We don’t have in ROOT such a function to compute the error on the full covariance matrix. You can compute the error only on the standard deviations (diagonal element of your covariance matrix).

I would suggest you to use a bootstrapping method to estimate these uncertainty, it would be probably anyway more reliable, since there are no prior assumption on the data distribution. You would need to generate many different histograms according to what you have in your original one and see the spread you get in the obtained covariance matrix.