MSP430 F5438

While the answer to your question has been already given (since MCLK cannot be higher than 24 MHz and toggling port pins requires more than 1 MCLK cycle each where you'd need 1/2 cycle to get 24MHz output, so directly using the hardware clock outputs is the only way), your code won't work as expected:

you clear the OFIFG flags and then test for them. This won't work as expected. After clearing the fault flags, they will stay clear for some time. The fault flags are triggered by sort of retriggerable monoflops. This means a digital gate that remains high for a certain tiem after it has been triggered the last time and then falls to low. The time constant after which the monoflop falls to low and set the fault flag, must be chosen to fit the lowest input frequency that shall be still allowed.

With 24MHz I guess you're using the 5438A or similar. On the 5438A, the time constant for XT1 in high-frequency mode is between 3.3 and 33 microseconds. So after clearing the fault flag(s), between 3.3 and 33 microseconds (actually even a bit more if the crystal starts oscillating but then fades away) may pass before the fault flag is set again. You'll need to wait this time before checking the fault flag and proceeding. Without waiting, your code would always continue even if there's no crystal at all.

You're writing in assembly language, so it doesn't apply to you, but I'll write it here anyway:
Don't use a simple for-loop as delay. It won't work. Optimizing compilers will unroll or even eliminate the loop if it doesn't do anything useful. (executing a no_operation() IS something useful for the compiler). If the loop variable isn't declared volatile (and not a local one), the loop might be partially or completely unrolled anyway. You can setup a timer with the DCOs default frequency to get an at least somewhat reliable delay.