[PATCH] [man-pages] adjtimex.2: document clock_adjtime

I was experimenting with some possible changes to adjtimex(2) and
clock_adjtime(2) and tried to look up the man page to see what the
documented behavior is when I noticed that clock_adjtime() appears
to be the only system call that is currently undocumented.

Before I do any changes to it, this tries to document what I
understand it currently does.

Signed-off-by: Arnd Bergmann <redacted>
---
This is my first patch to man-pages, I'm just guessing about
what should be in there and how to format it, please reword
or reformat as necessary.
---
 man2/adjtimex.2      | 25 +++++++++++++++++++++++--
 man2/clock_adjtime.2 |  1 +
 2 files changed, 24 insertions(+), 2 deletions(-)
 create mode 100644 man2/clock_adjtime.2

diff --git a/man2/adjtimex.2 b/man2/adjtimex.2
index fc6892d..b7c0a5f 100644
--- a/man2/adjtimex.2
+++ b/man2/adjtimex.2

@@ -35,6 +35,8 @@ adjtimex, ntp_adjtime \- tune kernel clock
 .PP
 .BI "int adjtimex(struct timex *" "buf" );
 .PP
+.BI "int clock_adjtimex(clockid_t " clk_id, " struct timex *" "buf" );
+.PP
 .BI "int ntp_adjtime(struct timex *" buf );
 .fi
 .SH DESCRIPTION

@@ -344,6 +346,12 @@ Attempts to set read-only
 .I status
 bits are silently ignored.
 .\"
+.SS clock_adjtime ()
+The
+.BR clock_adjtime ()
+system call (added in Linux 2.6.39) behaves like adjtimex() but takes an additional
+.IR clk_id
+argument to specify the particular clock on which to act.
 .SS ntp_adjtime ()
 The
 .BR ntp_adjtime ()

@@ -472,6 +480,11 @@ An attempt was made to set
 to a value other than those listed above.
 .TP
 .B EINVAL
+The
+.I clk_id
+specified is not supported on this system.
+.TP
+.B EINVAL
 An attempt was made to set
 .I buf.tick
 to a value outside the range

@@ -482,6 +495,10 @@ where
 .B HZ
 is the system timer interrupt frequency.
 .TP
+.B EOPNOTSUPP
+.I clk_id
+does not support adjustment
+.TP
 .B EPERM
 .I buf.modes
 is neither 0 nor

@@ -503,10 +520,12 @@ T{
 T}	Thread safety	MT-Safe
 .TE
 .SH CONFORMING TO
-Neither of these interfaces is described in POSIX.1
+None of these interfaces is described in POSIX.1
 .PP
 .BR adjtimex ()
-is Linux-specific and should not be used in programs
+and
+.BR clock_adjtimex ()
+are Linux-specific and should not be used in programs
 intended to be portable.
 .PP
 The preferred API for the NTP daemon is

@@ -533,6 +552,8 @@ is done by the kernel in timer context
 Thus, it will take one tick into the second
 for the leap second to be inserted or deleted.
 .SH SEE ALSO
+.BR clock_gettime (2)
+.BR clock_settime (2)
 .BR settimeofday (2),
 .BR adjtime (3),
 .BR ntp_gettime (3),

diff --git a/man2/clock_adjtime.2 b/man2/clock_adjtime.2
new file mode 100644
index 0000000..b08b9c8
--- /dev/null
+++ b/man2/clock_adjtime.2

@@ -0,0 +1 @@
+.so man2/adjtimex.2

-- 
2.9.0

On Mon, Nov 20, 2017 at 11:53:02PM +0100, Arnd Bergmann wrote:

 .B EINVAL
+The
+.I clk_id
+specified is not supported on this system.

We return EINVAL when the clockid is not valid.  That can mean two
things.  Either the SYS-V style hard coded positive clockid is out of
range, or the dynamic negative clockid does not refer to a valid
instance of a clock object.

Dynamic clocks might also return ENODEV in case a hot-plugable device
(like USB) has disappeared after its character device was opened.

Thanks,
Richard

On Tue, Nov 21, 2017 at 4:05 AM, Richard Cochran
[off-list ref] wrote:

On Mon, Nov 20, 2017 at 11:53:02PM +0100, Arnd Bergmann wrote:

quoted

 .B EINVAL
+The
+.I clk_id
+specified is not supported on this system.

