pdelay (peer delay) is a configuration option for a PTP domain. IT is
a huge mass of code, which we never actually used and is most likely
buggy.

This commits remove all related code, and configures our PTP engine as
end-to-end.  I doubt this commit can be reverted in the future, but at
least it leaves traces about what was there and is not there any more.

Something about pdelay is still there: the message names, one field in
a structure and one configuration variable, to ease recovery in the
future.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The master chooses whether it is one-step or two-step (page 66).  It
doesn't make sense to be one-step (but we accept one-step masters when
we are slave).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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

IN this way users can tee the messages to a file and avoid the
4k-at-a-time bursts.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This somehow voids the effort of two commits ago. But we really don't
need to save space when building a full exectutable linked with libc
and all the rest.  As a side effect, it allowed me to discover the
wrong varargs to some messages (see previous commit).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

printf with missing argument. Introduced in a8733f10, 52 commits ago.

I didn't notice because I was not using CONFIG_PPSI_RUNTIME_VERBOSITY
(as expected, the preprocessor hit again).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The pp_Vprintf (Verbose printf) macro, replaces PP_VPRINTF in the file
to ease generation of dumps.

All messages are build-time protected by "ifdef VERB_DUMP" anyways:
by removing the dependency on CONFIG_PPSI_RUNTIME_VERBOSITY it's
easier to enable the messages at build time because you just need
to set

   USER_CFLAGS="-DVERB_DUMP"

without the need to force RUNTIME_VERBOSITY as well.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The big dumping stanzas are already conditional under
if(pp_verbose_dump), so they only exists if VERB_DUMP or VERB_LOG_MSGS
is defined at build time.

This commit adds the same build-time protection to other messages
related to packet dumps (all of them rely on a build-time protected
subfunction anyways).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

The three "raw" fields claim to exist for debugging, but since we
have several TimeInternal structures around they cost quite some space.
The fields were only copied to TimeInternal from network frames,
and there is not explicit code to dump them.

Thus, this commit removes them but leaves the code in place, so it can
be re-enabled if the need arises. (I use the ugly "#if 0").
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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

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

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

sometimes I jumped to "no_incoming_msg" in case of error. This renames
the labels to "out", as the catch both error and normal exit paths.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

set_TimeInternal was partial anyways, since TimeInternal is more than
seconds and nanoseconds (even if only WR uses it).  It was only used
to zero the values, so use memset instead, within the new
inline clear_TimeInternal()
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

This stuff must be reviewed again anyways, because I'm not sure at
all about what these numbers add (I can't find proper docs or examples).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

This introduces "-DDIAG_ERR" at build time and pp_error() as a
conditionally-compiled and conditionally-enabled thing.  I'm not sure
error messages should be missing from the binary by default, but I'll
evaluate better at a later time, after debugging my own errors.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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

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

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

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

Now you can set VERB_DUMP, VERB_SERVO, VERB_FRAMES, VERB_TIME
separately.  For example

    make USER_CFLAGS="-DCONFIG_PPSI_RUNTIME_VERBOSITY -DVERB_SERVO"
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This has currently no effect, but later I'll have an option to enable
verbose messages for individual contexts (message dumps, servo
operations, frame send/receive, time operations).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This only removes the hairy ifdef, to convert them to normal if
functions.  There is no effect on the generated binary, but it's now
less easy for me to introduce errors.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This commit makes no technical changesin the output, but it creates
<ppsi/diag-macros.h> (always includes) to split out all the
diag-related stuff.  The macros in there are the same as in previous
commits, but they are documented.

This also introduces pp_verbose_messages (which is 1 or 0) to allow
the various "#ifdef VERB_LOG_MSGS" to become language if() instead.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

There is no way to set VERB_LOG_MSGS. This Makefile change allows
to set USER_CFLAGS, for example like this:

   make USER_CFLAGS="-DVERB_LOG_MSGS -DCONFIG_PPSI_RUNTIME_VERBOSITY"

and then use "-V -V" to enable all messages.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

pp_printf calls puts, but that's the "embedded" puts, the one that
doens't add a trailing newline. So to avoid extra lines and split
messages in the diagnostic output we need to link the fake puts, that
calls pp_puts instead (in turn, pp_puts for arch-gnu-linux relies on
fputs, which doesn't add the trailing newline).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The time operations are get, set and adjust. They are methods of a
single structure in order to simplify use of custom timers (and make
them separate works according to the LGPL).

This patch set introduces specific <arch>-time.c files where all the
timing stuff lives (not much, actually). Sometimes this turns
<arch>-io.c into an almost-empty file, but that how history changes
stuff.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Nobody was implementing this timer_adjust_all, and I'm going to
rewrite the timer management anyways, because it is a horrible
complication just to manage some timeouts for packet transmission.

So let it RIP, and let's happily leave without it.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Instead of having several external functions for networking, only
have one symbol: the structure of network operations. This simplifies
stuff a while, and allows more static symbols and better encapsulation.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This bug was not exposed, as our extension offers both the open and
close hook. Bug introduced in 9e4a90c7, 20 commits ago.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

I confess I repent having started this. And some of the files are really
in a terrible status and need rewrie anyways (I'm not pointing fingers,
I know the history of the project and I'm aware this is what I deserve).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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