What is the time unit that strace uses when displaying time spent in syscalls?

Question

When using the command strace with the flag -T, I would like to know what is the time unit used to display time spent in syscalls? I assume it should be in seconds, but I am not quite sure and it seems to be omitted from the manual.

Solution 1

From the source code:

if (Tflag) {
    ts_sub(ts, ts, &tcp->etime);
    tprintf(" <%ld.%06ld>",
        (long) ts->tv_sec, (long) ts->tv_nsec / 1000);
}

This means that the time is shown in seconds, with microseconds (calculated from the nanosecond value) after the decimal point.

Solution 2

If you run

strace -T  sleep 2

you will see

nanosleep({tv_sec=2, tv_nsec=0}, NULL)  = 0 <2.000230>

so it looks like the time spent is in seconds.

Solution 3

If you run the command strace using the "flag -c" it will show you a table and the time is reported in seconds:

strace -c -p 3569 # 3569 is PID
strace: Process 3569 attached
^Cstrace: Process 3569 detached
% time     seconds  usecs/call     calls    errors syscall
------ ----------- ----------- --------- --------- ----------------
99.73    0.016000           8      1971           poll
0.16    0.000025           0       509        75 futex
0.06    0.000010           0      1985      1966 recvmsg
0.06    0.000009           0      2336           mprotect
0.00    0.000000           0       478           read
0.00    0.000000           0        13           write
0.00    0.000000           0        29           mmap
0.00    0.000000           0         9           munmap
0.00    0.000000           0        18           writev
0.00    0.000000           0       351           madvise
0.00    0.000000           0         1           restart_syscall
------ ----------- ----------- --------- --------- ----------------
100.00    0.016044                  7700      2041 total

from strace's man

-c

Count time, calls, and errors for each system call and report a summary on program exit. On Linux, this attempts to show system time (CPU time spent running in the kernel) independent of wall clock time. If -c is used with -f or -F (below), only aggregate totals for all traced processes are kept.