MultiHal multithreaded polling

Change-Id: I3ebe380169eed1c8deeca2860d1788be6c14837e
This commit is contained in:
Aaron Whyte
2013-10-28 17:18:06 -07:00
committed by Mike Lockwood
parent ab6ec384c4
commit 92863c14b7
4 changed files with 146 additions and 157 deletions

View File

@@ -17,55 +17,28 @@
#include <hardware/sensors.h>
#include <algorithm>
#include <pthread.h>
#include <linux/input.h>
#include <cutils/atomic.h>
#include <cutils/log.h>
#include "SensorEventQueue.h"
SensorEventQueue::SensorEventQueue(int capacity) {
mCapacity = capacity;
mStart = 0;
mSize = 0;
mData = new sensors_event_t[mCapacity];
pthread_cond_init(&mDataAvailableCondition, NULL);
pthread_cond_init(&mSpaceAvailableCondition, NULL);
pthread_mutex_init(&mMutex, NULL);
}
SensorEventQueue::~SensorEventQueue() {
delete[] mData;
mData = NULL;
pthread_cond_destroy(&mDataAvailableCondition);
pthread_cond_destroy(&mSpaceAvailableCondition);
pthread_mutex_destroy(&mMutex);
}
void SensorEventQueue::lock() {
pthread_mutex_lock(&mMutex);
}
void SensorEventQueue::unlock() {
pthread_mutex_unlock(&mMutex);
}
void SensorEventQueue::waitForSpaceAndLock() {
lock();
while (mSize >= mCapacity) {
pthread_cond_wait(&mSpaceAvailableCondition, &mMutex);
}
}
void SensorEventQueue::waitForDataAndLock() {
lock();
while (mSize <= 0) {
pthread_cond_wait(&mDataAvailableCondition, &mMutex);
}
}
int SensorEventQueue::getWritableRegion(int requestedLength, sensors_event_t** out) {
if (mSize >= mCapacity || requestedLength <= 0) {
if (mSize == mCapacity || requestedLength <= 0) {
*out = NULL;
return 0;
}
@@ -88,9 +61,6 @@ int SensorEventQueue::getWritableRegion(int requestedLength, sensors_event_t** o
void SensorEventQueue::markAsWritten(int count) {
mSize += count;
if (mSize) {
pthread_cond_broadcast(&mDataAvailableCondition);
}
}
int SensorEventQueue::getSize() {
@@ -98,13 +68,21 @@ int SensorEventQueue::getSize() {
}
sensors_event_t* SensorEventQueue::peek() {
if (mSize <= 0) return NULL;
if (mSize == 0) return NULL;
return &mData[mStart];
}
void SensorEventQueue::dequeue() {
if (mSize <= 0) return;
if (mSize == 0) return;
if (mSize == mCapacity) {
pthread_cond_broadcast(&mSpaceAvailableCondition);
}
mSize--;
mStart = (mStart + 1) % mCapacity;
pthread_cond_broadcast(&mSpaceAvailableCondition);
}
void SensorEventQueue::waitForSpace(pthread_mutex_t* mutex) {
while (mSize == mCapacity) {
pthread_cond_wait(&mSpaceAvailableCondition, mutex);
}
}

View File

@@ -35,17 +35,11 @@ class SensorEventQueue {
int mStart; // start of readable region
int mSize; // number of readable items
sensors_event_t* mData;
pthread_cond_t mDataAvailableCondition;
pthread_cond_t mSpaceAvailableCondition;
pthread_mutex_t mMutex;
public:
SensorEventQueue(int capacity);
~SensorEventQueue();
void lock();
void unlock();
void waitForSpaceAndLock();
void waitForDataAndLock();
// Returns length of region, between zero and min(capacity, requestedLength). If there is any
// writable space, it will return a region of at least one. Because it must return
@@ -73,6 +67,9 @@ public:
// This will decrease the size by one, freeing up the oldest readable event's slot for writing.
// Only call while holding the lock.
void dequeue();
// Blocks until space is available. No-op if there is already space.
void waitForSpace(pthread_mutex_t* mutex);
};
#endif // SENSOREVENTQUEUE_H_

