This pair of commits costs around 300 bytes overall, in network-using
configurations.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

We are going to add more UDP services, so this commit adds the
generic management of UDP sockets, with udp protocol and port matching.
Thus, we now have two IPV4 sockets: one for ICMP and one for bootp;
whose receive part relies on generic code for the match.

We still miss the output helpers, so with this commit bootp still
builds the whole frame by itself.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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

I also change the prototype of the socket creation so the compile can
warn me where the locally-allocated socket must be changed.

The size saving is not huge, but this allows later creation of further
sockets with a smaller overhead (e.g.: the syslog socket that sends only
has no queue, and the snmp socket needs a very small buffer).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This has no effect, but greatly simplifies the later vlan-based operation.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The misbehavior can be reproduced  with

   ping -s 422 192.168.16.200

now, the first frame is fully stored to the queue:

   copy_out: head 368 avail 512 len 2
   copy_out: head 370 avail 510 len 14
   copy_out: head 384 avail 496 len 24
   copy_out: head 408 avail 472 len 450

but it is retrieved only partially:

   copy_in: tail 368 avail 22 len 2
   copy_in: tail 370 avail 24 len 14
   copy_in: tail 384 avail 38 len 24
   copy_in: tail 408 avail 62 len 400

The tail of the frame remains in the queue, and is
not correctly removed. So we have:

    update_rx_queues: queue for socket full; [avail 462 required 490]

(because 50 bytes are lost there). And the next rx frame,
the tail of the previous one is considered size, then truncated
to the payload buffer size (here 400 for icmp), so we never realign:

   copy_out: head 346 avail 462 len 2
   copy_out: head 348 avail 460 len 14
   copy_out: head 362 avail 446 len 24
   copy_out: head 386 avail 422 len 84
   Q: Size 84 head 470 Smac 0:4:25:12:34:56
   update_rx_queues: saved packet to queue [avail 338 n 1 size 124]
   copy_in: tail 296 avail 338 len 2
   copy_in: tail 298 avail 340 len 14
   copy_in: tail 312 avail 354 len 24
   copy_in: tail 336 avail 378 len 400

So the available space in the 512-big queue explodes (and ping gets no
replies):

   copy_in: tail 120 avail 1326 len 400

In the caller I got one reply and nothing more, not even the short pings:

   68 bytes from 192.168.16.200: icmp_req=1 ttl=63 (truncated)
   5 packets transmitted, 1 received, 80% packet loss, time 4009ms

With this commit everything flows

   68 bytes from 192.168.16.200: icmp_req=1 ttl=63 (truncated)
   68 bytes from 192.168.16.200: icmp_req=2 ttl=63 (truncated)
   68 bytes from 192.168.16.200: icmp_req=3 ttl=63 (truncated)

   64 bytes from 192.168.16.200: icmp_req=1 ttl=63 time=2.02 ms
   64 bytes from 192.168.16.200: icmp_req=2 ttl=63 time=1.61 ms
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Several advantages:
 - normal practice: you send a buffer pointer and its own len
 - rx values is already fixed, earlier the wrong result was wrongly used
 - this detaches us from the header length. Transparent vlan ahead...
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

We removed the pool of sockets, to only keep a list of pointers.
The caller provides the socket (and the buffer).

This saves from 440 to 1570 bytes in the binary (more in the ones
with no CONFIG_IP, beause sockets are not used.

Now we can move to a caller-defined queue size, when I find the time and
willingness.
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>

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>

This changes the pfilter rule-set, to make it ready for vlan addition.
Moreover, it removes the choice between three rule-sets and uses one
that works for everyone (we'll add the vlan option later on).

Moreover, we prepare for the option of peer-delay PTP and UDP-based PTP.

As a side effect, one Kconfig option is removed.

We know for sure that what goes to the "streamer" protocol
is re-checked, so the streamer can get some "everything else", exactly
like the 7S wr-nic (that sends to the host everything unused).

This is the current rule set:

     - everything tagged is dropped
     - CPU: arp broadcast, not unicast (i.e. requests only).
     - CPU: PTP ethtype. Any mac address.
     - CPU: ICMP unicast, not broadcast.
     - CPU: UDP (uni/multicast), ports  bootpc, ptp-event, ptp-general.
     - Etherbone (class 7): UDP (uni/multicast) and port 0xebd0
     - Other fabric (class 6): anything not Etherbone (inverted class 7 bit)

This counts up to 28 rules. We have 4 more rules available and we may
add options, either at build time or run time to add udp ports or
other ethertypes.  Or not...

2 caveats:

  - frames for the CPU will have class bits 0x41 instead of 0x01, because
  everything not etherbone will receive bit 6. This is not a problem, as
  the CPU is not checking the class bits.

  - the wr-nic gateware, when using this new sw code base, must be
  changed to use classes 6 and 7 like everybody else, not classes 7
  and 5.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The description of class management was not completely clear or correct,
so it's now rewritten with some references to gateware files.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Actually, the "new" information is very little.  But now the
explanation of how it works fits in 80 columns (I personally need
this) and is laid out a little better.  Also, it describes the new
situation, activated by the next commits.

No code change at this point.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Otherwise, the macro won't build on 64-bit hosts
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>

This removes TRACE_DEV(), turning it into pll_verbose() and wrc_verbose().

The functions are enabled and disabled by kconfig, with defaults matching
the previous behaviour.

So nothing changes, but the binary is smaller because the previous
wrc_debug_printf() was silent because of a run-time conditional
(though hardwired false at build time). The condional is now at build
time.  We save around 2k, between strings and calling code, in
all wrc builds (nothing is saved in wrs, which is verbose by default,
nor in devel_build_test_defconfig where all verbose options are on)).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

devel_build_test_defconfig: enable all verbose options
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The makefile included trace.h for everyone, which in turns included
wrc.h.  It's better to include wrc.h straight on (we can't rely on all
files including it, for example pp_printf does not).

Meanwhile, some redundant includes are removed, and some are added
(I prefer wrc.h to be explicitly listed in the files that use it,
even if the command line set in Makefile already prepends it).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

The misleading TRACE thing was only used in wrs builds, and expanded
to pp_printf.

Considering we don't want to remove messages from the softpll in wrpc,
this commit turns trace onto pp_printf.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

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

Otherwise, the first "make" builds tools before spitting the
error about an unconfigured system. I prefer to have the error alone.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This quiet=quiet_ has the effect of printing

  HOSTCC  scripts/kconfig/conf.o
  HOSTCC  scripts/kconfig/kxgettext.o
  SHIPPED scripts/kconfig/zconf.tab.c
  SHIPPED scripts/kconfig/lex.zconf.c
  SHIPPED scripts/kconfig/zconf.hash.c
  HOSTCC  scripts/kconfig/zconf.tab.o
  HOSTLD  scripts/kconfig/conf

instead of the full command lines. We could even use quite=silent_ to
have no output at all, but this would introduce a noticeable delay in
the build.

The aim of this change is making more visible the error "your wrpc is
not configured" at the first build. The full compilation of config
stuff makes the error less clear, and the initial error confuses
people (me included).
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This reverts commit 3a3c3806.

The previous line: every object depending on .config, was better.
We had the problem that a fresh clone was not building because of a
missing autoconf.h.

The depend rule was not working anyways, because it created an empty
.depend the first time over, due to the missing autoconf.h.  I admit I
don't know how to create a depend file after configuration: make
pretends to remake all its inclusions as the first step, so a rule
like ".depend: silentoldconfig" loops forever.