We return EINVAL when the clockid is not valid.  That can mean two
things.  Either the SYS-V style hard coded positive clockid is out of
range, or the dynamic negative clockid does not refer to a valid
instance of a clock object.

Dynamic clocks might also return ENODEV in case a hot-plugable device
(like USB) has disappeared after its character device was opened.

I copied that line from clock_gettime() man page. I suppose we want to
fix change this in both pages, right? Any suggestions for a good way to
express your explanation in the man page? I suppose we don't want to
go into details of the implementation there but still capture the possible
corner cases.

       Arnd

On Tue, Nov 21, 2017 at 09:06:37AM +0100, Arnd Bergmann wrote:

I copied that line from clock_gettime() man page. I suppose we want to
fix change this in both pages, right? Any suggestions for a good way to
express your explanation in the man page? I suppose we don't want to
go into details of the implementation there but still capture the possible
corner cases.

Dynamic clockids are a Linux specific extension.  This should be
explained with a paragraph or two on the gettime man page, along with
an example using the macros.

	#define CLOCKFD 3
	#define FD_TO_CLOCKID(fd)	((~(clockid_t) (fd) << 3) | CLOCKFD)
	#define CLOCKID_TO_FD(clk)	((unsigned int) ~((clk) >> 3))

Then, the adjtimex page can say, see gettime.

Let me try to come up with a text over the (USA) holiday weekend.

Thanks,
Richard

On Tue, Nov 21, 2017 at 5:06 PM, Richard Cochran
[off-list ref] wrote:

On Tue, Nov 21, 2017 at 09:06:37AM +0100, Arnd Bergmann wrote:

quoted

I copied that line from clock_gettime() man page. I suppose we want to
fix change this in both pages, right? Any suggestions for a good way to
express your explanation in the man page? I suppose we don't want to
go into details of the implementation there but still capture the possible
corner cases.

Dynamic clockids are a Linux specific extension.  This should be
explained with a paragraph or two on the gettime man page, along with
an example using the macros.

        #define CLOCKFD 3
        #define FD_TO_CLOCKID(fd)       ((~(clockid_t) (fd) << 3) | CLOCKFD)
        #define CLOCKID_TO_FD(clk)      ((unsigned int) ~((clk) >> 3))

Then, the adjtimex page can say, see gettime.

Let me try to come up with a text over the (USA) holiday weekend.

Thanks! There is no rush here, take your time. One more question:
I see that ptp_clock_adjtime doesn't call timekeeping_validate_timex(),
so a number of the error conditions are not caught there. Should we
document that as intended, or change it to behave the same way
as do_adjtimex()?

I also see that ptp_clock_adjtime() ignores all other flags whenever
ADJ_SETOFFSET is set, while __do_adjtimex() can do ADJ_SETOFFSET
and ADJ_FREQUENCY (or any other combination) in a single syscall,
which matches what is documented in the man page.

       Arnd

Hi Richard,

On 11/21/2017 05:06 PM, Richard Cochran wrote:

On Tue, Nov 21, 2017 at 09:06:37AM +0100, Arnd Bergmann wrote:

quoted

I copied that line from clock_gettime() man page. I suppose we want to
fix change this in both pages, right? Any suggestions for a good way to
express your explanation in the man page? I suppose we don't want to
go into details of the implementation there but still capture the possible
corner cases.

Dynamic clockids are a Linux specific extension.  This should be
explained with a paragraph or two on the gettime man page, along with
an example using the macros.

	#define CLOCKFD 3
	#define FD_TO_CLOCKID(fd)	((~(clockid_t) (fd) << 3) | CLOCKFD)
	#define CLOCKID_TO_FD(clk)	((unsigned int) ~((clk) >> 3))

Then, the adjtimex page can say, see gettime.

Let me try to come up with a text over the (USA) holiday weekend.

Might you have a chance to look at this still?

Cheers,

Michael



-- 
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/

On Mon, Dec 18, 2017 at 08:19:42PM +0100, Michael Kerrisk (man-pages) wrote:

quoted

Let me try to come up with a text over the (USA) holiday weekend.

Might you have a chance to look at this still?

Sorry for forgetting this.  I will have another chance over the
upcoming international holiday...

Thanks,
Richard

Linux has allowed passing open file descriptors to clock_gettime() and
friends since v2.6.39.  This patch documents these "dynamic" clocks
and adds a brief example of how to use them.