View File

@@ -30,6 +30,7 @@
#include <stdio.h>
#include <dlfcn.h>
#include <SensorEventQueue.h>
// comment out to disable debug-level logging
#define LOG_NDEBUG 0
@@ -41,6 +42,13 @@ static const int MAX_CONF_LINE_LENGTH = 1024;
static pthread_mutex_t init_modules_mutex = PTHREAD_MUTEX_INITIALIZER;
static pthread_mutex_t init_sensors_mutex = PTHREAD_MUTEX_INITIALIZER;
// This mutex is shared by all queues
static pthread_mutex_t queue_mutex = PTHREAD_MUTEX_INITIALIZER;
// Used to pause the multihal poll(). Broadcasted by sub-polling tasks if waiting_for_data.
static pthread_cond_t data_available_cond = PTHREAD_COND_INITIALIZER;
bool waiting_for_data = false;
/*
* Vector of sub modules, whose indexes are referred to ni this file as module_index.
*/
@@ -65,7 +73,7 @@ struct FullHandle {
return localHandle < that.localHandle;
}
bool operator=(const FullHandle &that) const {
bool operator==(const FullHandle &that) const {
return moduleIndex == that.moduleIndex && localHandle == that.localHandle;
}
};
@@ -75,13 +83,12 @@ std::map<FullHandle, int> full_to_global;
int next_global_handle = 1;
static int assign_global_handle(int module_index, int local_handle) {
ALOGD("assign_global_handle %d %d", module_index, local_handle);
int global_handle = next_global_handle++;
FullHandle *full_handle = new FullHandle();
full_handle->moduleIndex = module_index;
full_handle->localHandle = local_handle;
full_to_global[*full_handle] = global_handle;
global_to_full[global_handle] = *full_handle;
FullHandle full_handle;
full_handle.moduleIndex = module_index;
full_handle.localHandle = local_handle;
full_to_global[full_handle] = global_handle;
global_to_full[global_handle] = full_handle;
return global_handle;
}
@@ -90,12 +97,53 @@ static int get_local_handle(int global_handle) {
}
static int get_module_index(int global_handle) {
ALOGD("get_module_index %d", global_handle);
ALOGD("get_module_index for global_handle %d", global_handle);
FullHandle f = global_to_full[global_handle];
ALOGD("FullHandle moduleIndex %d, localHandle %d", f.moduleIndex, f.localHandle);
return f.moduleIndex;
}
static const int SENSOR_EVENT_QUEUE_CAPACITY = 20;
struct TaskContext {
sensors_poll_device_t* device;
SensorEventQueue* queue;
};
void *writerTask(void* ptr) {
ALOGD("writerTask STARTS");
TaskContext* ctx = (TaskContext*)ptr;
sensors_poll_device_t* device = ctx->device;
SensorEventQueue* queue = ctx->queue;
sensors_event_t* buffer;
int eventsPolled;
while (1) {
ALOGD("writerTask before lock 1");
pthread_mutex_lock(&queue_mutex);
ALOGD("writerTask before waitForSpace");
queue->waitForSpace(&queue_mutex);
ALOGD("writerTask after waitForSpace");
int bufferSize = queue->getWritableRegion(SENSOR_EVENT_QUEUE_CAPACITY, &buffer);
// Do blocking poll outside of lock
pthread_mutex_unlock(&queue_mutex);
ALOGD("writerTask before poll() - bufferSize = %d", bufferSize);
eventsPolled = device->poll(device, buffer, bufferSize);
ALOGD("writerTask poll() got %d events.", eventsPolled);
ALOGD("writerTask before lock 2");
pthread_mutex_lock(&queue_mutex);
queue->markAsWritten(eventsPolled);
ALOGD("writerTask wrote %d events", eventsPolled);
if (waiting_for_data) {
ALOGD("writerTask - broadcast data_available_cond");
pthread_cond_broadcast(&data_available_cond);
}
pthread_mutex_unlock(&queue_mutex);
}
// never actually returns
return NULL;
}
/*
* Cache of all sensors, with original handles replaced by global handles.
@@ -124,15 +172,31 @@ struct sensors_poll_context_t {
int close();
std::vector<hw_device_t*> sub_hw_devices;
std::vector<SensorEventQueue*> queues;
std::vector<pthread_t> threads;
int nextReadIndex;
sensors_poll_device_t* get_v0_device_by_handle(int global_handle);
sensors_poll_device_1_t* get_v1_device_by_handle(int global_handle);
int get_device_version_by_handle(int global_handle);
void copy_event_remap_handle(sensors_event_t* src, sensors_event_t* dest, int sub_index);
};
void sensors_poll_context_t::addSubHwDevice(struct hw_device_t* sub_hw_device) {
ALOGD("addSubHwDevice");
this->sub_hw_devices.push_back(sub_hw_device);
SensorEventQueue *queue = new SensorEventQueue(SENSOR_EVENT_QUEUE_CAPACITY);
this->queues.push_back(queue);
TaskContext* taskContext = new TaskContext();
