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

This adds the build choice about whether to use the older sockitowm or
the new "w1" implementation, which supports all three thermometers
but is still missing eeprom support (the API is defined, though, see
include/w1.h

This also adds a shell command, called "w1", which rescans the bus and
tries to read the temperature from all devices. In the example below
the third device is an eeprom, so it return an error for temperature
read-out.

   wrc# w1
   W1: 68000801dce56910
   W1: f70000001eda8242
   W1: 5f00000040e50143
   device 0: 68000801dce56910
   temp: 34.7500
   device 1: f70000001eda8242
   temp: 32.5000
   device 2: 5f00000040e50143
   temp: -32768.0000

The code is properly integrated in wrpc-sw, and the "first" thermometer is
used to build the MAC address of the device, like it used to be with
sockitowm.

The binary built with w1 is 3kB smaller than what is build with sockitowm.

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

This adds the files for an alternative to sockitowm. I called it "w1"
like in the Linux kernel, because "onewire" was already in use as file
name in this project.

The bus scanning code was a fun project of mine a few years ago, the
rest is what I did for my own crappy operating system in the last few
days.  This commit adds the files as modified to build in wrpc-sw, since
I have no "official" Bathos files yet to diff against.

This commit doesn't change the build, only adds the file, so actual
changes to existing code is all part of the next (smaller) commit.

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

This is the host part of the software-only uart. The lm32 part is
selected at compile time (CONFIG_UART_SW).

The tool, when run, opens one or more pty pairs, and reports to stdout
which is the name of the slave descriptor. You can then run
"minicom -p /dev/pts/XXX" or other commands, to access the software uart.

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

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

This introduces dev/uart-sw as an alternative to dev/uart . The
software-only thing is a pair of circular buffers that can be driven
by tools/wrpc-uart-sw .

The software uart can be enabled by Kconfig (and is disabled by
default).  The code size is the same as the uart, but it "wastes" 306
bytes of data space.

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

This adds "sleep" that is used to verify the usleep function.
It is not build by default nor by a Kconfig option, because it is
not expected to be generally useful.

To build, add "CONFIG_CMD_SLEEP=y" to the command line of make.

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

This adds a real usleep implementation. The init function calculates
the "loops per jiffy" variable, by watching the timer ticks before
interrupts are enabled by spll_init().  This init takes 40ms.

Then, the udelay runs the same loop used in calculating lpj, and
is thus correct (it will be slower because we have interrupts running
at run time, but this is not a problem).

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

This is a copy of ptp-noposix/libposix/freestanding-wrapper.c, moved
into this project because it actually is only used here. The Makefile
now builds this one.

This move is needed in order to remove this faulty usleep and provide a
better one, while keeping the other fake functions called by ptp-noposix
code.

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

1- wrc.h now includes syscon.h. This is a prerequisite to a non-inline
implementation of usleep (the current inline one is just a random
loop, 10x wrong on the spec). Without this we should modify sockitowm
where it calls usleep, but I'd better not touch sockitowm.

2- syscon.h includes <sys/types.h>.  This is needed to have
useconds_t.  I'd better have usleep take unsigned long, but some file
already includes sys/types.h, so our prototype must match that one.

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

This command can return the .config file to the user. It can be useful
to trace what is actually running, as addition to the "ver"
command.  It is optional because it costs .5kB of binary size.

Example:

  wrc# config
    Current WRPC-SW configuration:
  CONFIG_RAMSIZE=90112
  # CONFIG_PTP_NOPOSIX is not set
  CONFIG_PPSI=y
  # CONFIG_CHECK_RESET is not set
  CONFIG_PPSI_VERBOSITY=0
  CONFIG_PPSI_RUNTIME_VERBOSITY=y
  CONFIG_STACKSIZE=2048
  CONFIG_PP_PRINTF=y
  # CONFIG_PRINTF_FULL is not set
  CONFIG_PRINTF_XINT=y
  # CONFIG_PRINTF_MINI is not set
  # CONFIG_PRINTF_NONE is not set
  CONFIG_PRINT_BUFSIZE=128
  CONFIG_ETHERBONE=y
  # CONFIG_DETERMINISTIC_BINARY is not set
  CONFIG_CMD_CONFIG=y

(This configuration amounts to a binary size of 86252 bytes, at this
commit).

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

The help command currently only lists available commands. It was asked
by some developers some time ago, and this implementation only costs
150 bytes in the final exectuable.  I don't think it's worth a Kconfig
option.

