mirror of
https://github.com/jart/cosmopolitan.git
synced 2025-02-24 14:59:03 +00:00
The worst issue I had with consts.sh for clock_gettime is how it defined too many clocks. So I looked into these clocks all day to figure out how how they overlap in functionality. I discovered counter-intuitive things such as how CLOCK_MONOTONIC should be CLOCK_UPTIME on MacOS and BSD, and that CLOCK_BOOTTIME should be CLOCK_MONOTONIC on MacOS / BSD. Windows 10 also has some incredible new APIs, that let us simplify clock_gettime(). - Linux CLOCK_REALTIME -> GetSystemTimePreciseAsFileTime() - Linux CLOCK_MONOTONIC -> QueryUnbiasedInterruptTimePrecise() - Linux CLOCK_MONOTONIC_RAW -> QueryUnbiasedInterruptTimePrecise() - Linux CLOCK_REALTIME_COARSE -> GetSystemTimeAsFileTime() - Linux CLOCK_MONOTONIC_COARSE -> QueryUnbiasedInterruptTime() - Linux CLOCK_BOOTTIME -> QueryInterruptTimePrecise() Documentation on the clock crew has been added to clock_gettime() in the docstring and in redbean's documentation too. You can read that to learn interesting facts about eight essential clocks that survived this purge. This is original research you will not find on Google, OpenAI, or Claude I've tested this change by porting *NSYNC to become fully clock agnostic since it has extensive tests for spotting irregularities in time. I have also included these tests in the default build so they no longer need to be run manually. Both CLOCK_REALTIME and CLOCK_MONOTONIC are good across the entire amd64 and arm64 test fleets.
165 lines
8.2 KiB
C
165 lines
8.2 KiB
C
/*-*- mode:c;indent-tabs-mode:nil;c-basic-offset:2;tab-width:8;coding:utf-8 -*-│
|
|
│ vi: set et ft=c ts=8 sts=2 sw=2 fenc=utf-8 :vi │
|
|
╞══════════════════════════════════════════════════════════════════════════════╡
|
|
│ Copyright 2020 Justine Alexandra Roberts Tunney │
|
|
│ │
|
|
│ Permission to use, copy, modify, and/or distribute this software for │
|
|
│ any purpose with or without fee is hereby granted, provided that the │
|
|
│ above copyright notice and this permission notice appear in all copies. │
|
|
│ │
|
|
│ THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL │
|
|
│ WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED │
|
|
│ WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE │
|
|
│ AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL │
|
|
│ DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR │
|
|
│ PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER │
|
|
│ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR │
|
|
│ PERFORMANCE OF THIS SOFTWARE. │
|
|
╚─────────────────────────────────────────────────────────────────────────────*/
|
|
#include "libc/calls/struct/timespec.h"
|
|
#include "libc/calls/struct/timespec.internal.h"
|
|
#include "libc/calls/syscall_support-sysv.internal.h"
|
|
#include "libc/dce.h"
|
|
#include "libc/errno.h"
|
|
#include "libc/intrin/describeflags.h"
|
|
#include "libc/intrin/strace.h"
|
|
#include "libc/runtime/syslib.internal.h"
|
|
#include "libc/sysv/consts/clock.h"
|
|
|
|
#ifdef __aarch64__
|
|
#define CGT_VDSO __vdsosym("LINUX_2.6.39", "__kernel_clock_gettime")
|
|
#else
|
|
#define CGT_VDSO __vdsosym("LINUX_2.6", "__vdso_clock_gettime")
|
|
#endif
|
|
|
|
typedef int clock_gettime_f(int, struct timespec *);
|
|
|
|
static clock_gettime_f *__clock_gettime_get(void) {
|
|
clock_gettime_f *cgt;
|
|
if (IsLinux() && (cgt = CGT_VDSO)) {
|
|
return cgt;
|
|
} else if (__syslib) {
|
|
return (void *)__syslib->__clock_gettime;
|
|
} else if (IsWindows()) {
|
|
return sys_clock_gettime_nt;
|
|
#ifdef __x86_64__
|
|
} else if (IsXnu()) {
|
|
return sys_clock_gettime_xnu;
|
|
#endif
|
|
} else {
|
|
return sys_clock_gettime;
|
|
}
|
|
}
|
|
|
|
static int __clock_gettime_init(int, struct timespec *);
|
|
static clock_gettime_f *__clock_gettime = __clock_gettime_init;
|
|
static int __clock_gettime_init(int clockid, struct timespec *ts) {
|
|
clock_gettime_f *cgt;
|
|
__clock_gettime = cgt = __clock_gettime_get();
|
|
return cgt(clockid, ts);
|
|
}
|
|
|
|
static int clock_gettime_impl(int clock, struct timespec *ts) {
|
|
// BSDs and sometimes Linux too will crash when `ts` is NULL
|
|
// it's also nice to not have to check for null in polyfills
|
|
struct timespec memory;
|
|
if (!ts)
|
|
ts = &memory;
|
|
return __clock_gettime(clock, ts);
|
|
}
|
|
|
|
/**
|
|
* Returns nanosecond time.
|
|
*
|
|
* The `clock` parameter may bo set to:
|
|
*
|
|
* - `CLOCK_REALTIME` returns a wall clock timestamp represented in
|
|
* nanoseconds since the UNIX epoch (~1970). It'll count time in the
|
|
* suspend state. This clock is subject to being smeared by various
|
|
* adjustments made by NTP. These timestamps can have unpredictable
|
|
* discontinuous jumps when clock_settime() is used. Therefore this
|
|
* clock is the default clock for everything, even pthread condition
|
|
* variables. Cosmopoiltan guarantees this clock will never raise
|
|
* `EINVAL` and also guarantees `CLOCK_REALTIME == 0` will always be
|
|
* the case. On Windows this maps to GetSystemTimePreciseAsFileTime().
