Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

This branche adds POLLERR support (see commit messages).
Federico gave his acked-by, but the original commit is already in
production, so I am not changing the commit messages.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>
Acked-by: Federico Vaga <federico.vaga@gmail.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>
Acked-by: Federico Vaga <federico.vaga@gmail.com

For traceability purposes, we include commit descriptions
as MODULE_VERSION information in drivers that are deployed
in CERN control system.

Ideally, the commit description should be added to all the
binary blobs installed, but the kernel module is critical
in this connection, as it is crucial to trace back quickly
the whereabouts of installed .ko sources.
Signed-off-by: Juan David Gonzalez Cobas <dcobas@cern.ch>
Acked-by: Alessandro Rubini <rubini@gnudd.com>
Acked-by: Federico Vaga <federico.vaga@gmail.com>

Unfortunately, we sometimes have threads that block in the I/O
operation (the use case is input), and need to be notified when
another thread aborts the scheduled operation.  This sometimes applies
to a user-requested reconfiguration of the trigger.

This commit awakes processes sleeping on a cset when the trigger is
disabled by a sysfs store, and the chardev interface returns POLLERR
in the poll/select functions for control devices.

We chose not to affect the data device nor read and write. This means
all simple code (scripts, simple demos etc) is not affected.  But
full-featured applications, which rely on poll, have this additional
notification if they check the POLLERR bit.

Note: select(2) reports files both readable and writable when the drive
returns POLLERR, so applications should really use poll(2) instead.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Two different functions use the "can_read" and "can_write" names in
different ways: integer flags or function pointers. This is confusing.
So now flags are called "ret_ok", leavning can_read as the name of
a function pointer and nothing else.
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>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

The commits allows the vmalloc buffer to change the allocated size.
If mmap is active, it returns EBUSY, but we need to protect
against new maps to appear during the change. The vm_ops->open
method returns void, so we need to block the mmap() system call
directly, in chardev.c.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>
Acked-by: Federico Vaga <federico.vaga@gmail.com>

This commit adds a new trigger operation: arm.  The operation is
useful because it allows a trigger to define its own procedure to arm
itself. It is used by the fmc-adc-100Ms-14bit-4cha driver.

At the moment, nobody use the trigger operation data_done() and abort()
because the generic one work in most of the case, and tipically there
is nothing special to do if the arming procedure is always the same.
This patch make sense of these operations because if you need a custom
arm(), probably, you also need a custom data_done() and abort().
Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

Zero is the testing tool for zio features, and a show case for driver
writers (the "simple" one is zio-mini). So, now that the core supports
interleave, zio-zero should do as well.
Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

If a cset->flags has ZIO_CSET_INTERLEAVE_ONLY or
ZIO_CSET_CHAN_INTERLEAVE, then ZIO creates an additional channel in
the cset: the interleaved channel.  As for normal channels, the
interleaved channel has 2 char device:

/dev/zio/devname-0-i-ctrl
/dev/zio/devname-0-i-data

If cset flags include ZIO_CSET_INTERLEAVE_ONLY, ZIO creates only
interleaved channel char devices. From the sysfs you can
enable/disable only the interleaved channel; all the other channels
are always disabled.

If cset flags include ZIO_CSET_CHAN_INTERLEAVE, ZIO creates all char
devices for normal channels and also for the interleaved channel. By
default the interleaved channel is disabled. You can enable the
interleaved channel from sysfs; this choice automatically disable all
the other channels.

Interleave support is needed for the fmc-adc-100ms-14b-4cha driver.
Maybe this is not the final approach to interleave, but it is
working and well-tested, so the time is ripe to have it in master.

Maybe we'll change some details about interleave support, in the
future.
Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

This is a long-standing bug, when more than one process access
concurrently the char devices. This adds proper locking and solved the
problem.

Also, a new policy for writing control is in place: if you write
control, any partial block (unless empty) is pushed to the pipeline.
This matches the idea that after write(2) accepts data, it cannot be lost.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>
Acked-by: Federico Vaga <federico.vaga@gmail.com>

[output]
If the user fills the buffer with many blocks, trigger user must
be able to re-arm itself after data_done to immediately output all
blocks in the buffer.
Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

The data_done trigger operation returns an integer which mean if the
trigger must be automatically re-armed (1) or not (0). The rearm value
it is returned by zio_trigger_data_done to notify if the trigger was
rearmed or not.
Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubinig@gnudd.com>

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

trigger and buffer creation

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

The function zio_device_find_child() internally use the Linux kernel
function device_find_child() with increment the refcount of the child
device.
Once we have done our operation on the child device we must use
put_device() to decrese the refcount.
Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

The bug: if a device probe() fails, ZIO register the real ZIO device
anyway
The issue: on device probe() channel-set, trigger, channels and buffer
are not ready

The patch: do probe after the _complete_ registration of the real
device. In this way, it is guaranteed that all components are ready.
More over, if the probe fail, the device can be easly removed.
Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

At the moment, no known driver use these operations so we can "safely"
remove them from the framework.
We decide to remove init() and exit() because, in the current
implementation, they introduce more problems than they solve.

If you used these functions, we suggest you to move your init()
functions in the probe() function and your exit() functions before
deregister the device.
Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

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

Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

Raise a build error if the last attribute's enumarator (e.g.
_ZIO_BUF_ATTR_STD_NUM) is greater than the maximum numer of standard
attributes ZIO_MAX_STD_ATTR
Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>

The 'version' attribute is a standard attribute. It is placed at
the last position of the standard attribute array. In this way, it is
easier to identify it in an hexdump stream because it is the last
standard attribute. At the moment the last index of standard attributes
array is 15, but in a far away future may change.

The aim of this attribute is to identify the version of an attribute set
of a ZIO object (device, trigger, buffer). During development the set
of attribute exported by a driver may change. This attribute must be
used to notify these changes to user-space programs.
The attribute is optional.

When used in a device, we suggest to place it at device level.
Signed-off-by: Federico Vaga <federico.vaga@gmail.com>
Acked-by: Alessandro Rubini <rubini@gnudd.com>