clock_gettime: Add Linux aliases for CLOCK_*

[imp]

Summary

Linux standardized what we call CLOCK_{REALTIME,MONOTONIC}_FAST as
CLOCK_{REALTIME,MONOTONIC}_COARSE. In addition, Linux spells
CLOCK_UPTIME as CLOCK_BOOTTIME.

Add aliases to time.h and document these new aliases in
clock_gettime(2).

Sponsored by: Netflix

Test Plan

Cast a wide net for this review since rwatson introduced them, but others have touched adjacent code
Linux does not have a CLOCK_UPTIME_COARSE, so I didn't create an alias for CLOCK_UPTIME_FAST.
However, it does have CLOCK_BOOTTIME instead of CLOCK_UPTIME, so I did that too.

[kevans]

lib/libc/sys/clock_gettime.2

113

s/wit/with/

[imp]

lib/libc/sys/clock_gettime.2

113

Changed locally. Will update if there's other changes.

[kib]

We are already somewhat non-compliant to POSIX by providing more CLOCK_ symbols than defined by the standard. More, POSIX does not allow to pollute CLOCK_ from sys/time.h, but for time.h all symbols with CLOCK_ prefix are reserved.

I think that cleanup is due, but this change is fine until it.

[imp]

In D30988#697219, @kib wrote:

We are already somewhat non-compliant to POSIX by providing more CLOCK_ symbols than defined by the standard. More, POSIX does not allow to pollute CLOCK_ from sys/time.h, but for time.h all symbols with CLOCK_ prefix are reserved.

What's the proper symbol to hide these behind? Is it like errno.h (#ifndef _POSIX_SOURCE)? Or something else?

[phk]

no comments

[imp]

(tinderboxed)

[bz]

Linux also uses CLOCK_BOOTTIME (see D30985). Might it be possible to also add this along with this?

[imp]

In D30988#697785, @bz wrote:

Linux also uses CLOCK_BOOTTIME (see D30985). Might it be possible to also add this along with this?

I don't see why not, modulo kib@'s concerns about namespace pollution.

git 11:11am rebo:[15537]> git diff
diff --git a/sys/sys/time.h b/sys/sys/time.h
index 04b57cfea125..9fbae544cb27 100644
--- a/sys/sys/time.h
+++ b/sys/sys/time.h
@@ -482,6 +482,7 @@ struct clockinfo {
 #ifndef CLOCK_MONOTONIC
 #define        CLOCK_MONOTONIC 4
 #define        CLOCK_UPTIME    5               /* FreeBSD-specific. */
+#define        CLOCK_BOOTTIME          CLOCK_UPTIME            /* Linux compat */
 #define        CLOCK_UPTIME_PRECISE    7       /* FreeBSD-specific. */
 #define        CLOCK_UPTIME_FAST       8       /* FreeBSD-specific. */
 #define        CLOCK_REALTIME_PRECISE  9       /* FreeBSD-specific. */

is what you have in mind, eh?

[imp]

my only worry about adding it is that I requested an exp-run with this change already, so I'd have to request another one I think... and I'm out of time this morning to take care of it before the BBQ I'm headed to...

[imp]

In D30988#697788, @imp wrote:

my only worry about adding it is that I requested an exp-run with this change already, so I'd have to request another one I think... and I'm out of time this morning to take care of it before the BBQ I'm headed to...

ah, I did a half-assed version that's sufficient for exp-run testing and updated the bug:

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=256941

I'm not familiar with the process here, but if it hasn't started yet, then we'll be good.

[bz]

In D30988#697789, @imp wrote:

In D30988#697788, @imp wrote:

my only worry about adding it is that I requested an exp-run with this change already, so I'd have to request another one I think... and I'm out of time this morning to take care of it before the BBQ I'm headed to...

ah, I did a half-assed version that's sufficient for exp-run testing and updated the bug:

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=256941

I'm not familiar with the process here, but if it hasn't started yet, then we'll be good.

.. for as long as my original #define was correct?

[imp]

CLOCK_BOOTTIME adjustments

[imp]

In D30988#697795, @bz wrote:

.. for as long as my original #define was correct?

I think that it was. Please verify I got it correct though here.

[imp]

Update based on D31056

[kib]

sys/sys/_clock_id.h

69 ↗

(On Diff #91832)

I suggest to add 'Linux compat' herald comment right before '__BSD_VISIBLE' test for the whole block, and not repeat it for each line.

Might be, makes sense to do the same for 'FreeBSD-specific' comments above.

[imp]

update with kib's suggested comment

[imp]

rebase