Chapter 5 SVM why should center before LinearSVC

[gitgithan]

There was a tip

The LinearSVC class regularizes the bias term, so you should center the training set first by subtracting its mean

I couldn't find any resource discussing why this is so?
What happens if not centered?
What are the implications of (not) regularizing the bias term?