Signed-off-by: Richard Cochran <redacted>
---
 man2/clock_getres.2 | 39 ++++++++++++++++++++++++++++++++++++++-
 1 file changed, 38 insertions(+), 1 deletion(-)

diff --git a/man2/clock_getres.2 b/man2/clock_getres.2
index 0812d159a..30cbfe46a 100644
--- a/man2/clock_getres.2
+++ b/man2/clock_getres.2

@@ -183,6 +183,35 @@ Per-process CPU-time clock
 .TP
 .BR CLOCK_THREAD_CPUTIME_ID " (since Linux 2.6.12)"
 Thread-specific CPU-time clock.
+.PP
+Linux also implements dynamic clock instances as described below.
+.SH DYNAMIC CLOCKS " (since Linux 2.6.39)"
+In addition to hard coded SYS-V style clock ids, Linux also supports
+POSIX clock operations on certain character devices.  Such devices are
+called "dynamic" clocks.  Using the appropriate macros, open file
+descriptors may be converted into clock ids and passed to
+.BR clock_gettime (),
+.BR clock_settime (),
+and
+.BR clock_adj (2).
+The follow example shows how to convert a file descriptor into a
+dynamic clock id.
+.PP
+.in +4n
+.EX
+#define CLOCKFD 3
+#define FD_TO_CLOCKID(fd)       ((~(clockid_t) (fd) << 3) | CLOCKFD)
+#define CLOCKID_TO_FD(clk)      ((unsigned int) ~((clk) >> 3))
+
+struct timeval tv;
+clockid_t clkid;
+int fd;
+
+fd = open("/dev/ptp0", O_RDWR);
+clkid = FD_TO_CLOCKID(fd);
+clock_gettime(clkid, &tv);
+.EE
+.in
 .SH RETURN VALUE
 .BR clock_gettime (),
 .BR clock_settime (),

@@ -200,11 +229,19 @@ points outside the accessible address space.
 .B EINVAL
 The
 .I clk_id
-specified is not supported on this system.
+specified is invalid for one of two reasons.  Either the SYS-V style
+hard coded positive value is out of range, or the dynamic clock id
+does not refer to a valid instance of a clock object.
 .\" Linux also gives this error on attempts to set CLOCK_PROCESS_CPUTIME_ID
 .\" and CLOCK_THREAD_CPUTIME_ID, when probably the proper error should be
 .\" EPERM.
 .TP
+.B ENODEV
+The hot-plugable device (like USB for example) represented by a
+dynamic
+.I clk_id
+has disappeared after its character device was opened.
+.TP
 .B EPERM
 .BR clock_settime ()
 does not have permission to set the clock indicated.

-- 
2.11.0

From: Arnd Bergmann <arnd@arndb.de>

I was experimenting with some possible changes to adjtimex(2) and
clock_adjtime(2) and tried to look up the man page to see what the
documented behavior is when I noticed that clock_adjtime() appears
to be the only system call that is currently undocumented.

Before I do any changes to it, this tries to document what I
understand it currently does.

[ RC: Add better explanations of the usage and error codes
  and correct some typographical mistakes. ]

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Richard Cochran <richardcochran@gmail.com>
---
 man2/adjtimex.2      | 63 +++++++++++++++++++++++++++++++++++++++++++++++++---
 man2/clock_adjtime.2 |  1 +
 2 files changed, 61 insertions(+), 3 deletions(-)
 create mode 100644 man2/clock_adjtime.2

diff --git a/man2/adjtimex.2 b/man2/adjtimex.2
index fc6892d7e..71b5c4a5a 100644
--- a/man2/adjtimex.2
+++ b/man2/adjtimex.2

@@ -35,6 +35,8 @@ adjtimex, ntp_adjtime \- tune kernel clock
 .PP
 .BI "int adjtimex(struct timex *" "buf" );
 .PP
+.BI "int clock_adjtime(clockid_t " clk_id, " struct timex *" "buf" );
+.PP
 .BI "int ntp_adjtime(struct timex *" buf );
 .fi
 .SH DESCRIPTION

