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 branch adds WRS shared memory, and makes the HAL use it.
Clients (both locally with libwr and ppsi without libwr) now
access port status with shared memory.

We also have a shmem-dumping tool, that helped auditing the
data structure (but changes to them are not applied yet).

As a side effect libwr and the hal are passed throught Lindent,
but before any change, so you can still blame these changes.

Adam Wujek authored 10 years ago and Alessandro Rubini committed 10 years ago

merge load-fpga.c and load-virtex.c

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

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 moves the query_ports() and get_port_state() to be shared memory
lookups.

The hal_client.c file now accesses shared memory at init time, and
performes the two functions above by looking in shared memory.

The thing is sill suboptimal, because the "exported" hexp_port_state_t
structure is only slightly different from struct hal_port_state. And
most fields being looked up are never filled.

Last but not least, halexp_get_timing_state() is not ready to be
converted to shared memory at this point in time (timing data is not
stored to shmem yet).

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

This has no effect on the code functionality.

Even if it affects libwr, this is a hal-only change.  The functions
in libwr that the hal itself uses, now receive a pointer to ports[],
which is thus back a static pointer within wrsw_hal/hal_ports.c.

This change allows moving the RPC queries to be shared memory lookups,
by using the ports[] pointer in the client; so both the client and
the server can access the same shared mamory using libwr functions.

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

They are the usual 0 and -1: some functions returned the names, but
no caller was checking the names. Also, PORT_BUSY (which was 1) was
never used specially, so it is turned into a normal -1 error.

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

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

Even though the idea of using a single helper process for all
web queries is sound, this is not it. No more comments from 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>

Adam Wujek authored 10 years ago and Alessandro Rubini committed 10 years ago

Also fix makefiles and documentation

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

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

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

This allows other process to access port information without
making several RPC calls.

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

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

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

The stdout and stderr of the basic WR tools are piled to the "logger"
process, to get actual logs saved, if so configured for.

Thus, we should not reopen the output channels to /dev/null

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

This is the first step to move the "get_port_state" query to shared
memory.  The internal structures of hal_ports.c will be allocated
in shared memory, so the calls to halexp_* will just convert the data,
without context-switching to the hal process.

I couldn't add the functions to libwr/hal_client.c where they belong,
because the choice of using the same "halexp" names for local and remote
calls would lead to duplicate symbol definitions when linking the hal
(the hal process links libwr, but can't link hal_client.o).
Thus, libwr/hal_shmem.c is created, to host the parts of hal/hal_ports.c
that are needed to convert data.

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

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>
Acked-by: Maciej Lipinski <maciej.lipinski@cern.ch>

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

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

LOCK_STATE_* macros were defined locally and not used
(only LOCK_STATE_NONE was used, but 0 is better since the
relevant function says "returns 0 or 1").

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

HAL_PORT_MODE_* were defined locally and never used. The code
uses HEXP_PORT_MODE_* instead (defined in <hal/hal_exports.h> and
used in several places).

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

After this commit, all symbols in hal_ports are of the form hal_port_*
(all but one, that is not used and I'll remove it in another commit).
I'm not smart enough to track the origin of the various names, so I
need this.  I need this especially in this file because I'm going to
export stuff using shmem, to avoid a lot of context switches.

Thus, the "private" structure hal_port_state is not a typedef any
more, as a first step in becoming public. It doesn't make sense to
have two almost-identical structures, where one is copied to the other
one item at a time to deal with the minor differences. (It happened
for some reason: I'm not complaining hard, but it's time to get rid of
this).

Unfortunately, two functions were called "halexp", and relied on a
prototype shared by the RPC server (this file) and the RPC client.
While the convention made sense at first, it now shows its limits, so
the exported one is called hal_port_* like everything in the file, but
for this I need to add structure names (but I left the typedef at this
point, to avoid massive name changes around).

If I didn't introduce bugs, the code is the same as it was. I did it
one symbol at a time while build-checking each steps (and I even have
the commits to track any error).

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

pps_gen.c had a missing semicolon because of this bug. Fix the
macro to behave properly withing if/else. And fix the user, adding
the missing semicolon.

We all know that these two functions should be static inline, and the
fpga base pointer should be void *, but that's for later.

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

This header is almost obsolete, as we'll move to SDB soon.
Meanwhile, sort out the tab/space mess and indent with spaces only,
so it looks good to everybody (including greg who insists on one
tab every 2 spaces -- which works for indentation, but not for
for alignment at end-of-line).

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

We'd better use less CPU power to run the HAL, even if this shows down
response times for management queries (until we move most of it to
shmem).

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

The code was hardwired to /dev/ttyS2 and 115200 baud. So we are sure
nobody is using it, and it actually can't be used at all.

If needed, we know it's in the history.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>
Ackec-by: Grzegorz Daniluk <grzegorz.daniluk@cern.ch>

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