> On a UP box it still need to switch to the tracing context once (or
> more) per system call. Ouch.
Yes, it would. And you get a 100% user-space solution that works
on 2.0 and 2.2.
As I wrote in my first message, and you didn't bother addressing, you
could add one ptrace call to determine whether the child is about to
core dump on the signal it just received. Then you get a 98% user-space
solution with one simple policy-free change to the kernel, and that
solution runs child system calls at full speed.
Or you could add another flavor of PTRACE_CONT that runs full speed
through system calls and handled signals, but traps out on unhandled
signals. Then you get a 98% user-space solution that runs every child
process at full speed.
And look at this, the user-space program can easily try the ptrace
enhancements and fall back if they are not present, so you get a 'ucore'
that works on every kernel version and runs at full speed when it finds
kernel support.
Michael Elizabeth Chastain
<mailto:mec@shout.net>
"love without fear"
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/