lkml.org 
[lkml]   [2010]   [Jun]   [28]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: [PATCHv2 1/3] mfd: add STMPE I/O Expander support
    Hi Samuel,

    On Mon, Jun 28, 2010 at 01:55:16 +0200, Samuel Ortiz wrote:
    > On Tue, Jun 22, 2010 at 07:25:27PM +0530, Rabin Vincent wrote:
    > > +int stmpe_reg_read(struct stmpe *stmpe, u8 reg)
    > > +{
    > > + int ret;
    > > +
    > > + ret = i2c_smbus_read_byte_data(stmpe->i2c, reg);
    > > + if (ret < 0)
    > > + dev_err(stmpe->dev, "failed to read reg %#x: %d\n",
    > > + reg, ret);
    > > +
    > > + dev_vdbg(stmpe->dev, "rd: reg %#x => data %#x\n", reg, ret);
    > > +
    > > + return ret;
    > > +}
    > > +EXPORT_SYMBOL(stmpe_reg_read);
    > I think your locking is broken here.
    > If your exporting this routine (and the next ones below), you'd better make
    > sure you're under stmpe->lock for the stmpe register concurrent access.

    stmpe_reg_read() and stmpe_reg_write() are just a call to one
    i2c_smbus_* function, and the I2C core takes a bus_lock internally
    preventing concurrent accesses.

    The only place where the I2C core locking is not sufficient is the
    read/modify/write sequence, and we provide stmpe_set_bits() for that,
    which takes a lock. If someone uses reg_read()/reg_write() sequences on
    registers where they should be using set_bits(), adding extra locking in
    reg_read()/reg_write() will not provide any additional safeguard.

    The same scheme is used by adp5520.

    Could you please explain why more locking is needed?

    > What I suggest is that you'd have the exported routines taking your stmpe
    > lock, and then have an internal version (e.g. named with a __stmpe prefix)
    > without lock taken for your core code. In your case, you could probably call
    > the i2c I/O routines directly, that's up to you.
    >
    > > +/**
    > > + * stmpe_set_bits() - set the value of a bitfield in a stmpe register
    > > + * @stmpe: device to write to
    > > + * @reg: register to write
    > > + * @mask: mask of bits to set
    > > + * @val: value to set
    > > + */
    > > +int stmpe_set_bits(struct stmpe *stmpe, u8 reg, u8 mask, u8 val)
    > > +{
    > > + int ret;
    > > +
    > > + mutex_lock(&stmpe->lock);
    > > +
    > > + ret = stmpe_reg_read(stmpe, reg);
    > That one for example would be __stmpe_read().
    >
    > > +/**
    > > + * stmpe_block_write() - write multiple stmpe registers
    > > + * @stmpe: device to write to
    > > + * @reg: first register
    > > + * @length: number of registers
    > > + * @values: values to write
    > > + */
    > > +int stmpe_block_write(struct stmpe *stmpe, u8 reg, u8 length,
    > > + const u8 *values)
    > > +{
    > > + int ret;
    > > +
    > > + dev_vdbg(stmpe->dev, "wr: regs %#x (%d)\n", reg, length);
    > > +#ifdef VERBOSE_DEBUG
    > > + print_hex_dump_bytes("stmpe wr: ", dump_prefix_offset, values, length);
    > > +#endif
    > I don't really enjoy this part for 2 reasons:
    > - You should use a less generic ifdef switch, prefixed with STMPE_ for
    > example.

    The dev_vdbg() in the previous line is activated via VERBOSE_DEBUG, so
    the idea was to have this dump use the same config. I'll fix it as your
    recommended, though. Will fix your other comments too.

    Rabin


    \
     
     \ /
      Last update: 2010-06-29 08:53    [W:0.040 / U:2.188 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site