Reattempt failed flash operations

Author: todd-herbert

libraries/InternalFileSytem/src/flash/flash_nrf5x.c

@@ -44,11 +44,46 @@ extern uint32_t __flash_arduino_start[];
 // MACRO TYPEDEF CONSTANT ENUM DECLARATION
 //--------------------------------------------------------------------+
 static SemaphoreHandle_t _sem = NULL;
+static bool _flash_op_failed = false;
 
 void flash_nrf5x_event_cb (uint32_t event)
 {
-//  if (event != NRF_EVT_FLASH_OPERATION_SUCCESS) LOG_LV1("IFLASH", "Flash op Failed");
-  if ( _sem ) xSemaphoreGive(_sem);
+  if ( _sem ) {
+    // Record the result, for consumption by fal_erase or fal_program
+    // Used to reattempt failed operations
+    _flash_op_failed = (event == NRF_EVT_FLASH_OPERATION_ERROR);
+
+    // Signal to fal_erase or fal_program that our async flash op is now complete
+    xSemaphoreGive(_sem);
+  } 
+}
+
+// How many retry attempts when performing flash operations 
+#define MAX_RETRY 20
+
+// Check whether a flash operation was successful, or should be repeated
+static bool retry_flash_op (uint32_t op_result, bool sd_enabled) {
+  // If busy
+  if (op_result == NRF_ERROR_BUSY) {
+    delay(1);
+    return true; // Retry
+  }
+
+  // If unspecified error
+  if (op_result != NRF_SUCCESS)
+    return true; // Retry
+
+  // If the soft device is enabled, flash operations run async
+  // The callback (flash_nrf5x_event_cb) will give semaphore when the flash operation is complete
+  // The callback also checks for NRF_EVT_FLASH_OPERATION_ERROR, which is not available to us otherwise
+  if (sd_enabled) {
+    xSemaphoreTake(_sem, portMAX_DELAY);
+    if (_flash_op_failed) 
+      return true; // Retry
+  }
+
+  // Success
+  return false;
 }
 
 // Flash Abstraction Layer
@@ -107,30 +142,28 @@ static bool fal_erase (uint32_t addr)
   // Init semaphore for first call
   if ( _sem == NULL )
   {
-    _sem = xSemaphoreCreateCounting(10, 0);
+    _sem = xSemaphoreCreateBinary();
     VERIFY(_sem);
   }
 
-  // retry if busy
-  uint32_t err;
-  while ( NRF_ERROR_BUSY == (err = sd_flash_page_erase(addr / FLASH_NRF52_PAGE_SIZE)) )
-  {
-    delay(1);
-  }
-  VERIFY_STATUS(err, false);
-
-  // wait for async event if SD is enabled
+  // Check if soft device is enabled
+  // If yes, flash operations are async, so we need to wait for the callback to give the semaphore
   uint8_t sd_en = 0;
   (void) sd_softdevice_is_enabled(&sd_en);
 
-  if ( sd_en ) xSemaphoreTake(_sem, portMAX_DELAY);
-
-  return true;
+  // Make multiple attempts to erase
+  uint8_t attempt = 0;
+  while (retry_flash_op(sd_flash_page_erase(addr / FLASH_NRF52_PAGE_SIZE), sd_en)) {
+    if (++attempt > MAX_RETRY)
+      return false; // Failure
+  }
+    return true; // Success
 }
 
 static uint32_t fal_program (uint32_t dst, void const * src, uint32_t len)
 {
-  // wait for async event if SD is enabled
+  // Check if soft device is enabled
+  // If yes, flash operations are async, so we need to wait for the callback to give the semaphore
   uint8_t sd_en = 0;
   (void) sd_softdevice_is_enabled(&sd_en);
 
@@ -140,27 +173,26 @@ static uint32_t fal_program (uint32_t dst, void const * src, uint32_t len)
   // https://devzone.nordicsemi.com/f/nordic-q-a/40088/sd_flash_write-cause-nrf_fault_id_sd_assert
   // Workaround: write half page at a time.
 #if NRF52832_XXAA
-  while ( NRF_ERROR_BUSY == (err = sd_flash_write((uint32_t*) dst, (uint32_t const *) src, len/4)) )
-  {
-    delay(1);
+  uint8_t attempt = 0;
+  while (retry_flash_op(sd_flash_write((uint32_t*) dst, (uint32_t const *) src, len/4), sd_en)) {
+    if (++attempt > MAX_RETRY)
+      return 0; // Failure
   }
-  VERIFY_STATUS(err, 0);
-
-  if ( sd_en ) xSemaphoreTake(_sem, portMAX_DELAY);
 #else
-  while ( NRF_ERROR_BUSY == (err = sd_flash_write((uint32_t*) dst, (uint32_t const *) src, len/8)) )
-  {
-    delay(1);
+
+  // First part of block
+  uint8_t attempt = 0;
+  while (retry_flash_op(sd_flash_write((uint32_t*) dst, (uint32_t const *) src, len/8), sd_en)) {
+    if (++attempt > MAX_RETRY)
+      return 0; // Failure
   }
-  VERIFY_STATUS(err, 0);
-  if ( sd_en ) xSemaphoreTake(_sem, portMAX_DELAY);
 
-  while ( NRF_ERROR_BUSY == (err = sd_flash_write((uint32_t*) (dst+ len/2), (uint32_t const *) (src + len/2), len/8)) )
-  {
-    delay(1);
+  // Second part of block
+  attempt = 0;
+  while (retry_flash_op(sd_flash_write((uint32_t*) (dst+ len/2), (uint32_t const *) (src + len/2), len/8), sd_en)) {
+    if (++attempt > MAX_RETRY)
+      return 0; // Failure
   }
-  VERIFY_STATUS(err, 0);
-  if ( sd_en ) xSemaphoreTake(_sem, portMAX_DELAY);
 #endif
 
   return len;