I happened to be comparing the in-process system call tracer against strace, and noticed that the latter is visibly faster. Since ptrace is not known for being quick, something is surprisingly slow in our trap/emulate path. It's likely that a trampoline approach would solve this, but stlil it seems worth understanding where the current costs are coming from.
I happened to be comparing the in-process system call tracer against
strace, and noticed that the latter is visibly faster. Sinceptraceis not known for being quick, something is surprisingly slow in our trap/emulate path. It's likely that a trampoline approach would solve this, but stlil it seems worth understanding where the current costs are coming from.