There are three ways how you could proceed:
If both arrays are already sorted for example in ascending order, then you start at the beginning of the first array and the end of the last array; if the sum is too small you skip to the next element of the first array, if the sum is too large you skip to the previous element of the second array, and so on.
If one array only is sorted, then you try each element x of the other array, lookup the element closest to M - x using binary search, and use the best pair.
If neither array is sorted, you sort either one or both. Calculate the runtime of either approach; it seems best to sort the smaller array, which then takes O (n log m) where m is the size of the larger array, and m the size of the smaller array.
If the larger array is already sorted, sorting the smaller array as well may be faster, because the sorting is done in O (m log m), and the first method takes only linear time. And if the arrays are of comparable size (say $m = n^{1/2}$) so log m vs. log n doesn't affect asymptotic behaviour, then what is fastest on some implementation will depend on the actual implementation.