Ken Van Hoeylandt 74127a5f6c
Add kernel drivers for SPI and UART and make locking APIs more consistent (#489)
- Add kernel support for SPI driver
- Add kernel support for UART driver
- Implemented ESP32 UART kernel driver
- Update existing UART-related code in Tactility to use new kernel driver
- Remove UART from tt::hal::Configuration
- Remove tt_hal_uart functionality but keep functions for now
- Update devicetrees for UART changes
- Kernel mutex and recursive mutex: improved locking API design
- Other kernel improvements
- Added device_exists_of_type() and device_find_by_name()
2026-02-07 21:28:11 +01:00

142 lines
3.7 KiB
C++

// SPDX-License-Identifier: Apache-2.0
#include <queue>
#include <tactility/concurrent/dispatcher.h>
#include "tactility/error.h"
#include <atomic>
#include <tactility/concurrent/event_group.h>
#include <tactility/concurrent/mutex.h>
#include <tactility/log.h>
#define TAG "Dispatcher"
static constexpr EventBits_t BACKPRESSURE_WARNING_COUNT = 100U;
static constexpr EventBits_t WAIT_FLAG = 1U;
struct QueuedItem {
DispatcherCallback callback;
void* context;
};
struct DispatcherData {
Mutex mutex = { 0 };
std::queue<QueuedItem> queue = {};
EventGroupHandle_t eventGroup = nullptr;
std::atomic<bool> shutdown{false}; // TODO: Use EventGroup
DispatcherData() {
event_group_construct(&eventGroup);
mutex_construct(&mutex);
}
~DispatcherData() {
event_group_destruct(&eventGroup);
mutex_destruct(&mutex);
}
};
#define dispatcher_data(handle) static_cast<DispatcherData*>(handle)
extern "C" {
DispatcherHandle_t dispatcher_alloc(void) {
return new DispatcherData();
}
void dispatcher_free(DispatcherHandle_t dispatcher) {
auto* data = dispatcher_data(dispatcher);
data->shutdown.store(true, std::memory_order_release);
mutex_lock(&data->mutex);
mutex_unlock(&data->mutex);
delete data;
}
error_t dispatcher_dispatch_timed(DispatcherHandle_t dispatcher, void* callbackContext, DispatcherCallback callback, TickType_t timeout) {
auto* data = dispatcher_data(dispatcher);
// Mutate
if (!mutex_try_lock(&data->mutex, timeout)) {
#ifdef ESP_PLATFORM
LOG_E(TAG, "Mutex acquisition timeout");
#endif
return ERROR_TIMEOUT;
}
if (data->shutdown.load(std::memory_order_acquire)) {
mutex_unlock(&data->mutex);
return ERROR_INVALID_STATE;
}
data->queue.push({
.callback = callback,
.context = callbackContext
});
if (data->queue.size() == BACKPRESSURE_WARNING_COUNT) {
#ifdef ESP_PLATFORM
LOG_W(TAG, "Backpressure: You're not consuming fast enough (100 queued)");
#endif
}
mutex_unlock(&data->mutex);
if (event_group_set(data->eventGroup, WAIT_FLAG) != ERROR_NONE) {
#ifdef ESP_PLATFORM
LOG_E(TAG, "Failed to set flag");
#endif
return ERROR_RESOURCE;
}
return ERROR_NONE;
}
error_t dispatcher_consume_timed(DispatcherHandle_t dispatcher, TickType_t timeout) {
auto* data = dispatcher_data(dispatcher);
// TODO: keep track of time and consider the timeout input as total timeout
// Wait for signal
error_t error = event_group_wait(data->eventGroup, WAIT_FLAG, false, true, nullptr, timeout);
if (error != ERROR_NONE) {
if (error == ERROR_TIMEOUT) {
return ERROR_TIMEOUT;
} else {
return ERROR_RESOURCE;
}
}
if (data->shutdown.load(std::memory_order_acquire)) {
return ERROR_INVALID_STATE;
}
// Mutate
bool processing = true;
do {
if (mutex_try_lock(&data->mutex, 10)) {
if (!data->queue.empty()) {
// Make a copy, so it's thread-safe when we unlock
auto entry = data->queue.front();
data->queue.pop();
processing = !data->queue.empty();
// Don't keep lock as callback might be slow and we want to allow dispatch in the meanwhile
mutex_unlock(&data->mutex);
entry.callback(entry.context);
} else {
processing = false;
mutex_unlock(&data->mutex);
}
} else {
#ifdef ESP_PLATFORM
LOG_W(TAG, "Mutex acquisition timeout");
#endif
}
} while (processing && !data->shutdown.load(std::memory_order_acquire));
return ERROR_NONE;
}
}