A question about the loss

Why log loss[-sum(log f  (yis; xsi ))] is used in multi classification problem instead of cross entropy？
For example, I have three categories. When the prediction y of the network is correct, such as equal to (1,0,0) or (0,1,0), the loss is infinite.