I think GBM can hit every point on that circle! -- well, possibly excepting one, which doesn't matter.
DavidJN's cite may be spot-on, although I haven't looked at that paper in years.
Off-hand, the horizontal line through A identifies two key times (t1,t2), and the GBM's location at t1 splits the problem into two exclusive possibilities. (Which are, hopefully, obvious).
Approximating the circle with purely horizontal pieces might be too crude. I vaguely recall parabolic segments (for BM) are solvable -- that might be useful.
Or maybe just a brute force lattice of some sort with a zillion nodes. Or a PDE. Many possibilities.
But, again, whatever the numerics: solve two separate problems, starting from t1, and conditioning on the location of the motion at t1. Once you have those solutions, just integrate them over the transition density of reaching those starting points, which is just an elementary lognormal.
Anyway, that's how I'd do it.