Specifically, it seems as though do_slow_gettimeoffset() returns
the wrong value during the period between the timer interrupt
and when update_wall_timer_one_tick() updates xtime.tv_usec
in kernel/sched.c. Obviously, since sys_gettimeofday() does not
call do_gettimeofday() during this period, the system call is
not affected. However, several drivers call do_gettimeofday(),
possibly some in interrupts (as mine does).
Is it better to simply say "don't call do_gettimeofday() from
interrupts?"
dave...