1
0
mirror of https://bitbucket.org/librepilot/librepilot.git synced 2024-11-29 07:24:13 +01:00

Refactor how new packets are scheduled to avoid trying to start a transaction

from an ISR.  This was causing lock ups.  It's a bit higher latency probably
(trivially slow) but if we aren't keeping up with the Overo transactions then
using a proper DMA double buffer will be required.
This commit is contained in:
James Cotton 2012-02-04 20:10:56 -06:00
parent 4e190f99e2
commit 179e66b754

View File

@ -37,7 +37,7 @@
#define OVEROSYNC_PACKET_SIZE 256
#define MAX_QUEUE_SIZE 3
#define STACK_SIZE_BYTES 512
#define TASK_PRIORITY (tskIDLE_PRIORITY + 2)
#define TASK_PRIORITY (tskIDLE_PRIORITY + 0)
// Private types
@ -67,6 +67,7 @@ struct overosync {
uint32_t loading_transaction_id;
xSemaphoreHandle transaction_lock;
xSemaphoreHandle buffer_lock;
volatile bool transaction_done;
uint32_t write_pointer;
uint32_t sent_objects;
uint32_t failed_objects;
@ -98,7 +99,7 @@ int32_t OveroSyncInitialize(void)
*/
int32_t OveroSyncStart(void)
{
overosync = (struct overosync *) pvPortMalloc(sizeof(overosync));
overosync = (struct overosync *) pvPortMalloc(sizeof(*overosync));
if(overosync == NULL)
return -1;
@ -158,7 +159,7 @@ static void overoSyncTask(void *parameters)
UAVObjEvent ev;
// Kick off SPI transfers (once one is completed another will automatically transmit)
transmitData();
overosync->transaction_done = true;
// Loop forever
while (1) {
@ -166,6 +167,10 @@ static void overoSyncTask(void *parameters)
if (xQueueReceive(queue, &ev, portMAX_DELAY) == pdTRUE) {
// Process event. This calls transmitData
UAVTalkSendObject(uavTalkCon, ev.obj, ev.instId, false, 0);
if(overosync->transaction_done)
transmitData();
overosync_transfers++;
}
}
@ -175,13 +180,18 @@ int32_t transactionsDone = 0;
static void transmitDataDone(bool crc_ok, uint8_t crc_val)
{
uint8_t *rx_buffer;
static signed portBASE_TYPE xHigherPriorityTaskWoken;
transactionsDone ++;
rx_buffer = overosync->transactions[overosync->active_transaction_id].rx_buffer;
// Release the semaphore and start another transaction (which grabs semaphore again but then
// returns instantly)
xSemaphoreGive(overosync->transaction_lock);
transmitData();
// returns instantly). Because this is called by the DMA ISR we need to be aware of context
// switches.
xSemaphoreGiveFromISR(overosync->transaction_lock, &xHigherPriorityTaskWoken);
portEND_SWITCHING_ISR(xHigherPriorityTaskWoken);
overosync->transaction_done = true;
// Parse the data from overo
for (uint32_t i = 0; rx_buffer[0] != 0 && i < sizeof(rx_buffer) ; i++)
@ -202,24 +212,24 @@ static int32_t packData(uint8_t * data, int32_t length)
// Get the lock for manipulating the buffer
xSemaphoreTake(overosync->buffer_lock, portMAX_DELAY);
// Check this packet will fit
if ((overosync->write_pointer + length) > sizeof(overosync->transactions[overosync->loading_transaction_id].tx_buffer)) {
overosync->failed_objects ++;
xSemaphoreGive(overosync->buffer_lock);
return -1;
}
// Get offset into buffer and copy contents
tx_buffer = overosync->transactions[overosync->loading_transaction_id].tx_buffer +
overosync->write_pointer;
memcpy(tx_buffer,data,length);
overosync->write_pointer += length;
overosync->sent_objects++;
xSemaphoreGive(overosync->buffer_lock);
return length;
}
@ -228,7 +238,10 @@ static int32_t transmitData()
uint8_t *tx_buffer, *rx_buffer;
// Get lock to manipulate buffers
xSemaphoreTake(overosync->buffer_lock, portMAX_DELAY);
if(xSemaphoreTake(overosync->buffer_lock, 1) == pdFALSE)
return -1;
overosync->transaction_done = false;
// Swap buffers
overosync->active_transaction_id = overosync->loading_transaction_id;