taskContext->device = (sensors_poll_device_t*) sub_hw_device;
taskContext->queue = queue;
pthread_t writerThread;
pthread_create(&writerThread, NULL, writerTask, taskContext);
this->threads.push_back(writerThread);
}
sensors_poll_device_t* sensors_poll_context_t::get_v0_device_by_handle(int handle) {
@@ -168,34 +232,60 @@ int sensors_poll_context_t::setDelay(int handle, int64_t ns) {
return retval;
}
int sensors_poll_context_t::poll(sensors_event_t *data, int count) {
ALOGD("poll");
// This only gets the first device. Parallel polling of multiple devices is coming soon.
int sub_index = 0;
sensors_poll_device_t* v0 = (sensors_poll_device_t*) this->sub_hw_devices[sub_index];
ALOGD("poll's blocking read begins...");
int retval = v0->poll(v0, data, count);
ALOGD("...poll's blocking read ends");
ALOGD("rewriting %d sensor handles...", retval);
// A normal event's "sensor" field is a local handles. Convert it to a global handle.
void sensors_poll_context_t::copy_event_remap_handle(sensors_event_t* dest, sensors_event_t* src,
int sub_index) {
memcpy(dest, src, sizeof(struct sensors_event_t));
// A normal event's "sensor" field is a local handle. Convert it to a global handle.
// A meta-data event must have its sensor set to 0, but it has a nested event
// with a local handle that needs to be converted to a global handle.
FullHandle full_handle;
full_handle.moduleIndex = sub_index;
for (int i = 0; i < retval; i++) {
sensors_event_t *event = &data[i];
// If it's a metadata event, rewrite the inner payload, not the sensor field.
if (event->type == SENSOR_TYPE_META_DATA) {
full_handle.localHandle = event->meta_data.sensor;
event->meta_data.sensor = full_to_global[full_handle];
} else {
full_handle.localHandle = event->sensor;
event->sensor = full_to_global[full_handle];
// If it's a metadata event, rewrite the inner payload, not the sensor field.
if (dest->type == SENSOR_TYPE_META_DATA) {
full_handle.localHandle = dest->meta_data.sensor;
dest->meta_data.sensor = full_to_global[full_handle];
} else {
full_handle.localHandle = dest->sensor;
dest->sensor = full_to_global[full_handle];
}
}
int sensors_poll_context_t::poll(sensors_event_t *data, int maxReads) {
ALOGD("poll");
int empties = 0;
int queueCount = (int)this->queues.size();
int eventsRead = 0;
pthread_mutex_lock(&queue_mutex);
while (eventsRead == 0) {
while (empties < queueCount && eventsRead < maxReads) {
SensorEventQueue* queue = this->queues.at(this->nextReadIndex);
ALOGD("queue size: %d", queue->getSize());
sensors_event_t* event = queue->peek();
if (event == NULL) {
empties++;
} else {
empties = 0;
this->copy_event_remap_handle(&data[eventsRead++], event, nextReadIndex);
queue->dequeue();
}
this->nextReadIndex = (this->nextReadIndex + 1) % queueCount;
}
if (eventsRead == 0) {
// The queues have been scanned and none contain data.
// Wait for any of them to signal that there's data.
ALOGD("poll stopping to wait for data");
waiting_for_data = true;
pthread_cond_wait(&data_available_cond, &queue_mutex);
waiting_for_data = false;
empties = 0;
ALOGD("poll done waiting for data");
}
}
return retval;
pthread_mutex_unlock(&queue_mutex);
ALOGD("...poll's blocking read ends. Returning %d events.", eventsRead);
return eventsRead;
}
int sensors_poll_context_t::batch(int handle, int flags, int64_t period_ns, int64_t timeout) {
@@ -436,9 +526,13 @@ static void lazy_init_sensors_list() {
static int module__get_sensors_list(struct sensors_module_t* module,
struct sensor_t const** list) {
ALOGD("module__get_sensors_list");
ALOGD("module__get_sensors_list start");
lazy_init_sensors_list();
*list = global_sensors_list;
ALOGD("global_sensors_count: %d", global_sensors_count);
for (int i = 0; i < global_sensors_count; i++) {
ALOGD("sensor type: %d", global_sensors_list[i].type);
}
return global_sensors_count;
}
@@ -480,6 +574,8 @@ static int open_sensors(const struct hw_module_t* hw_module, const char* name,
dev->proxy_device.batch = device__batch;
dev->proxy_device.flush = device__flush;
dev->nextReadIndex = 0;
// Open() the subhal modules. Remember their devices in a vector parallel to sub_hw_modules.
for (std::vector<hw_module_t*>::iterator it = sub_hw_modules->begin();
it != sub_hw_modules->end(); it++) {

