- backup the existing .config file, if present, and restore it before exiting
- make clean before exiting

Since check_packet is using network channels (file descriptors), it is
better declared as a network operations.  This allows, for example,
to make a simulator using ARCH=unix and only changing TIME= .
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This simplifies the build system, by relying in weak symbols instead
of the link order. We need weak symbols anyways for configuration, so
we can get rid of the libraries.  The size of binaries returned at
build time is increased, because arch/ and lib/ stuff is not part of
ppsi.o (as it actually should).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

The tool now builds all configs/* setups, but it has fewer options than
it used to have (no command-line arguments as yet, so "./MAKEALL bare"
will not just check bare configs.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

This adds the needed configuration items: architectures, extension,
cross_compile and wrpc_root.  So, what we passed on environment and
command line is now in the configuration file.

This changes the priorities of such options, so the Makefile is
changed to force $(CC), $(LD) and $(CROSS_COMPILE) down to subprojects
(i.e. pp_printf and arch-wrs/mini-rpc).  Also, we need to remove
quotes from .config items, and this is ugly linesin ./Makefile
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Now Kconfig works, but we have no active configuration at all.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Add the lines from the .gitignore found in bst-kconfig. As in the
previous commit, use bst-kconfig commit v2.6.34-4-g8168c0b

(cherry-picked from wrpc-sw d923523a)
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This is the combination of the following commits (oldest on top)
from wrpc-sw:

  89cdfcd added kconfig infrastructure, the trivial way
  5fda0c8 kconfig: use pathnames from Linux-2.6.34
  ab30804 kconfig: use 'Kconfig' as file name, not the bst-kconfig ones
  7c78880 kconfig: trivially write "WRPC" in place of "Linux Kernel"
  2c30cb4 kconfig: trivially fix most 'kernel' words in messages
  70eb3ba kconfig: add a configs/ directory and use it
  64cc10f scripts/kconfig: don't complain for default in choices

After picking them, I replaced all "WRPC" with "PPSi", and I removed
the configs/  files (unrelated to this project).

In practice, at this point we have the Kconfig infrastructure of
Linux-2.6.34, which I picked from the "bst-kconfig" package, which did
the extraction work.

One day I forward-ported this from v2.6.34 to v3.4, by applying a few
hundred kernel patches. While the final code was working, we decided
the port was not needed for wrpc-sw. Here in PPSi I keep the same version
we have in wrpc-sw.

As of this commit, the code is not used; the default configuration
file and Makefile targets are added in the next commits.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Bug introduced by

   ce8485bf bmc: remove field ppi->frgn_rec_i

I forgot to run ./MAKEALL during the last commits.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This is needed for the wrpc-sw release, as Etherbone won't fit with
a full ppsi in the default RAM size for SPEC cards.

ptpdump support will soon become optional in ppsi, through Kconfig,
and a bigger RAM size will become the default for SPEC images as well.
But currently we need this hack.
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 was already expexted to happen, but it was done in the wrong
place (my fault).  Most likely we should move this
"enable_timing_output" in the ptp-wide operations, to be used
outside of WR as well.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The field is a duplicate of ppi->frgn_rec_num, and got out of sync.
This removal fixes WR syncronization as a slave.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Also, report how may are there in bmc diagnostics
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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>