Not sure how I did this, I can't recreate it with toy programs. I was playing with recording gdb, and after doing "record stop" I did "continue" and got a SIGTRAP. It didn't appear to be a case of an int3 left in the instruction stream. It turned out to be due to eflags.TF being set.
I'd suspect your recording session stepped into a signal handler while recording, and then you tripped on the kernel bug behind Bug 17511. A kernel patch for that bug just made it to the mm tree this week...