LinuxQuestions.org - clock_gettime(CLOCK_THREAD_CPUTIME

My question on behavior of calling clock_gettime(CLOCK_THREAD_CPUTIME_ID). I looked at linux source code. There is a Posix implementation that coverts clock_id (CLOCK_THREAD_CPUTIME_ID) a specific kclock struct. Then does the per thread CPU tracking. But it seems like posix implementation will result in syscall. The code snippet is as follow https://elixir.bootlin.com/linux/lat...imers.c#L1085:

Code:

SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock,

        struct __kernel_timespec __user *, tp)

{

    const struct k_clock *kc = clockid_to_kclock(which_clock);

    struct timespec64 kernel_tp;

    int error;



    if (!kc)

        return -EINVAL;



    error = kc->clock_get_timespec(which_clock, &kernel_tp);



    if (!error && put_timespec64(&kernel_tp, tp))

        error = -EFAULT;



    return error;

}

The second path is vdso which doesn't incur syscall. However, it doesn't seem to support CLOCK_THREAD_CPUTIME_ID. My linux is built for x86. which path will it take and how to rationalize that in the source code? https://elixir.bootlin.com/linux/lat...eofday.c#L228:

Code:

static __always_inline int

__cvdso_clock_gettime_common(const struct vdso_data *vd, clockid_t clock,

                struct __kernel_timespec *ts)

{

    u32 msk;



    /* Check for negative values or invalid clocks */

    if (unlikely((u32) clock >= MAX_CLOCKS))

        return -1;



    /*

    * Convert the clockid to a bitmask and use it to check which

    * clocks are handled in the VDSO directly.

    */

    msk = 1U << clock;

    if (likely(msk & VDSO_HRES))

        vd = &vd[CS_HRES_COARSE];

    else if (msk & VDSO_COARSE)

        return do_coarse(&vd[CS_HRES_COARSE], clock, ts);

    else if (msk & VDSO_RAW)

        vd = &vd[CS_RAW];

    else

        return -1;



    return do_hres(vd, clock, ts);

}