There is still some factorization to make, and other irrelevant
size-related things, but I'd better wait a little and get confident
this stuff is solid.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Commit only build for arch-unix (and no tools either)
This simplifies a lot the calculations.

  - all times are now living in a single structure
  - all fields are signed and 64-bit long
  - the 2038 bug is over (we still had 32-bit seconds)
  - no overflow in math or unexplained "unsigned" appearing
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

The idea that somebody wants to steer time for >1s offsets is
perverse.  Let's simplify (if any, we may want to jump for <1s offsets
that are bigger than a few milliseconds).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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

While it may make sense to setup constant delay somewhere (for WR we
define them, for instance), the legacy implementation was unused and
wrong.

Unused because nobody sets it (even if we could do it in the config
file).  Wrong because inbound is never used, and both are global and
not per-port.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Now this is the behaviour for the choice:

    arch-bare-*

         if you select P2P, it's built and selected. Default is E2E_ONLY.

    arch-sim
    arch-unix
    arch-wrs

         The P2P is built by default configuration, and the mechanism
         is selected at configuration time. Default is E2E. If you
         choose E2E_ONLY and configure pdelay, it will ignore the config.
         (Suboptimal but I'm lazy: do *not* use E2E_ONLY for those archs.

    arch-wrpc

        The default is e2e until changed at run time.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

Here in ppsi,

  ./MAKEALL wrpc_defconfig wrpc_pdelay_defconfig

shows a very small difference, but most pdelay code is then
discarded by the wrpc-sw link time, due to --gc-sections.
This is the result:

   laptopo% ./MAKEALL spec_defconfig spec_pdelay_defconfig
   ##### Building with 'spec_defconfig'
   /opt/lm32-gcc-4.5.3/bin/lm32-elf-ar: creating libsdbfs.a
      text    data     bss     dec     hex filename
     87688    3492    6352   97532   17cfc wrc.elf
   ##### Building with 'spec_pdelay_defconfig'
   /opt/lm32-gcc-4.5.3/bin/lm32-elf-ar: creating libsdbfs.a
      text    data     bss     dec     hex filename
     93140    3492    6360  102992   19250 wrc.elf

What is missing now is the run-time choice between e2e and p2p. Later....
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This fixes a bug introduced by f1d048c3
(conf.c: use f_simple_int() for assigning simple integer options):
we used ASSIGN_INT_FIELD for assigning values to variables shorter than
int, while the destination pointer was set to int *, so neighbouring
fields were overwritten.

Another commit in a non-merged branch adds a build-time check to
avoid repeating this bug, but field names changed, so let's avoid
cherry-picking it here at this point.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

This commit ensures all architectures do the proper things to work
with either end-to-end and peer-to-peer, and all frames use the
proper addresses

 * we avoid ifdef by using HAS_PTP and HAS_E2E
 * no need to -D config items in Makefile
 * wrpc now correctly sets ppg->delay_mech at startup
 * the simulator now correctly sets ppg->delay_mech at startup
 * bare archs correctly set ppg->delay_mech at startup
 * bare archs properly use one or the other multicast address
 * pp_instance_cfg->delay_mech is removed as not yet used
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

the calculation of the link delay and offset to master is moved from
the fat wr_servo_update to tree functions which will be called
according to the delay mech. In the case of pdelay, the link delay
calculated and then the offset.
Signed-off-by: C.Prados <c.prados@gsi.de>

Conflicts:
	include/ppsi/pp-instance.h
	include/ppsi/ppsi.h
	proto-ext-whiterabbit/wr-api.h
	proto-ext-whiterabbit/wr-servo.c

pclock handles/issues the PDelay messages.
Signed-off-by: C.Prados <c.prados@gsi.de>

I will keep the t1..t4 for calculating the
peer path delay and these new are used only
for synchronization
Signed-off-by: C.Prados <c.prados@gsi.de>

In PDelay mechanism, the Master has to reply to PDelay_Req from pclocks
with PDelay_respond and PDelay_follow_up. Add pack/issue functions
for this messages
Signed-off-by: C.Prados <c.prados@gsi.de>

[WARNING: this commit was changed by Alessandro, adding a missing
"pdelay_req_hdr" field in pp_instance, and exporting msg_copy_header
that had been turned to static -- this must be fixed]

functions for set/get the mechanism.
Signed-off-by: C.Prados <c.prados@gsi.de>

With UDP we have two channels but we talk with the same peers.
Moreover, we reply with a general message to an event message, so
having two peers was plain wrong.

I'm well aware this commit should happen before vlans are introduced,
but I don't want to deal with all the conflicts in moving it back.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The vlanhdr is not something I can find in standard include files,
so here is a custom definition, called pp_vlanhdr.

Adding peer_vid is needed so a master can reply to its own slave,
while listening to several vlans.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This step is in preparation for vlan support, where the tx and rx
headers have a different length.

This commit doesn't build for arch-wrpc, fixed in the next commit
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This "netpath" is a pain, completely needless. The NP(ppi)
is overhead in the code and must be removed. It is now turned to
an identity function, so external patches won't conflict.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

Add counting of tx and rx PTP packets.
__recv_and_count and __send_and_log are arch independent functions.
Increase WRS_PPSI_SHMEM_VERSION due to add of tx_count and tx_count firlds in
pp_instance.
Update WRS_PPSI_SHMEM_VERSION to 5
Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

Part of a process to move wrs-servo internal data into shmem
Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

In order to fit all pp_instance items withing WRS shared memory,
we need to make them a little smaller.  Thus, the tx and rx buffers
are now allocated separately.  And, while I am at it, I make them
smaller, because 128 bytes are more than enough for PTP with
extensions (all frames are under 100 bytes). Given we are always
showr of memory in wrpc-sw, this is benefical.

As a side effect, clean up some out-of-memory exit paths, and fix a
use-before-check buglet in arch-unix (no, we never go out of memory,
but the check was wrong).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This field is not needed, because we issue the response immediately,
using the last received header. The copy is not needed, nor the field.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

This paves the way to the addition of a third protocol. Besides the
code is slightly smaller (around 50 bytes on all archs).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This reverts commit 9a3e00784afea6ab46510315b9c50cb510175f2d.

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

Bitfields are bad, so this turns them into flags and constants.
There should be no change at all in behaviour (although the code
is a little smaller).

The next commit does a change for real.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

I'm going to introduce a "flags" field in pp_instance, so diagnostic
flags are better called d_flags. The same applies, for consistency,
to the global variable.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

Bitfields are a bad practice. And this removes a few bytes in the wrpc
builds (no change in other builds).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Otherwise, if we have two PTP ports for the same interface, like
Ethernet and UDP, we are not able to tell which message is for which.

This now works as expected:

./ppsi -C "port eth0;  iface eth0; role master; proto raw; diagnostics 111" \
       -C "port eth0u; iface eth0; role master; proto udp; diagnostics 111"
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This adds frame dropping as a fault-injection mechanism. It activates
it for arch-unix in the same commit, because I tested it.

Not documented yet.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

There is no need to remove outliers and to make a running
average on ofm. Indeed we know that ofm = t2 - t1 -mpd.
We are already removing outliers and averaging on mpd.
So, if after mpd has been "cleaned", ofm still shows outliers
or irregularities they are for sure due to t1 and t2. So they
are errors coming from the clocks and they have not to be
ignored. Instead they must be corrected ASAP, and to do that
the software must be able to see them instantly. So no
running avg nor outlier removal has to be performed on ofm values.