#heteroscedasticity

In the regression model $y = wx + \epsilon$ with $\epsilon \sim \mathcal{N}(0, \sigma^2 x^2)$, the noise grows with the input ($x \in [1, 4]$, $\sigma$ known). Given the training data $(x, y) = (1, 3), (2, 5), (4, 10)$, what is the maximum-likelihood estimate of $w$?