Skip to content

K
Kernel

openeuler / Kernel
1 年多 前同步成功

通知 8
Star 0
Fork 0
K
Kernel
切换分支/标签
  1. 28 8月, 2013 20 次提交
  3. 27 8月, 2013 7 次提交
  5. 26 8月, 2013 13 次提交
    • I
      staging: comedi: comedi_bond: use correct minor device numbers in name · 7322b320
      由 Ian Abbott 提交于 
      The board name for "comedi_bond" is constructed from a space-separated
list of items of the form "minor:subdevice" where "minor" is a minor
device number and "subdevice" is a subdevice number.  Currently, all the
"minor" device numbers are for the "comedi_bond" device itself and the
"subdevice" numbers are for the bonded devices.  It makes makes more
sense for the "minor" device numbers to come from the bonded devices as
well so that the string is a list of bonded "minor:subdevice" pairs.
Fix it.
Signed-off-by: NIan Abbott <abbotti@mev.co.uk>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      7322b320
    • I
      staging: comedi: comedi_bond: use krealloc() and fix memory leak · a55de0f4
      由 Ian Abbott 提交于 
      `do_dev_config()` is called from the comedi 'attach' handler,
`bonding_attach()`.  The device private data structure contains a
dynamically allocated array of pointers to "bonded" device structures
which grows during the `do_dev_config()` call.  The length of this array
is in `devpriv->ndevs`.  It currently uses a local function `realloc()`
to allocate a new array, copy the old contents over and free the old
array.  It should be more efficient to use `krealloc()` as it may be
able to use slack space at the end of the previous array and avoid a
copy.

The old `realloc()` function always freed the old buffer which meant
that if it failed to allocate the new buffer it would lose the contents
of the old buffer.  Unfortunately, that contained pointers to more
dynamically allocated memory, leading to a memory leak.  If `krealloc()`
fails, keep the old buffer and avoid the memory leak.  The
aforementioned pointers to more dynamically allocated memory will be
cleaned up by the 'detach' handler, `bonding_detach()` which will be
called by the comedi core as a consequence of `krealloc()` failing in
`do_dev_config()`.
Signed-off-by: NIan Abbott <abbotti@mev.co.uk>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      a55de0f4
    • I
      staging: comedi: comedi_bond: handle base channel for insn_bits · 0f3ce1a6
      由 Ian Abbott 提交于 
      If a DIO subdevice has more than 32 channels, its 'insn_bits' handler is
supposed to take account of the base channel from
`CR_CHAN(insn->chanspec)`.  (The comedi core will adjust the base
channel to 0 and shift the mask and data to compensate if the subdevice
has less than or equal to 32 channels.)  The "comedi_bond" driver
currently ignores the base channel and assumes it is 0.

Replace `comedi_dio_bitfield()` in the "kcomedilib" module with
`comedi_dio_bitfield2()` that takes account of the base channel, and
rewrite the "comedi_bond" driver's 'insn_bits' handler
(`bonding_dio_insn_bits()`) to take account of the base channel and use
the new function.

No other modules use `comedi_dio_bitfield()` so it is safe to replace it
with `comedi_dio_bitfield2()`.  The name follows that of the equivalent
function in the user-space comedilib.  If the base channel is non-zero
and the subdevice has less than or equal to 32 channels it needs to
adjust things in the same way as the comedi core (same as `parse_insn()`
in "comedi_fops.c") due to most drivers ignoring the base channel.
Signed-off-by: NIan Abbott <abbotti@mev.co.uk>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      0f3ce1a6
    • I
      staging: comedi: comedi_bond: get INSN_CONFIG_DIO_QUERY info from horse's mouth · 16d2d3cb
      由 Ian Abbott 提交于 
      The DIO subdevice of the "comedi_bond" device attempts to remember the
directions of DIO channels itself in the `io_bits` member of the
subdevice, but that is only large enough for the first 32 channels and
it might not be accurate anyway as changing the direction of one channel
may have affected a whole group of channels and we have no idea of the
initial directions before the "bonded" device was linked to the the
"comedi_bond" device.  It would be better to ask the bonded device for
this information when handling a `INSN_CONFIG_DIO_QUERY` configuration
instruction.  Add new function `comedi_dio_get_config()` to the
"kcomedilib" module to allow us to get the DIO direction of a channel
and use it.
Signed-off-by: NIan Abbott <abbotti@mev.co.uk>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      16d2d3cb
    • I
      staging: comedi: comedi_bond: just check devprivs->devs once on detach · 7b8cbe92
      由 Ian Abbott 提交于 
      The `while` loop in `bonding_detach()` doesn't need to check
`devpriv->devs` each time round the loop.  Move the test outside the
loop.  The enclosing `if (devpriv)` can be changed to `if (devpriv &&
devpriv->devs)` as everything in this `if` statement is associated with
`devpriv->devs` anyway.
Signed-off-by: NIan Abbott <abbotti@mev.co.uk>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      7b8cbe92
    • I
      staging: comedi: comedi_bond: no need to free dev->private on detach · c2af5b96
      由 Ian Abbott 提交于 
      The comedi core will free `dev->private` if it is non-NULL after calling
the "detach" handler (`bonding_detach()`), so don't bother freeing it in
`bonding_detach()`.
Signed-off-by: NIan Abbott <abbotti@mev.co.uk>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      c2af5b96
    • I
      staging: comedi: comedi_bond: remove unused subdev_type · ad9f81f0
      由 Ian Abbott 提交于 
      The `subdev_type` member of `struct bonded_device` is set but not used.
Remove it.
Signed-off-by: NIan Abbott <abbotti@mev.co.uk>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      ad9f81f0
    • I
      staging: comedi: comedi_bond: don't map channels individually · 10b799d2
      由 Ian Abbott 提交于 
      The private data structure (`struct comedi_bond_private`) for the
overall "comedi_bond" device maps each channel individually to a pointer
to the `struct bonded_device` it belongs to via array member
`chan_id_dev_map[MAX_CHANS]`. This speeds up look-ups from channel
number to bonded device a bit, but the length of the array used to look
this up is currently fixed at `MAX_CHANS` (256) and there are no
overflow checks when filling the array.