This is it in action with my configuration:

   wrc# help
   Available commands:
     ver
     pll
     sfp
     stat
     ptp
     mode
     calibration
     time
     gui
     sdb
     mac
     init
     ptrack
     ip
     verbose
     help

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

The environment was drafted but not actually used, so this commit
removes it.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>
Acked-by: Grzegorz Daniluk <grzegorz.daniluk@cern.ch>
Acked-by: Tomasz Wlostowski <tomasz.wlostowski@cern.ch>

This commit creates a .cmd section, so there is no global list of
commands in shell.c any more, and no ifdef in the code for conditional
commands.  Also, (unrelated) makes shell errors slightly more friendly.

The technique is widely used, in the Linux kernel and all boot
loaders, as well as a lot of my own personal code, so this is safe
(and tested).

As a side effect, this allows (in later commits) to move the "verbose"
ppsi-specific command back to ppsi/arch-wrpc/ where it really belongs.
It allows new commands to be easily added by just adding their
respective file to the build.

Even though this increases code size by around 50 lines, it makes it
more modular and only costs 8 bytes in the final executable (in my
configuration: 128 bytes less of text and 136 bytes more of data).

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

This commit makes no effect, but it prepares for the following one.
In general, using an intermediate platform-agnostic linker script
when making wrc.o is not bad, so let's do it here.

The main reason why we need it is to force the "KEEP" command for the
".cmd" ELF section, introduced by the next commit.  The alternative to
this would be chaning the main linker script, which is worse in my
opinion.

If we add initcalls or similar stuff later, this bigobj.lds is the
place to do it.

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

The parameter used to be off by default in order to save space in the
generated binay.  Now ppsi is smaller and there's no reason to
prevent diagnostic code to be there.

This parameter, though, will disappear over time, as we have a new
diagnostic system, more modular (which still offers a compile-time
option to remove all diagnostic code).

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

The new configuation option CONFIG_CHECK_RESET adds a function to dump
the stack (to the serial port) if a reset is detected. This happens
when a null pointer is followed or when the program is reloaded
(spec-sw/tools/spec-cl, for example).  For this reason, the feature is
disabled by default (actually, spec-cl may clear the RAM first, but it
currently doesn't do that and the check is triggered).

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

Clearing the screen is in general bad practice: it hides information
for no advantage (ok, I can accept if for the "gui" interface).

In this case it removes from the user the build date.

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

I personally dislike "common" variables, as they hid bugs a few times
for me. Thus, ppsi has "-fno-common" by now.  This commit adds
extern to memlayout.h and creates the variables in dev/sdb.c .
This builds whether or not -fno-common is there and has no effect
on a ptp-noposix build.

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

Alessandro Rubini authored 12 years ago and Grzegorz Daniluk committed 12 years ago

the pp_printf_defconfig was the trial version for pp_pritnf as an
alternative to mprintf, so it is not needed any more.  All other
defconfig were working already, but I'd better regenerate them with
the current defaults.

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

Alessandro Rubini authored 12 years ago and Grzegorz Daniluk committed 12 years ago

mprintf is of uncertain copyright status, and that's why
we added pp_printf (actually, that's why I picked pp_printf
fro ptp-proposal and made it a standalone thing).

Now that everyone involved confirms pp_printf is well tested,
let's remove mprintf.  We still accept mprintf in the callers,
because the name is used by our ptp-noposix submodule, and I'm
sure people has it in their fingers and it will appear again.

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

Alessandro Rubini authored 12 years ago and Grzegorz Daniluk committed 12 years ago

The ppsi submodule (which is being cleaned up) now has a separate
structure for whiterabbit-specific data associated to the port. This
commit includes such change and adapts its own structures accordingly.

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

Alessandro Rubini authored 12 years ago and Grzegorz Daniluk committed 12 years ago

The ppsi subproject now builds pp_printf by default, so the Makefile here
passes ~CONFIG_NO_PRINTF=y". Also, assign_TimeInternal() is not there
any more because it is just an assignment.

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

The minic driver reads the TX timestamp from the device after sending.
Unfortunately, it only waited for the data to be enqueued to be sent.
Sometimes this lead to a race condition where the old TX stamp was read.

As a result, the PTP calculation goes wrong by as much as an entire second.
When in track_phase, this manifests as a bad phase offset sent to softpll.
The softpll takes years to reach that phase offset, so it is always busy.
Thus the clock drifts further and further away, never correcting.

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

Aurelio Colosimo authored 12 years ago and Alessandro Rubini committed 12 years ago

Signed-off-by: Aurelio Colosimo <aurelio@aureliocolosimo.it>

Aurelio Colosimo authored 12 years ago and Alessandro Rubini committed 12 years ago

Signed-off-by: Aurelio Colosimo <aurelio@aureliocolosimo.it>

Aurelio Colosimo authored 12 years ago and Alessandro Rubini committed 12 years ago

Signed-off-by: Aurelio Colosimo <aurelio@aureliocolosimo.it>