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

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

I got wrong several times with this. Time to fix.

The correctionField is part of the timestamp, so it's better dealt
with when the timestamp is recovered.

NOTE: correctionfields in the "response" direction should be subtracted.
We always added isntead, and we keep addition for compatibility.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Commit cc014e83 centralized the checks, but this was lost
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

We are not yet at a full 64-bit clean implementation. Actually,
printing secs as unsigned 32 bits will survive 2038, but pp_printf
with the default XINT implementation doesn't support unsigned. And
XINT is still kept because it's smaller, important issue on our nodes.
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: Adam Wujek <adam.wujek@cern.ch>

Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

CONFIG_HAS_P2P is always defined as 0 or 1
revert 6de6467f
Use #if CONFIG_HAS_P2P == 1 instead
Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

This tool can be used to force overflows of the input fifo.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Currently it only exports the rxdrop/txdrop feature.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Avoid this for dropped frames:

diag-frames-1-wr1: RECV -2 bytes at        88.000000112 (type 8, follow_up)
diag-frames-1-wr1: RECV -2 bytes at        88.000000112 (type 8, follow_up)
diag-frames-1-wr1: RECV -2 bytes at    109520.000000088 (type 8, follow_up)
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

See gcc info pages: "Macros with a Variable Number of Arguments."
The ISO-C stadanrd doesn't allow varargs macros to get zero or more
arguments. So this fixes the problem using the gcc extension.

A standard-compliant alternative would be to have "fmt" included in
the variadic part, like this:

  #define assert(cond, ...) \
        if (CONFIG_HAS_ASSERT && !(cond)) \
                __assert(__func__, __LINE__, 1 /* forever */, __VA_ARGS__)

But the reader wouldn't  now it's a fmt+args.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

For wrpc-sw, this depends on a similar commit in wrpc-sw, because
arch-wrpc relies on that implementation (to avoid building two of them).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

arch-wrs/Makefile was building arch-unix/conf.o . Make a local copy
instead, as we'll add more fault-injection stuff.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

In WRPC calibration took too long, it was possible that some frames received
(including PTP) on a network interface were lost due to RX fifo beeing full.
Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

Another try to fix makefile
Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

A stupid error when changing data type. Pity me.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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

At frame level 1, we should print tx/rx timestamps we get.
Also, fix the printout to use %llx for the "secs" field, now that
is is 64 bits (2038 is coming).

Meanwhile, remove some redundant "if (t)", because all frame operations
have called with a valid timestamp pointer.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

bah! next commit
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This bug was introduced in commit

   4c6953fb unix and wrs socket: open in non-blocking mode and report ...

because I unwillingly removed the ptpdump feature for outbount udp
frames. This puts back in place the lost message.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

To avoid triggering .config target at clean.
Put explicit target for $(CURDIR)/.config
Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

This change is needed by WRPC, since it wants to include file with shmem
version in the assembly file.
Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

To avoid triggering .config target at clean.
All *_defconfig and *_config from configs directory will fall into this target,
but not .config.

NOTE: target for .config is executed at include of a file (see makefile manual)
Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

In wrpc's SNMP we store an offset between beginning of a structure and
particullar field. To save memory we store this offset as a uint8_t, which
limits the offset of a filed to 255 bytes. It was necessary to move following
fields below 255 offset:
--n_err_state
--n_err_offset
--n_err_delta_rtt

Compiling wrpc-sw with an offset greater than 255, a compiler prints warnings
like:
lib/snmp.c:405:2: warning: large integer implicitly truncated to unsigned type
Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

To avoid triggering .config target at clean
Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

Signed-off-by: Adam Wujek <adam.wujek@cern.ch>

Intermediate commits in the branch do not work or do not even build,
but I'd better keep the various parts separate.

We now have a unified "struct pp_time" everywhere, so se avoid a
number of conversions, some code is more readable and we are
eventually free from the 2038 bug.

All time is now represented as 64b seconds and 64b scaled_nsecs. The
scaling is 16 bits, like the correctionField, which is now added directly
to the time stamps.

Please expect some more housekeeping edits in the area, like
factorizing commong code and constants, or removing fields that are
not needed any more.