Signed-off-by: Adam Wujek <dev_public@wujek.eu>

To be able to return number of nseconds

Signed-off-by: Adam Wujek <dev_public@wujek.eu>

Signed-off-by: Adam Wujek <dev_public@wujek.eu>

Before there was overflow in number of us.

Signed-off-by: Adam Wujek <dev_public@wujek.eu>

Signed-off-by: Adam Wujek <dev_public@wujek.eu>

Signed-off-by: Adam Wujek <dev_public@wujek.eu>

Signed-off-by: Adam Wujek <dev_public@wujek.eu>

As RTS_MODE_* values are exported via IPC not SPLL_MODE_*

Signed-off-by: Adam Wujek <dev_public@wujek.eu>

Signed-off-by: Adam Wujek <dev_public@wujek.eu>

Signed-off-by: Adam Wujek <dev_public@wujek.eu>

This fix solves the problem of jumps of a fraction of a second during
the startup. It is believed that the problem is in HDL, this commit is
just a workaround implemented in software.

Signed-off-by: Adam Wujek <dev_public@wujek.eu>

Jean-Claude BAU authored 5 years ago and Maciej Lipinski committed 5 years ago

- Add a new tool (wrs_leapsec) to check for an incoming leap second. It
is done by parsing the leap second file. If a leap second is detected
then adjustment parameters for the clock algoritm are set in the kernel.
- Start the tool at startup
- With cron, execute the tool every 3 hours
- As many tools need to parse the leap seconds file, functions have been
added in the library and then called by these tools

Jean-Claude BAU authored 5 years ago and Maciej Lipinski committed 5 years ago

The -n option in the wr_date tool is used for a dry run. It execute the
code without applying changes in the system and the hardware.
However calling "wr_date -n set host" was not setting the date but
applying the TAI offset. To make things more consistant, a new option
has been added. Calling "wr_date set host tai" now sets only the TAI
offset.
The wr_date script has been changed to take into account this
modification of the behavior.

Jean-Claude BAU authored 5 years ago and Adam Wujek committed 5 years ago

1/ wr_date script:
- Started before PPSi
- Set number of NTP retries to 10 if GM otherwise 1. GM info is
extracted from dot-config
- Clock source is controlled directly with wr_date tool
- Can be started any time without any risk to set a wrong date (see
point 2)
2/ wr_date
- 'set host' parameters take into account the time mode ( FR: set
sec+ns, BC: time cannot be set, GM: set seconds only+wait to be in a
middle of a second)
- function substrat_substrat() moved to time_lib.h. Also used by wr_mon
3/ wr_mon
- Print TAI-UTC
- Update Timing Mode & PLL locking state displayed values

Benoit Rat authored 5 years ago and Adam Wujek committed 5 years ago

By doing this we force to read again the FPGA register inside the loop
     and the compiler can not skip this loop thinking he is reading the
same value (that was not modified by him).

Conflicts:
	userspace/tools/wr_date.c

Benoit Rat authored 5 years ago and Adam Wujek committed 5 years ago

Fix TAI offset into kernel was done withing get function but this has
been removed in commit #091f4c8d. As it is still needed by slave, we
force its use using -n option. This will only fix tai offset into kernel
without applying ToD to WR/FPGA time.

Conflicts:
	userspace/tools/wr_date.c

Benoit Rat authored 5 years ago and Adam Wujek committed 5 years ago

Use gmtime instead of localtime (TAI should never used localtime)
Add verbose option to also display linux time

Benoit Rat authored 5 years ago and Adam Wujek committed 5 years ago

The output will be something like:

wr_date -v diff
TAI(HW)-UTC(SW): +36.602286
UTC(HW)-UTC(SW): -0.397714

Conflicts:
	userspace/tools/wr_date.c

Benoit Rat authored 5 years ago and Adam Wujek committed 5 years ago

In the previous version `wr_date get` would override kernel TAI time
(using local file) and this could cause some problems in case leap
second from /etc/leap-seconds.list was not the same than from PTP/WR
protocol.

baujc authored 6 years ago and Adam Wujek committed 5 years ago

- Add new option "stat" to display statistic about the evolution between
WR and Unix time
- When we set the WR time with the current UNIX time, the driver
wr_clocksource must be removed first and then reinstalled when the time
is set. If it is not done, the WR and Unix time are not well
synchronized.
- When the WR time is set, the sec and ns and set in one call. We don't
need to call twice the wr_date tool.

Keep created wr_date.conf in /etc not in /wr/etc.
There is no need to keep it in the flash instead of ramdisk.

Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

Now wr_date depends on libwr.

Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

Also, run "wr_date get" at boot once, so it parses leap-seconds.list
and configures the kernel for the proper tai_offset.

With the previous commit it worked with "wr_date get tohost",
becase after the host is correctly in 2014 or so, all leap seconds are
in the past.  But if we are not synced, the host is in 1970 and
no leap second has already happened.

If this code is unchanged in 2018, we'd better be off by 2 seconds
than 37, so if linux reports a date earlier than 2014, use 2014 and
fix tai_offset at 35.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This is a short-term fix to actual user needs. We'll soon
have a kernel module to track WR time, but most likely people
won't use that module until the next release.

Meanwhile, the normal "wr_date get" is enhanced to report both UTC and
TAI.

This wr_date is a simple plug-in replacement for installed switches.

Example, after setting a wrong host time on purpose:
(the example above includes extra blank lines to separate the output
from the various commands)

   wrs-192.168.16.242#wr_date get; date; wr_date get tohost; date

   1416482191.851373280 TAI
   2014-11-20 11:16:31.851373280 TAI
   2014-11-20 11:15:56.851373280 UTC

   Mon Nov 10 09:08:14 UTC 2014

   1416482191.895071872 TAI
   2014-11-20 11:16:31.895071872 TAI
   2014-11-20 11:15:56.895071872 UTC

   Thu Nov 20 11:15:56 UTC 2014

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

I'm confused by TAI/UTC: let's wr_date make it clear that it is
reporting TAI.  Nobody parses this output so far: the web interface
uses it, but it prints back the whole output.

So this change is safe.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Every other place calls the file wr_date.conf.

userspace/tools/wr_date.c will open the config file (it does not,
currently) but the name is different.

So we fix the executable and leave the rest as it is, since it works:
documentation matches S70wr_date and use of .config for build-time settings.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

moreover, adjust fractional part if integer part is already ok

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>