mirrors/cosmopolitan
1
0
Fork 0
mirror of https://github.com/jart/cosmopolitan.git synced 2025-10-09 22:28:24 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

Delve into clock rabbit hole

Browse source 
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.
This commit is contained in:
Justine Tunney 2024-09-04 00:38:44 -07:00
parent 8f8145105c
commit dd8544c3bd
No known key found for this signature in database
GPG key ID: BE714B4575D6E328
87 changed files with 939 additions and 900 deletions
Download patch file Download diff file Expand all files Collapse all files

51
libc/calls/clock_gettime-mono.c
View file

@ -18,42 +18,63 @@
╚─────────────────────────────────────────────────────────────────────────────*/
#include "libc/atomic.h"
#include "libc/calls/struct/timespec.h"
#include "libc/calls/struct/timespec.internal.h"
#include "libc/calls/struct/timeval.h"
#include "libc/cosmo.h"
#include "libc/errno.h"
#include "libc/dce.h"
#include "libc/nexgen32e/rdtsc.h"
#include "libc/nexgen32e/x86feature.h"
/**
* @fileoverview Fast Monotonic Clock Polyfill for XNU/NT.
* @fileoverview Monotonic clock polyfill.
*
* This isn't quite `CLOCK_MONOTONIC` and isn't quite `CLOCK_BOOTTIME`
* either; however it is fast and almost always goes in one direction.
*
* Intel architecture guarantees that a mapping exists between rdtsc &
* nanoseconds only if the cpu advertises invariant timestamps support
* however this shouldn't matter for a monotonic clock since we really
* don't want to have it tick while suspended. Sadly that shall happen
* since nearly all x86 microprocessors support invariant tsc which is
* why we try to avoid this fallback when possible.
*/
int sys_sysctl(int *, unsigned, void *, size_t *, void *, size_t) libcesque;
static struct {
atomic_uint once;
struct timespec base_wall;
uint64_t base_tick;
unsigned long base;
struct timespec boot;
} g_mono;
static struct timespec get_boot_time_xnu(void) {
struct timeval t;
size_t n = sizeof(t);
int mib[] = {1 /* CTL_KERN */, 21 /* KERN_BOOTTIME */};
if (sys_sysctl(mib, 2, &t, &n, 0, 0) == -1)
__builtin_trap();
return timeval_totimespec(t);
}
static void sys_clock_gettime_mono_init(void) {
g_mono.base_wall = timespec_real();
g_mono.base_tick = rdtsc();
g_mono.base = rdtsc();
if (IsXnu()) {
g_mono.boot = get_boot_time_xnu();
} else {
__builtin_trap();
}
}
int sys_clock_gettime_mono(struct timespec *time) {
uint64_t nanos;
uint64_t cycles;
#ifdef __x86_64__
// intel architecture guarantees that a mapping exists between rdtsc &
// nanoseconds only if the cpu advertises invariant timestamps support
if (!X86_HAVE(INVTSC))
return -EINVAL;
#endif
cosmo_once(&g_mono.once, sys_clock_gettime_mono_init);
cycles = rdtsc() - g_mono.base_tick;
// ensure we get the full 64 bits of counting, which avoids wraparound
cycles = rdtsc() - g_mono.base;
// this is a crude approximation, that's worked reasonably well so far
// only the kernel knows the actual mapping between rdtsc and nanosecs
// which we could attempt to measure ourselves using clock_gettime but
// we'd need to impose 100 ms of startup latency for a guess this good
nanos = cycles / 3;
*time = timespec_add(g_mono.base_wall, timespec_fromnanos(nanos));
*time = timespec_add(g_mono.boot, timespec_fromnanos(nanos));
return 0;
}