In practice, there will only be a few bonded devices (actually bonded
subdevices) and it is practical to just skip through the list until we
reach the one containing the desired channel.

The only place where the bonded device is looked up from the channel
number is in `bonding_dio_insn_config()`.  Change it to do the look-up
by skipping through the list of bonded devices and remove the
`chan_id_dev_map[]` member.  The `chanid_offset` member of `struct
bonded_device` is also no longer needed as the value can be derived
while skipping through the list of bonded devices, so remove that member
as well.
Signed-off-by: NIan Abbott <abbotti@mev.co.uk>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      10b799d2
    • I
      staging: comedi: comedi_bond: use bitmap to record opened/closed minors · f59ebeda
      由 Ian Abbott 提交于 
      `do_dev_config()` currently records the comedi minor devices it has
opened by setting `devs_opened[minor]` to the pointer returned by
`comedi_open()`.  This is checked to avoid opening the same minor device
twice.  The pointer values in `devs_opened[]` aren't needed; we only
need to record which minor device numbers are being used.  Change
`devs_opened` to a bitmap (declared with `DECLARE_BITMAP()`) of length
`COMEDI_NUM_BOARD_MINORS` as the minor device numbers are range-checked
to fit in a bitmap of this length.  Use `test_and_set_bit()` to record
the minor device numbers we attempt to open with `comedi_open()`.

`bonding_detach()` calls `comedi_close()` to close the comedi minor
devices.  Since the minor device numbers may be repeated in its list of
bonded subdevices, it currently uses a simple `unsigned long
devs_closed` variable as a bitmap to keep track of which minor device
numbers it has already closed to avoid closing them twice.  As a single
`unsigned long` consists of less than `COMEDI_NUM_BOARD_MINORS` bits on
a 32-bit machine, change `devs_closed to a bitmap of this length using
`DECLARE_BITMAP()` and use `test_and_set_bit()` to avoid calling
`comedi_close()` more than once for each minor device number in use.
Signed-off-by: NIan Abbott <abbotti@mev.co.uk>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      f59ebeda
    • I
      staging: comedi: comedi_bond: change return value of bonding_attach() · 8fe73691
      由 Ian Abbott 提交于 
      `bonding_attach()` is the comedi "attach" handler for the driver.  Any
non-negative return value is treated as successful, but 0 is the
preferred return value on success.
Signed-off-by: NIan Abbott <abbotti@mev.co.uk>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      8fe73691
    • I
      staging: comedi: comedi_bond: return error code in do_dev_config() · fcf26cff
      由 Ian Abbott 提交于 
      Change `do_dev_config()` to return an error code on failure and 0 on
success, instead of 0 on failure and 1 on success.
Signed-off-by: NIan Abbott <abbotti@mev.co.uk>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      fcf26cff
    • I
      staging: comedi: comedi_bond: no need to initialize file[] · a7e240a4
      由 Ian Abbott 提交于 
      The `char file[]` variable in `do_dev_config()` doesn't need to be
initialized as it gets overwritten with a `snprintf()`.  It just needs
to be long enough.
Signed-off-by: NIan Abbott <abbotti@mev.co.uk>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      a7e240a4
    • I
      staging: comedi: comedi_bond: rename CamelCase identifiers · c2b03cec
      由 Ian Abbott 提交于 
      Signed-off-by: NIan Abbott <abbotti@mev.co.uk>
Signed-off-by: NGreg Kroah-Hartman <gregkh@linuxfoundation.org>
      c2b03cec