cudaGetErrorString what to do with return string

Hello all,

cudaGetErrorString does not require a pointer to allocated memory for the return string to be passed in as a parameter. That means it either passes back a pointer to memory that has been malloc’ed, or to memory that was declared a static, or to a string literal. Since the return value is const, I’m tempted to guess string literal, but I’d like to know for sure, and I’d like it very much if the CUDA Reference Manual would tell me.