I believe simply `(x-m)/qnorm(p)` will do it, where `m` is the known mean and `x` is the known percentile `p`. `qnorm(p)` gives you how many standard deviations away from the mean the value is, and then you scale it by the given difference.