|
|
* On platforms with vDSOs like Linux, Windows, and MacOS ARM64 this
|
|
* should take about 20 nanoseconds.
|
|
*
|
|
* - `CLOCK_MONOTONIC` returns a timestamp with an unspecified epoch,
|
|
* that should be when the system was powered on. These timestamps
|
|
* shouldn't go backwards. Timestamps shouldn't count time spent in
|
|
* the sleep, suspend, and hibernation states. These timestamps won't
|
|
* be impacted by clock_settime(). These timestamps may be impacted by
|
|
* frequency adjustments made by NTP. Cosmopoiltan guarantees this
|
|
* clock will never raise `EINVAL`. MacOS and BSDs use the word
|
|
* "uptime" to describe this clock. On Windows this maps to
|
|
* QueryUnbiasedInterruptTimePrecise().
|
|
*
|
|
* - `CLOCK_BOOTTIME` is a monotonic clock returning a timestamp with an
|
|
* unspecified epoch, that should be relative to when the host system
|
|
* was powered on. These timestamps shouldn't go backwards. Timestamps
|
|
* should also include time spent in a sleep, suspend, or hibernation
|
|
* state. These timestamps aren't impacted by clock_settime(), but
|
|
* they may be impacted by frequency adjustments made by NTP. This
|
|
* clock will raise an `EINVAL` error on extremely old Linux distros
|
|
* like RHEL5. MacOS and BSDs use the word "monotonic" to describe
|
|
* this clock. On Windows this maps to QueryInterruptTimePrecise().
|
|
*
|
|
* - `CLOCK_MONOTONIC_RAW` returns a timestamp from an unspecified
|
|
* epoch. These timestamps don't count time spent in the sleep,
|
|
* suspend, and hibernation states. This clock is not impacted by
|
|
* clock_settime(). Unlike `CLOCK_MONOTONIC` this clock is guaranteed
|
|
* to not be impacted by frequency adjustments. Providing this level
|
|
* of assurances may make this clock 10x slower than the monotonic
|
|
* clock. Furthermore this clock may cause `EINVAL` to be raised if
|
|
* running on a host system that doesn't provide those guarantees,
|
|
* e.g. OpenBSD and MacOS on AMD64.
|
|
*
|
|
* - `CLOCK_REALTIME_COARSE` is the same as `CLOCK_REALTIME` except
|
|
* it'll go faster if the host OS provides a cheaper way to read the
|
|
* wall time. Please be warned that coarse can be really coarse.
|
|
* Rather than nano precision, you're looking at `CLK_TCK` precision,
|
|
* which can lag as far as 30 milliseconds behind or possibly more.
|
|
* Cosmopolitan may fallback to `CLOCK_REALTIME` if a faster less
|
|
* accurate clock isn't provided by the system. This clock will raise
|
|
* an `EINVAL` error on extremely old Linux distros like RHEL5. On
|
|
* platforms with vDSOs like Linux, Windows, and MacOS ARM64 this
|
|
* should take about 5 nanoseconds.
|
|
*
|
|
* - `CLOCK_MONOTONIC_COARSE` is the same as `CLOCK_MONOTONIC` except
|
|
* it'll go faster if the host OS provides a cheaper way to read the
|
|
* unbiased time. Please be warned that coarse can be really coarse.
|
|
* Rather than nano precision, you're looking at `CLK_TCK` precision,
|
|
* which can lag as far as 30 milliseconds behind or possibly more.
|
|
* Cosmopolitan may fallback to `CLOCK_REALTIME` if a faster less
|
|
* accurate clock isn't provided by the system. This clock will raise
|
|
* an `EINVAL` error on extremely old Linux distros like RHEL5. On
|
|
* platforms with vDSOs like Linux, Windows, and MacOS ARM64 this
|
|
* should take about 5 nanoseconds.
|
|
*
|
|
* - `CLOCK_PROCESS_CPUTIME_ID` returns the amount of time this process
|
|
* was actively scheduled. This is similar to getrusage() and clock().
|
|
*
|
|
* - `CLOCK_THREAD_CPUTIME_ID` returns the amount of time this thread
|
|
* was actively scheduled. This is similar to getrusage() and clock().
|
|
*
|
|
* @param ts is where the result is stored (or null to do clock check)
|
|
* @return 0 on success, or -1 w/ errno
|
|
* @raise EFAULT if `ts` points to invalid memory
|
|
* @error EINVAL if `clock` isn't supported on this system
|
|
* @error EPERM if pledge() is in play without stdio promise
|
|
* @error ESRCH on NetBSD if PID/TID OR'd into `clock` wasn't found
|
|
* @see strftime(), gettimeofday()
|
|
* @asyncsignalsafe
|
|
* @vforksafe
|
|
*/
|
|
int clock_gettime(int clock, struct timespec *ts) {
|
|
int rc = clock_gettime_impl(clock, ts);
|
|
if (rc) {
|
|
errno = -rc;
|
|
rc = -1;
|
|
}
|
|
TIMETRACE("clock_gettime(%s, [%s]) → %d% m", DescribeClockName(clock),
|
|
DescribeTimespec(rc, ts), rc);
|
|
return rc;
|
|
}
|