状态残留问题 fix(log): 改进UDP日志发送的堆内存管理 - 当堆内存不足时记录UDP发送跳过的告警信息到ring buffer - 降低UDP发送跳过的内存阈值从10KB到6KB - 降低日志排水最小内存阈值从8KB到4KB docs(readme): 更新OTA配置参数文档说明 移除已废弃的OTA_PACKET_DELAY_MS参数，更新包间节奏控制说明为 基于CREDIT背压机制的描述。
This commit is contained in:
小龙虾
+15
-181
@@ -1,6 +1,5 @@
|
||||
/*
|
||||
* http_handlers.cpp - HTTP 请求处理器实现
|
||||
* 从 wifi_ota.cpp 提取的 HTTP handler 函数
|
||||
*/
|
||||
#include <string.h>
|
||||
#include <stdarg.h>
|
||||
@@ -28,6 +27,7 @@
|
||||
#include "storage/gd32_flash_store.h"
|
||||
#include "health_monitor.h"
|
||||
#include "monitor/debug_log.h"
|
||||
#include "monitor/cpu_stats.h"
|
||||
#include "spi_master.h"
|
||||
#include "canfd_bridge.h"
|
||||
#include "monitor/can_feedback_task.h"
|
||||
@@ -208,7 +208,7 @@ esp_err_t ota_upload_handler(httpd_req_t *req)
|
||||
DEBUG_INFO("WIFI", "收到固件上传请求, 内容长度: %d 字节", req->content_len);
|
||||
|
||||
esp_wifi_set_ps(WIFI_PS_NONE);
|
||||
// ★ L1: WiFi OTA 开始时恢复 BLE 共存状态（通过协调器）
|
||||
// WiFi OTA 开始时恢复 BLE 共存状态（通过协调器）
|
||||
OtaCoordinator::instance().notifyBleCoexEnd();
|
||||
|
||||
char peek;
|
||||
@@ -307,7 +307,7 @@ esp_err_t ota_upload_handler(httpd_req_t *req)
|
||||
const uint32_t free_pre = esp_get_free_heap_size();
|
||||
if (free_pre < 8192U) {
|
||||
(void)esp_task_wdt_reset();
|
||||
vTaskDelay(pdMS_TO_TICKS(free_pre < 4096U ? 5U : 2U));
|
||||
vTaskDelay(pdMS_TO_TICKS(free_pre < 4096U ? 5U : 1));
|
||||
}
|
||||
}
|
||||
|
||||
@@ -615,12 +615,10 @@ esp_err_t can_feedback_handler(httpd_req_t *req)
|
||||
valid[i] = getLatestCanFeedback(i, &results[i]);
|
||||
}
|
||||
|
||||
/* 估算缓冲大小：每通道 ~2KB (统计+ID条目) */
|
||||
char *resp = (char *)calloc(1, 8192);
|
||||
if (!resp) {
|
||||
httpd_resp_send_err(req, HTTPD_500_INTERNAL_SERVER_ERROR, "no memory");
|
||||
return ESP_FAIL;
|
||||
}
|
||||
/* ★ 修复：改用静态缓冲区，避免每次请求 malloc 8KB 加剧堆碎片化 */
|
||||
static char s_feedback_json[8192];
|
||||
char *resp = s_feedback_json;
|
||||
memset(resp, 0, sizeof(s_feedback_json));
|
||||
|
||||
char *p = resp;
|
||||
p += sprintf(p, "{\"channels\":[");
|
||||
@@ -670,7 +668,6 @@ esp_err_t can_feedback_handler(httpd_req_t *req)
|
||||
|
||||
httpd_resp_set_type(req, "application/json");
|
||||
httpd_resp_send(req, resp, (int)(p - resp));
|
||||
free(resp);
|
||||
return ESP_OK;
|
||||
}
|
||||
|
||||
@@ -882,7 +879,6 @@ esp_err_t spi_send_handler(httpd_req_t *req)
|
||||
static constexpr uint32_t INTERNAL_HEAP_CAPS = MALLOC_CAP_INTERNAL | MALLOC_CAP_8BIT;
|
||||
/* 状态 JSON 用静态缓冲：OTA 后 largest free 常 <2KB，calloc(8KB) 会 500 "no memory"。 */
|
||||
static constexpr size_t SYSTEM_STATUS_JSON_CAP = 6144;
|
||||
static constexpr size_t CPU_TASK_SNAPSHOT_MAX = 32;
|
||||
static char s_status_json[SYSTEM_STATUS_JSON_CAP];
|
||||
static SemaphoreHandle_t s_status_json_mux = nullptr;
|
||||
static portMUX_TYPE s_status_json_init_mux = portMUX_INITIALIZER_UNLOCKED;
|
||||
@@ -899,167 +895,6 @@ static SemaphoreHandle_t statusJsonMutex(void)
|
||||
return s_status_json_mux;
|
||||
}
|
||||
|
||||
struct CpuTaskSample {
|
||||
UBaseType_t task_number;
|
||||
configRUN_TIME_COUNTER_TYPE runtime;
|
