Numerical problems such as finding eigenvalues, singular value decomposition, linear programming, are traditionally solved with algorithms that can be interpreted as discrete-time processes. One can also find in the literature continuous-time methods for these same problems where solutions are the equilibrium points to which converge stable differential equations. The paper exposes one such continuous-time method for solving linear matrix inequalities. The proposed differential equations are those of an adaptive control feedback loop on an LTI system. The adaptive law is passivity-based with additional structural constraints of two types. The first constraint imposes the gain to be block-diagonal at all times. It can be interpreted as a decentralized control structure. The second constraint is only required asymptotically. It for example reads as requiring the feedback gain to be symmetric when time goes to infinity. Point-wise global stability is proved with quadratic Lyapunov functions. Results are illustrated on LMIs related to an H1 norm computation problem. Solutions to the LMIs are obtained by simulations in Simulink.