Release invalid blocks on DMA error
Signed-off-by: Federico Vaga <federico.vaga@gmail.com>

It is fair and necessary for peaceful coexistence of the driver
with other peers like FD and TDC.

This avoids page-split dma_items in a very dirty way

Debugging DMA error required to spot what next dma_item
adresses were being communicated to the DMA core; the
trace message has been deferred accordingly

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 set of commits completely reviews and simplifies the API, fixing
the sequence of allocation/fill/deallocation for buffers.

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

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>

Signed-off-by: Federico Vaga <federico.vaga@gmail.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: Federico Vaga <federico.vaga@gmail.com>

At open time, users are expected to pass the total number of samples
they expect to need in their buffer, not the total number of bytes.
The name now matches current use in fald-simple-acq.
Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

We need mapaddr and mapsize for munmap() at buffer_release time.
And we need the device pointer (not drivername) in order to route
the get_tstamp method. drivername, if needed, is in the device anyways.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

The prototypes for buffer management so far were wrong.

request_buffer() (which up to now both allocates and fille) is going
to only allocate the buffer, so it doesn't need a timeout.

fill_buffer() on the other hand can have a timeout -- although if
you used ack_poll() or acq_start() with no timetout, the filling is
expected to complete immediately.

The current request_buffer(), which still fills is too,
now uses NULL as a timeout, which isn't an argument any more.
./tools/fald-simple-acq works unchanged, as it already passed NULL
as timeout.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This reverts commit b6b81e29.

I don't now why, yet, but now multi-shot acquisition acquires a
1-sample error at each shot. So either hw documentation is wrong, or
this patch is incomplete (for example, the hardware value is read back
and used -- this is most likely but I have no time to check now).

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

The file needs a number of code stanzas to be turned into data
structures, however this is delayed to after-1.0.

the commit changes some details only.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

Tipos and minor simplifications.

The only real change here is not stopping acquisition at close time.
I got mad with the auto-start not working because of this. I really think
the program should leave stuff configured, unless explicitly closing it.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This is a validation for the fmcadc_acq_poll() function.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This commit deals with the acq_start and acq_stop, and adds acq_poll.
It id mainly s simplifying step, removing unneeded checks and
splitting acq_poll out of acq_start (now start calls poll at the end,
as a side effect).

../libtools/fald-simple-acq works unaffected.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This commit makes the library consistent with the official design as
far as open and close are concerned. But buffer selection is still missing.

- open_by_lun() is not there in the low-level implementation, because
the function is expected to be library only, and is expected to fall
back to open, after finding the hardware-specific dev_if from the
administratively-set lun value.

- redundant error checking has been removed. For example, NULL names
are as bad as wild-pointer names, so the program should crash, not
receive an error and manage it. I used this policy in several places.

Other changes brought in by this:

- the board list is now static in boards.c

I'll still move something, at the end: open and close are zio-generic, not
board-specific.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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

This picks a master commit of zio, which now includes all the things
needed for this driver.  Since the data_done method now returns integer
(instead of void), the code is fixed in that respect, partly undoing
commit:

   18f2d110 kernel: remove all warnings
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This set of patches moves functions around, trying to split what is
library-wide from what is zio-wide (i.e., generic code to manage zio
drivers) and board-specific.

This set of patches has no technical change, nor even in the names of
the functions (though some that were static are not static any more).

Later commits in lib/ will change the semantics to match
documentation, rename function and add the missing interfaces. This
set of commits is meant to make later diffs shorter, and thus easier
to review, but I surely forgot something here.

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 commit makes not change at all, it just copies part of the
original unified header to a new file. And apps use it.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

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