||||
};
|
||||
|
||||
struct CpuTaskMetric {
|
||||
const char* name;
|
||||
UBaseType_t task_number;
|
||||
uint64_t runtime_delta;
|
||||
double usage_pct;
|
||||
};
|
||||
|
||||
struct CpuUsageSnapshot {
|
||||
bool available;
|
||||
bool valid;
|
||||
uint32_t sample_ms;
|
||||
double usage_pct;
|
||||
double idle_pct;
|
||||
CpuTaskMetric tasks[CPU_TASK_SNAPSHOT_MAX];
|
||||
size_t task_count;
|
||||
};
|
||||
|
||||
static portMUX_TYPE s_cpu_stats_mux = portMUX_INITIALIZER_UNLOCKED;
|
||||
static CpuTaskSample s_cpu_last_samples[CPU_TASK_SNAPSHOT_MAX];
|
||||
static size_t s_cpu_last_count = 0;
|
||||
static configRUN_TIME_COUNTER_TYPE s_cpu_last_total = 0;
|
||||
static int64_t s_cpu_last_time_us = 0;
|
||||
static bool s_cpu_has_last = false;
|
||||
|
||||
static uint64_t runtime_delta(configRUN_TIME_COUNTER_TYPE current, configRUN_TIME_COUNTER_TYPE previous)
|
||||
{
|
||||
return current >= previous ? (uint64_t)(current - previous) : 0U;
|
||||
}
|
||||
|
||||
static bool task_name_is_idle(const char* name)
|
||||
{
|
||||
return name && strncmp(name, "IDLE", 4) == 0;
|
||||
}
|
||||
|
||||
static void reset_cpu_usage_baseline_locked(const TaskStatus_t* tasks, size_t task_count,
|
||||
configRUN_TIME_COUNTER_TYPE total_runtime,
|
||||
int64_t now_us)
|
||||
{
|
||||
size_t copy_count = std::min(task_count, CPU_TASK_SNAPSHOT_MAX);
|
||||
for (size_t i = 0; i < copy_count; ++i) {
|
||||
s_cpu_last_samples[i].task_number = tasks[i].xTaskNumber;
|
||||
s_cpu_last_samples[i].runtime = tasks[i].ulRunTimeCounter;
|
||||
}
|
||||
s_cpu_last_count = copy_count;
|
||||
s_cpu_last_total = total_runtime;
|
||||
s_cpu_last_time_us = now_us;
|
||||
s_cpu_has_last = true;
|
||||
}
|
||||
|
||||
// ★ 改为输出参数模式：CpuUsageSnapshot 608B + CpuTaskSample[24] 192B = 800B
|
||||
// 之前全在栈上，system_status_handler 又拷一份 608B → 叠加 1.4KB，
|
||||
// 加上 OtaSnapshot/TaskStackMetric/wifi_ap_record_t 约 2.3KB 栈消耗。
|
||||
// HTTP 主线程上传 100% 后调用此 handler 时极易栈溢出。改堆分配。
|
||||
static void sample_cpu_usage(CpuUsageSnapshot& result)
|
||||
{
|
||||
result = {};
|
||||
#if defined(CONFIG_FREERTOS_GENERATE_RUN_TIME_STATS)
|
||||
result.available = true;
|
||||
|
||||
UBaseType_t task_count = uxTaskGetNumberOfTasks();
|
||||
if (task_count == 0) {
|
||||
return;
|
||||
}
|
||||
task_count = std::min<UBaseType_t>(task_count + 2U, CPU_TASK_SNAPSHOT_MAX);
|
||||
|
||||
TaskStatus_t* tasks = (TaskStatus_t*)calloc(task_count, sizeof(TaskStatus_t));
|
||||
if (!tasks) {
|
||||
return;
|
||||
}
|
||||
|
||||
configRUN_TIME_COUNTER_TYPE total_runtime = 0;
|
||||
UBaseType_t captured_count = uxTaskGetSystemState(tasks, task_count, &total_runtime);
|
||||
int64_t now_us = esp_timer_get_time();
|
||||
if (captured_count == 0 || captured_count > task_count) {
|
||||
DEBUG_WARN("STATUS", "CPU stats: uxTaskGetSystemState 返回 0 (actual_tasks=%lu, buf_slots=%lu)",
|
||||
(unsigned long)uxTaskGetNumberOfTasks(), (unsigned long)task_count);
|
||||
free(tasks);
|
||||
return;
|
||||
}
|
||||
|
||||
// ★ previous_samples[24] = 192B，改堆分配减少栈压力
|
||||
