The pole masses of the heavy quarks (c, b and t) are relatively well defined in QCD (i.e. the solution of m²(p²) = p² extrapolated using the beta function and the available data from other values of µ usually obtained based upon model dependent decompositions of hadron masses that include these heavy quarks). Generally speaking, when we talk about the masses of the light quarks (u, d, and s) we use the masses at µ corresponding to 1 GeV or 2 GeV in an MS renormalization scheme (or some similar alternative) and those are the masses referenced by the particle data group. Naively extrapolating light quark masses to lower energy scales gives much larger hypothetical pole masses for these particles (approaching constituent quark approximation masses), but it isn't obvious that these pole masses are meaningful because confinement implies that light quarks are always present in hadrons, and there are no hadrons lighter than the pion (ca. 140 MeV) and the protons (a bit under 1 GeV). So, perhaps pole masses for these particles are simply "non-physical". The discussion of the issue in one paper (http://arxiv.org/pdf/hep-ph/9712201v2.pdf) states: Is there a non-perturbative way to determine the light quark pole masses below µ ∼ 1 GeV? Is it simply too hard to calculate but well defined? Or, are these quantities ill defined or truly non-physical?