Federico Vaga authored 12 years ago and Alessandro Rubini committed 12 years ago

This patch introduces ZIO_RW_PERM and ZIO_RO_PERM, which are used in
all attributes. Prevoously we had write-all for writeable attributes,
now we only allos the owner and the group.

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

The merge-data attribute tells the buffer to try a merge of a
newly-stored block with the previous one. When this is possible, the
buffer updates the previous block (nsamples, datalen) and releases the
new control immediately. This allows saving some memory without
breaking the ZIO data model. The feature only makes sense for input,
but there is no policy in place to enforce that.

When user space reads this new block, it will get attributes (and
timestamp) for the first data items, but all available items are
returned in this bigger block.

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

This is an incompatible change for buffer authors. We don't think
there are any out-of-tree buffers, though.  The allocation
function must allocate the control by itself, instead of
receiving it from outside.  The reason why the control was passed
doesn't hold any more, and the upcoming PF_ZIO buffer will
greatly benefit from this.

Also, now alloc_block returns NULL on error, so it can be used directly
by the caller (it used to be ERR_PTR).

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>

Just like drivers should update current_ctrl and not block->ctrl,
the same happens to buffers. This only applies to vmalloc, though.

Bug introduced by my commit 6a7378a5

:
      zio-sys: delay copying current_ctrl to block->ctrl

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

This patch removes the union field priv in the zio_attribute and it
introduces the id field. From our experience, the priv.ptr pointer
is useless; the priv.addr field can be used for every purpose, also
if you have your own private structure.

The new field id replaces the role of priv.addr. It can represent a
register address, a register offset, an index of a vector of private
structure which describe how to gain access to a register, or whatever
can describe uniquely a register (or register field)

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>

If the user changes the buffer size, this might make more space for a
sleeping writer. Similarly, the buffer may be fuller than the new
value after a change, so we check for ">= limit" and not "== limit".

This doesn't apply to the vmalloc buffer, that currently doesn't allow
a size change.

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

buffer->store_block calls trigger->push_block when it receives the
first block. The trigger user fires directly , and if the device
accepts the data immediately, our data_done() will retr_block from the
buffer.  This happens with zio-zero, cset 1, and leads to a deadlock.

To allow immediate success of push_block down to device level, release
the lock before calling push_block. The code here also handles
pathologic situations like two processes storing at the same time.

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

Now all the zio objects are handled with the device structure

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

This buffer uses vmalloc and allows mmap from user space.  It started
as a copy of buf-kmalloc, so you can diff to change what is needed
to support mmap over a read/write buffer type.

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>

The vmalloc buffer uses the same names for structure (to ease "diff"
users in understanding it) so the slab name can't be the structure name.

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

We will have buffers that use the default operations (the majority) and
buffers without file operations for in-kernel activities, like the
input subsystem. In the latter case we want the cdev to refuse being
openend, so NULL is reserved for that case.

Also, add v_op to the buffer, so it can support mmap.

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

The argument is a relic of original design and was used incorrectly.

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

This saves another 0.1 usec for each block lifetime.

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

Some of them don't apply any more (we made our choices, or the problem
has been addressed and fixed elsewhere). Some do but I reworded them.
Some still apply unchanged, and are not touched by this patch.

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>

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>

Federico Vaga authored 13 years ago and Alessandro Rubini committed 13 years ago

A zio object can be enabled or disabled

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

This is a general change that affects triggers and buffers as well.
It introduces incompatibilities but the old name was actually a bug.

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

The buffer was calling trig->pull_block even for output transfers.
Bug exposed by zio-trig-user.c, yet to be committed.

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

Manohar Vanga authored 13 years ago and Alessandro Rubini committed 13 years ago

Signed-off-by: Manohar Vanga <manohar.vanga@cern.ch>

Federico Vaga authored 13 years ago and Alessandro Rubini committed 13 years ago

The standard buffer attribute 'max-buffer-len' is now supported
by zio-buf-kmalloc

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

This was missing, but is part of the suggested procedure

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>

A trigger instance has flags too, and the create functions
set the flag in bot bi and ti. And channels inherit from cset.

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>

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

Federico Vaga authored 13 years ago and Alessandro Rubini committed 13 years ago

From this commit onwards the whole thing compiles and works (from
2.6.34 onwards). We chose to save you from the past history of
internal zio development.

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

This is the default buffer: it manages a list of data blocks
(the lenght is fixed at 16 blocks by now) for any device and
any trigger.  The actual file operations in this case are
generic, and live in zio-buf.c (zio-core.ko).

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