The error message is from taking the inner product of two CVX variables, which is a non-convex objective, not a linear objective. If you can;t overcome this, then CVX is not an appropriate tool for your model.
The constraint norm(Px,2) <= 1 would be o.k., but norm(Px,2) == 1 is non-convex and would be rejected by CVX.
I don;t see an l1 norm of x being maximized, unless you’re trying to do some dual norm thing, in which case I don’t think you have it right. Maximizing an l1, or any other, norm is non-convex, and therefore not possible in CVX unless you can accomplish it using binary or integer variables to handle the non-convexity.