CpuTaskSample* previous_samples = (CpuTaskSample*)calloc(CPU_TASK_SNAPSHOT_MAX, sizeof(CpuTaskSample));
|
||||
if (!previous_samples) {
|
||||
free(tasks);
|
||||
return;
|
||||
}
|
||||
|
||||
size_t previous_count = 0;
|
||||
configRUN_TIME_COUNTER_TYPE previous_total = 0;
|
||||
int64_t previous_time_us = 0;
|
||||
bool had_previous = false;
|
||||
|
||||
portENTER_CRITICAL(&s_cpu_stats_mux);
|
||||
previous_count = s_cpu_last_count;
|
||||
previous_total = s_cpu_last_total;
|
||||
previous_time_us = s_cpu_last_time_us;
|
||||
had_previous = s_cpu_has_last;
|
||||
for (size_t i = 0; i < previous_count; ++i) {
|
||||
previous_samples[i] = s_cpu_last_samples[i];
|
||||
}
|
||||
reset_cpu_usage_baseline_locked(tasks, captured_count, total_runtime, now_us);
|
||||
portEXIT_CRITICAL(&s_cpu_stats_mux);
|
||||
|
||||
bool can_compute = had_previous && total_runtime >= previous_total && now_us > previous_time_us;
|
||||
uint64_t total_delta = can_compute ? runtime_delta(total_runtime, previous_total) : 0U;
|
||||
uint64_t idle_delta = 0U;
|
||||
if (can_compute && total_delta > 0U) {
|
||||
for (UBaseType_t i = 0; i < captured_count; ++i) {
|
||||
configRUN_TIME_COUNTER_TYPE last_runtime = 0;
|
||||
bool found_previous = false;
|
||||
for (size_t j = 0; j < previous_count; ++j) {
|
||||
if (previous_samples[j].task_number == tasks[i].xTaskNumber) {
|
||||
last_runtime = previous_samples[j].runtime;
|
||||
found_previous = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
if (!found_previous) {
|
||||
continue;
|
||||
}
|
||||
uint64_t delta = runtime_delta(tasks[i].ulRunTimeCounter, last_runtime);
|
||||
double pct = (double)delta * 100.0 / (double)total_delta;
|
||||
if (task_name_is_idle(tasks[i].pcTaskName)) {
|
||||
idle_delta += delta;
|
||||
}
|
||||
if (result.task_count < CPU_TASK_SNAPSHOT_MAX) {
|
||||
CpuTaskMetric& metric = result.tasks[result.task_count++];
|
||||
metric.name = tasks[i].pcTaskName ? tasks[i].pcTaskName : "";
|
||||
metric.task_number = tasks[i].xTaskNumber;
|
||||
metric.runtime_delta = delta;
|
||||
metric.usage_pct = pct;
|
||||
}
|
||||
}
|
||||
|
||||
result.valid = true;
|
||||
result.sample_ms = (uint32_t)((now_us - previous_time_us) / 1000);
|
||||
result.idle_pct = std::min(100.0, (double)idle_delta * 100.0 / (double)total_delta);
|
||||
result.usage_pct = std::max(0.0, 100.0 - result.idle_pct);
|
||||
}
|
||||
|
||||
if (!result.valid) {
|
||||
DEBUG_DBG("STATUS", "CPU stats: valid=false had_prev=%d total=%lu prev_total=%lu "
|
||||
"can_compute=%d delta=%lu captured=%lu",
|
||||
(int)had_previous,
|
||||
(unsigned long)total_runtime, (unsigned long)previous_total,
|
||||
(int)can_compute, (unsigned long)total_delta,
|
||||
(unsigned long)captured_count);
|
||||
}
|
||||
|
||||
free(previous_samples);
|
||||
free(tasks);
|
||||
#else
|
||||
result.available = false;
|
||||
#endif
|
||||
}
|
||||
|
||||
static size_t get_task_stack_high_watermark_bytes(TaskHandle_t handle)
|
||||
{
|
||||
if (handle == NULL) return 0;
|
||||
@@ -1089,6 +924,7 @@ static int append_jsonf(char* buf, size_t cap, int offset, const char* fmt, ...)
