MSP430I2040: how to write large chunk of data into flash?

Hi,

I am trying to save a large chunk of data, a struct for some user generated settings, into flash to persistent them. The size if 112B. Somehow sometimes only the first 3x ~ 10x bytes are actually saved when read back the entire flash, the remaining bytes will be 0xFF as a result of the entire section of flash being erased before writing.

uint8_t tlv[64];
    uint8_t *Flash_ptr;
    // Read TLV calibration data before erasing information memory
	memcpy((void *)&tlv[0], (void*)TLV_START, sizeof(tlv));
    // Save settings struct into flash or other persistent medium
    Flash_ptr = (uint8_t *)SETTINGS_ADDR;
    // Erase information memory
    WDTCTL = WDTPW | WDTHOLD;
    FCTL2 = FWKEY | FSSEL_1 | FN1 | FN3 | FN5;  // MCLK/42 for Flash Timing Generator
    while (FCTL3 & BUSY);                       // Make sure flash controller is not busy
    FCTL3 = FWKEY;                              // Clear Lock bit
    if(FCTL3 & LOCKSEG) {                       // If Info Seg is still locked
        FCTL3 = FWKEY | LOCKSEG;                // Clear LOCKSEG bit
    }
    FCTL1 = FWKEY | ERASE;                      // Set Erase bit
    *Flash_ptr = 0;                             // Dummy write to erase info segment
    while (FCTL3 & BUSY);                       // Make sure flash controller is not busy
    FCTL3 = FWKEY;
    if(FCTL3 & LOCKSEG) {                       // If Info Seg is still locked
        FCTL3 = FWKEY | LOCKSEG;                // Clear LOCKSEG bit
    }
    FCTL1 = FWKEY | WRT;
    memcpy((void *)SETTINGS_ADDR, &settings, sizeof(Settings));
    // Save TLV calibration data back into flash
    memcpy((void *)TLV_START, &tlv[0], sizeof(tlv));
    FCTL1 = FWKEY;                              // Clear WRT bit
    FCTL3 = FWKEY | LOCKSEG;                    // Set LOCK bit

At first I thought this only happens to information flash, but later I found it also happens when SETTINGS_ADDR is pointed to main flash such as 0xF800. Datasheet mentions total accumulative programming time shouldn't exceed 8 ms. But my calculation indicates 112B takes less than 3 ms. with MCLK/42 as clock.

Could someone advise what I am missing here?

Thanks.