@@ -158,8 +160,26 @@ includes the
 .B ADJ_NANO
 flag, then
 .I buf.time.tv_usec
-is interpreted as a nanosecond value;
+is interpreted as a nanosecond value,
 otherwise it is interpreted as microseconds.
+.IP
+The value of
+.I buf.time
+is the sum of its two fields, but the
+field
+.I buf.time.tv_usec
+must always be non-negative.  The following example shows how to
+normalize a timeval with nanosecond resolution.
+.PP
+.in +12n
+.EX
+while (buf.time.tv_usec < 0) {
+    buf.time.tv_sec  -= 1;
+    buf.time.tv_usec += 1000000000;
+}
+.EE
+.in
+.PP
 .TP
 .BR ADJ_MICRO " (since Linux 2.6.26)"
 .\" commit eea83d896e318bda54be2d2770d2c5d6668d11db

@@ -344,6 +364,12 @@ Attempts to set read-only
 .I status
 bits are silently ignored.
 .\"
+.SS clock_adjtime ()
+The
+.BR clock_adjtime ()
+system call (added in Linux 2.6.39) behaves like adjtimex() but takes an additional
+.IR clk_id
+argument to specify the particular clock on which to act.
 .SS ntp_adjtime ()
 The
 .BR ntp_adjtime ()

@@ -472,6 +498,19 @@ An attempt was made to set
 to a value other than those listed above.
 .TP
 .B EINVAL
+The
+.I clk_id
+given to
+.BR clock_adjtime ()
+is invalid for one of two reasons.  Either the SYS-V style hard coded
+positive value is out of range, or the dynamic
+.I clk_id
+does not refer to a valid instance of a clock object.
+See
+.BR clock_gettime (2)
+for a discussion of dynamic clocks.
+.TP
+.B EINVAL
 An attempt was made to set
 .I buf.tick
 to a value outside the range

@@ -482,6 +521,20 @@ where
 .B HZ
 is the system timer interrupt frequency.
 .TP
+.B ENODEV
+The hot-plugable device (like USB for example) represented by a
+dynamic
+.I clk_id
+has disappeared after its character device was opened.
+See
+.BR clock_gettime (2)
+for a discussion of dynamic clocks.
+.TP
+.B EOPNOTSUPP
+The given
+.I clk_id
+does not support adjustment.
+.TP
 .B EPERM
 .I buf.modes
 is neither 0 nor

@@ -503,10 +556,12 @@ T{
 T}	Thread safety	MT-Safe
 .TE
 .SH CONFORMING TO
-Neither of these interfaces is described in POSIX.1
+None of these interfaces is described in POSIX.1
 .PP
 .BR adjtimex ()
-is Linux-specific and should not be used in programs
+and
+.BR clock_adjtime ()
+are Linux-specific and should not be used in programs
 intended to be portable.
 .PP
 The preferred API for the NTP daemon is

@@ -533,6 +588,8 @@ is done by the kernel in timer context
 Thus, it will take one tick into the second
 for the leap second to be inserted or deleted.
 .SH SEE ALSO
+.BR clock_gettime (2)
+.BR clock_settime (2)
 .BR settimeofday (2),
 .BR adjtime (3),
 .BR ntp_gettime (3),

diff --git a/man2/clock_adjtime.2 b/man2/clock_adjtime.2
new file mode 100644
index 000000000..b08b9c801
--- /dev/null
+++ b/man2/clock_adjtime.2

@@ -0,0 +1 @@
+.so man2/adjtimex.2

-- 
2.11.0

Hello Richard,

On 1/1/18 7:28 AM, Richard Cochran wrote:

Linux has allowed passing open file descriptors to clock_gettime() and
friends since v2.6.39.  This patch documents these "dynamic" clocks
and adds a brief example of how to use them.

This fell on the floor, I'm sorry. It's applied now.

Thanks you!

Michael

-- 
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/

Hello Richard, Arnd,

On 1/1/18 7:28 AM, Richard Cochran wrote:

From: Arnd Bergmann <arnd@arndb.de>

I was experimenting with some possible changes to adjtimex(2) and
clock_adjtime(2) and tried to look up the man page to see what the
documented behavior is when I noticed that clock_adjtime() appears
to be the only system call that is currently undocumented.

Before I do any changes to it, this tries to document what I
understand it currently does.

[ RC: Add better explanations of the usage and error codes
  and correct some typographical mistakes. ]

And this patch too is now applied.

Thank you!

Cheers,

Michael

-- 
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
Linux/UNIX System Programming Training: http://man7.org/training/