View File

@@ -2,6 +2,8 @@
#include <stdlib.h>
#include <hardware/sensors.h>
#include <pthread.h>
#include <cutils/atomic.h>
#include "SensorEventQueue.cpp"
// Unit tests for the SensorEventQueue.
@@ -78,93 +80,9 @@ bool testWrappingWriteSizeCounts() {
return true;
}
static const int TTOQ_EVENT_COUNT = 10000;
struct TaskContext {
bool success;
SensorEventQueue* queue;
};
void* writerTask(void* ptr) {
printf("writerTask starts\n");
TaskContext* ctx = (TaskContext*)ptr;
SensorEventQueue* queue = ctx->queue;
int totalWrites = 0;
sensors_event_t* buffer;
while (totalWrites < TTOQ_EVENT_COUNT) {
queue->waitForSpaceAndLock();
int writableSize = queue->getWritableRegion(rand() % 10 + 1, &buffer);
queue->unlock();
for (int i = 0; i < writableSize; i++) {
// serialize the events
buffer[i].timestamp = totalWrites++;
}
queue->lock();
queue->markAsWritten(writableSize);
queue->unlock();
}
printf("writerTask ends normally\n");
return NULL;
}
void* readerTask(void* ptr) {
printf("readerTask starts\n");
TaskContext* ctx = (TaskContext*)ptr;
SensorEventQueue* queue = ctx->queue;
int totalReads = 0;
while (totalReads < TTOQ_EVENT_COUNT) {
queue->waitForDataAndLock();
int maxReads = rand() % 20 + 1;
int reads = 0;
while (queue->getSize() && reads < maxReads) {
sensors_event_t* event = queue->peek();
if (totalReads != event->timestamp) {
printf("FAILURE: readerTask expected timestamp %d; actual was %d\n",
totalReads, (int)(event->timestamp));
ctx->success = false;
return NULL;
}
queue->dequeue();
totalReads++;
reads++;
}
queue->unlock();
}
printf("readerTask ends normally\n");
return NULL;
}
// Create a short queue, and write and read a ton of data through it.
// Write serial timestamps into the events, and expect to read them in the right order.
bool testTwoThreadsOneQueue() {
printf("TEST testTwoThreadsOneQueue\n");
SensorEventQueue* queue = new SensorEventQueue(100);
TaskContext readerCtx;
readerCtx.success = true;
readerCtx.queue = queue;
TaskContext writerCtx;
writerCtx.success = true;
writerCtx.queue = queue;
pthread_t writer, reader;
pthread_create(&reader, NULL, readerTask, &readerCtx);
pthread_create(&writer, NULL, writerTask, &writerCtx);
pthread_join(writer, NULL);
pthread_join(reader, NULL);
printf("testTwoThreadsOneQueue done\n");
return readerCtx.success && writerCtx.success;
}
int main(int argc, char **argv) {
if (testSimpleWriteSizeCounts() &&
testWrappingWriteSizeCounts() &&
testTwoThreadsOneQueue()) {
testWrappingWriteSizeCounts()) {
printf("ALL PASSED\n");
} else {
printf("SOMETHING FAILED\n");