This works around a design problem that must be checked carefully.
The "is from current parent" flags set at receive time reveales bad,
as we should only check when needed.

So by now just say it is the current parent if we don't have a current
parent yet (i.e., we are not in error, even if it is not from current
parent).

Also, forget about current parent and foreign masters when the servo
is initialized (e.g., at link-up on the wr node).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

wr_ops->enable_timing_output checks the current status and only acts
on hardware when the pps must be changed (turned on or off).  Thus,
all internal uses must refer to this function, or we'll get out of
sync between actually pps activation and internall-known state.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This was wrong (and segfaulted) since we introduced multi-link operation,
but I only tested bare-i386 today.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

This only works with wrpc at this point, due to the missing poll
for WRS lock to 10MHz.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The locking_poll operation needs to differentiate between slave
and grandmaster mode (we use this in the wrpc to change the clock
class, according to whether we are locked to the 10MHz or not).

The WR-Switch implementation is missing at this point.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This is the last bit of the overhaul made by the previous commit,
and has no technical effect. (In a perfect world, the previous commit
would have been one per function, to easily find errors, but I had not
enough time; this is different because it was a two-level call).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This is a massive change, but one that we needed since time 0.
Aliases were a bad choice, while use of an operation structure shows
better the symmetry between the two white rabbit implementations.
But, mainly, maintainance is easier.

I really needed this because I need to change one prototype, to
add support for wrpc acting as grandmaster.

There are still some asymmetries to be fixed, but the thing is
acceptable at this point. While I was at it, I removed some
WR_DSPORT(ppi) using a local pointer "wrp".

The code after this commit is shorter by a dozen bytes or so,
both on wrpc and wrs.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Currently implementations of WR functions do not include wr-api.h.  I
need such consistency check to have compiler help in creating the
wr_operations structures.  Anyways, files should always include its
own header.

I add wr-api.h in arch.h for laziness, so I get all files for sure.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

timeint_to_wr was actually copying data to a local uninitialized
variable of the same type as the input one.  The uninitialized
structure generates warnings about unitialized fields (at least on
newer compilers than the one we use on wr-switch).

Instead of clearing the local variable, just use structure assignment
and get rid of the function.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

When adjusting the time, we expect stamps to be invalid for a while.
Don't print errors unless the fact repeats unexpectedly.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This centralizes a "software" enable timing output, that calls the
"hardware" one (previous code) while keeping track of the status.

With non-wr the pps output is always enabled if the offset from
master is less than 1s.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

This sets the clock class in arch-wrpc in a way more aligned to the
standard. A grandmaster is 6 if locked to atomic or gps, but
52 if it is designated grandmaster but doesn't lock to a primary source
(this is was we poor developers experience).  A "master" is 187,
the old default, and a slave-only remains 255 as it has always been.

This also removes the ability to stop the program while it waits
for the pll to lock. It can only succeed or timeout.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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

This defines PP_CLASS_SLAVE_ONLY and similar names, and uses them properly.

Also, the global "struct default_rt_opts" was not following the
conventions about naming, so rename it to be a "pp" thing.  Moreover,
since we put all globals in ppg and we have ppg->rt_opts, try to use
that instead of a separate global.  Unfortunately the initialization
sequence is not well ordered, so __pp_default_rt_opts still remains
as a global variable, double-underscored to show it's not meant to be used.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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

wrc_ptp_ppsi.o was not picked any more from libarch.a into ppsi.o,
now that we don't refer to pp_diag_verbosity any more. So force
linking it, or the final wrc.elf build will fail for undefined symbols.

Also, remove a no-more-needed -I directive.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

We are already not using "const" verbosity in wrpc-sw, which is our most
size-constrained architecture. So remove this legacy support.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This is a long-standing bug introduced when fsm.c was changed to
immediately enter a new state (before that, we had a grace period in
state-faulty).  The problem is exposed by running the suggested
config file, if eth1 and/or eth2 exist but are currently down.

This patch introduces a specific timeout, so we wait 4 seconds before
reinitializing the pp instance; we thus avoid burning 100% CPU Time in
trying again and again.

As a side effect, this reduces timeouts provided to extensions from 4
to 3. I could have reused one of the other timeouts, as faulty state
does not use them, but I'd better have proper timeout names in
diagnostics.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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

The "port" config word is not creating a new port any more, it locates
an existing port, if any is there.  This allows, for example, to set
diagnostics from the command line:

	    -C  "port eth0; diagnostics 022"
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This option sets configuration options from command line,
as specified in CONFIG_ITEM.
CONFIG_ITEM must be a valid configuration string, so you'll likely
need to quote it on the command line.

The semicolon can be used as a separator, to stick several
configuration items in the same argument.  This option can also be
used more than once on the same command line, to set different
parameters. If used twice to set the same parameter, the last value
overrides previous values -- and later config files, specified with
"-f" may override -C values.

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

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

The structures are going to be exported, so use the pp_ prefix.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

This undoes part of the strangeness we had in configuration files. Now
"-f" is recognized on the command line, and more than one file can be
specified. Only if no file is found, the default one is used
(wr-switch looks in /wrc/etc/ppsi.conf first).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

This is a rewrite of the configuration parser, more based on data
structures se we'll soon be able to read more than one configuration
file (to add special cases while developing) and protocol extensions
will be able to extend the table of allowed options.

Both the configuration file format and the layout of the code calling
lib/conf.c::pp_config_file() is unchanged in this commit. This means
that even if you enable config diagnostics on the command line it
won't have effect, because the config file is parsed before the
command line. This is fixed by a later commit.

The only visible change in this commit is that "port" is allowed as a
synonim for "link", to be deprecated soon (but it will not be removed
from the parsing of the config file). Thus, pp_instance->link_name is
renamed to port_name.

The compiled size of this commit is the same as the previous one.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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