The function st_com_restart_annrec_timer is now just a one-liner, and
it really is just a timeout operation.

So call it pp_timeout_restart_annrec instead, and have it inline in
ppsi.h, removing it from common-fun.c
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

The frames should be sent with some dispersion around the timeout
point.  The new function calculates the correct dispersion, but we
need to change the setting of next_delay in order to actually spread
the distribution of announce and other messages.

Also, this commit doesn't spread all message times, some work on the
field is still needed.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

It look like the timeouts were not restarted after they expired (and
the respecti message is sent).  The previous commit, that disables
expired timers automatically, exposed this bug.  Unless I misunderstood,
the 1s timing of announce message was due to the 1s-long delay
between re-invocations of the state-master code.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

commit 2ce296a5 removed this stanza, but it looks like we actually
need to force IP_MULTICAST_LOOP to zero.

By zeroing the timeout whenever it expires, a number of
state files get simplified a little. So pp_timeout_z() returns
1 or 0 like pp_timeout(), but also clears the timeout when it is
expired.

Moreover, when a timeout expires a message is printed if the
build-time option pp_verbose_time is set (and run-time verbosity is at
level 2). In this way all timeouts are reported, or none of them.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This kills all WR-related timeouts because one of them
is enough (they are used in each state as a timeout for
completion of that very state).

Moreover, since ppsi should be extension-independent,
we now define 4 timeouts for the extension in use (and WR
only uses timeout 0).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This bug has no effect, but I'd better commit it separately for
documentation purposes.  Actually, there is not need to clear the
timeouts when they are state-specific and we are changing state.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

This fixes the bug exposed by the previous commit.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The timer implementatin was incredibly complex.  Now there is no such
thing as a timer in ptp. We only have timeouts.

A timeout is a point in time. We calculate; we can be earlier; we
can be later. The only thing which is arch-specific is calculating.

This replaces every "timer" operation with the equivalent "timeout"
one. All stuff is renamed to avoid confusion between "time" and
"timer".

Unfortunately, the resulting program as of this commit is not working.
the pdelay_request timeout happens continuously, I'm sure I've exposed
a latent bug, which I'm now going to fix.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

We had bare_timeval with both nsec and usec, but this doesn't work.
As a result, our timers were only second-grained and not
millisecond-grained (we got no nsec from clock_gettime).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This is a subset of <linux/jiffies.h> with typecheck from
<linux/typecheck.h>.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This small patch-set removes the MSG_SEND_AND_RET function and its
variants.  I find it extremely hard to read multi-line macros,
and also having a function returns implicitly by calling a macro.

I understand why this was born, but now I feel really better with the
code.

This removes the unmanageable MSG_SEND_AND_RET macro, turning it to a
normal function (inline, at the time being, to avoid moving from the
header to a different place).

The function is called __send_and_log because it logs too, but I'd better
move logging to the actual send and recv methods.

BTW: the message "-> FAULTY" (now removed) was wrong, because not all
errors of this function result in FAULTY state in the fsm.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This kills MSG_SEND_AND_RET_VARLEN, turning everything into
MSG_SEND_AND_RET. Such macro now receives the length argument and
doesn't receive the "pdelay" argument any more, because it is zero by
design in this implementation.

This commit has no effect on the compiled files.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

I'm sorry, but I need names better than x,y,z,w . This is my first
step in getting rid of this complex macro.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

No effect: the final program is still called ptpdump.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This allows to pass "-V -V" or "-g" (slave only) and so on.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Such stuff is needed for command line parsing.
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 is accomplished by making msg_unpack_header return an error code
(and by warning users who ignore it).  We *really* don't need our
own frames back -- even if they come from a different port.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The second time you do ADD_MEMBERSHIP to a multicast address, the
ioctl returns ADDRINUSE and ptp goes to faulty state.  By removing it
stuff works.  The issue is there since commit f44fa35b, the initial
UDP implementation, which hasn't been used in practice.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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>