|
||||
return offset + (int)written;
|
||||
}
|
||||
|
||||
/** 将 CpuStats 1Hz 快照写入 JSON（不在 HTTP 线程采样） */
|
||||
static int append_cpu_status_json(char* buf, size_t cap, int offset, const CpuUsageSnapshot& cpu)
|
||||
{
|
||||
offset = append_jsonf(buf, cap, offset,
|
||||
@@ -1168,7 +1004,9 @@ esp_err_t system_status_handler(httpd_req_t *req)
|
||||
connected = true;
|
||||
}
|
||||
|
||||
// ★ PR1: 一次性快照，替代散读 OtaCoordinator / OtaRelay 多个 getter
|
||||
// 先自愈孤儿忙态（如 SHA 失败后卡在 ESP32_FINALIZING），再取快照给 UI
|
||||
OtaCoordinator::instance().healOrphanStates();
|
||||
// 一次性快照，替代散读 OtaCoordinator / OtaRelay 多个 getter
|
||||
OtaSnapshot snap = OtaCoordinator::instance().getSnapshot();
|
||||
|
||||
size_t free_heap = esp_get_free_heap_size();
|
||||
@@ -1190,20 +1028,17 @@ esp_err_t system_status_handler(httpd_req_t *req)
|
||||
}
|
||||
}
|
||||
|
||||
/* CPU 采样在堆上；失败则跳过（available=false），不再 500 */
|
||||
/* CPU：只读 cpu_stats 任务的 1Hz 快照，禁止在 httpd 里 uxTaskGetSystemState */
|
||||
CpuUsageSnapshot cpu_local = {};
|
||||
CpuUsageSnapshot *cpu = &cpu_local;
|
||||
CpuUsageSnapshot *cpu_heap = nullptr;
|
||||
if (largest_free_block >= (sizeof(CpuUsageSnapshot) + 1024U)) {
|
||||
if (largest_free_block >= (sizeof(CpuUsageSnapshot) + 512U)) {
|
||||
cpu_heap = (CpuUsageSnapshot *)calloc(1, sizeof(CpuUsageSnapshot));
|
||||
if (cpu_heap) {
|
||||
cpu = cpu_heap;
|
||||
sample_cpu_usage(*cpu);
|
||||
}
|
||||
} else {
|
||||
cpu_local.available = false;
|
||||
cpu_local.valid = false;
|
||||
}
|
||||
CpuStats::instance().getSnapshot(*cpu);
|
||||
|
||||
char *resp = s_status_json;
|
||||
memset(resp, 0, SYSTEM_STATUS_JSON_CAP);
|
||||
@@ -1309,11 +1144,10 @@ esp_err_t system_status_handler(httpd_req_t *req)
|
||||
}
|
||||
|
||||
// ============================================================
|
||||
// GET /system/probe — GD32 APP 单次 SPI 探测（PR2）
|
||||
// GET /system/probe — GD32 APP 单次 SPI 探测
|
||||
// ============================================================
|
||||
/**
|
||||
* @brief 显式触发一次 GD32 APP 存活探测。
|
||||
* 从 system_status_handler 移出的副作用，改为独立端点。
|
||||
* 成功时更新 OtaRelay 缓存的 alive/version 并返回版本信息。
|
||||
*/
|
||||
esp_err_t system_probe_handler(httpd_req_t *req)
|
||||
|
||||
Reference in New Issue
Block a user