Re: [Outreachy kernel] Re: [PATCH] staging: iio: ade7753: replace mlock with driver private lock

[Jonathan Cameron]

On 19/03/17 13:16, Gargi Sharma wrote:
> On Sun, Mar 19, 2017 at 4:01 PM, Jonathan Cameron <jic23@xxxxxxxxxx> wrote:
>> On 17/03/17 09:32, Gargi Sharma wrote:
>>> On Mon, Mar 13, 2017 at 5:30 PM, Lars-Peter Clausen <lars@xxxxxxxxxx> wrote:
>>>>
>>>> On 03/12/2017 02:32 PM, simran singhal wrote:
>>>>> The IIO subsystem is redefining iio_dev->mlock to be used by
>>>>> the IIO core only for protecting device operating mode changes.
>>>>> ie. Changes between INDIO_DIRECT_MODE, INDIO_BUFFER_* modes.
>>>>>
>>>>> In this driver, mlock was being used to protect hardware state
>>>>> changes.  Replace it with a lock in the devices global data.
>>>>>
>>>>> Fix some coding style issues related to white space also.
>>>>>
>>>>> Signed-off-by: simran singhal <singhalsimran0@xxxxxxxxx>
>>>>> ---
>>>>>  drivers/staging/iio/meter/ade7753.c | 14 ++++++++------
>>>>>  1 file changed, 8 insertions(+), 6 deletions(-)
>>>>>
>>>>> diff --git a/drivers/staging/iio/meter/ade7753.c b/drivers/staging/iio/meter/ade7753.c
>>>>> index dfd8b71..ca99d82 100644
>>>>> --- a/drivers/staging/iio/meter/ade7753.c
>>>>> +++ b/drivers/staging/iio/meter/ade7753.c
>>>>> @@ -81,12 +81,14 @@
>>>>>   * @tx:         transmit buffer
>>>>>   * @rx:         receive buffer
>>>>>   * @buf_lock:       mutex to protect tx and rx
>>>>> + * @lock:    protect sensor state
>>>>
>>>> It might make sense to reuse the existing lock which currently protects the
>>>> read/write functions. You can do this by introducing a variant of
>>>> ade7753_spi_{read,write}_reg_16() that does not take a lock and use these to
>>>> implement the read-modify-write cycle in a protected section.
>>>
>>> There are other read/write functions for example,
>>> ade7753_spi_{read/write}_reg_8 that use the mutex as well. Should a
>>> variant of these functions be introduced as well? Also, how does one
>>> go about implementing RMW inside a protected section.
>> Hmm. Simran has also been progressing with patches for this.
>>
> I was trying to work through a patch for ade7754. So ran into the same
> problem :)
> 
>> You raise a good question. There are other read/modify/write sequences in
>> the driver.  They don't have the same issue with potentially deadlocking
>> against the buf lock as they are all using the spi subsystems provisions
>> for small write/read cycles where buffer protection is handled internally.
>>
>> So let us address the cases in turn:
>>
>> static int ade7753_reset(struct device *dev)
>> {
>>         u16 val;
>>         int ret;
>>
>>         ret = ade7753_spi_read_reg_16(dev, ADE7753_MODE, &val);
>>         if (ret)
>>                 return ret;
>>
>>         val |= BIT(6); /* Software Chip Reset */
>>
>>         return ade7753_spi_write_reg_16(dev, ADE7753_MODE, val);
>> }
>> This is only called in the device initialization.  At that point
>> we should be fine in assuming no parallel calls.  Crucial point
>> is it is before the call to iio_device_register which exposes
>> the userspace interfaces.
>>
>> static int ade7753_set_irq(struct device *dev, bool enable)
>> {
>>         int ret;
>>         u8 irqen;
>>
>>         ret = ade7753_spi_read_reg_8(dev, ADE7753_IRQEN, &irqen);
>>         if (ret)
>>                 goto error_ret;
>>
>>         if (enable)
>>                 irqen |= BIT(3); /* Enables an interrupt when a data is
>>                                   * present in the waveform register
>>                                   */
>>         else
>>                 irqen &= ~BIT(3);
>>
>>         ret = ade7753_spi_write_reg_8(dev, ADE7753_IRQEN, irqen);
>>
>> error_ret:
>>         return ret;
>> }
>>
>> This one is actually safe because it is the only function that
>> modifies that particular register.
>>
>> /* Power down the device */
>> static int ade7753_stop_device(struct device *dev)
>> {
>>         u16 val;
>>         int ret;
>>
>>         ret = ade7753_spi_read_reg_16(dev, ADE7753_MODE, &val);
>>         if (ret)
>>                 return ret;
>>
>>         val |= BIT(4);  /* AD converters can be turned off */
>>
>>         return ade7753_spi_write_reg_16(dev, ADE7753_MODE, val);
>> }
>>
>> Only called in remove (after userspace interfaces have been
>> removed by the iio_device_unregister call so also should not
>> be running concurrently with much else.
>>
> 
> The only nested lock here is ade7754_spi_write_reg_16, so as long as
> that is refactored, it'll be fine.
> 
>> So I think all the other cases are safe.  Perhaps it would have
>> been better to have had a lock around them, purely to make
>> the code more resilient against future changes though.
>> Probably a job to do as part of a larger scale pile of work
>> on that driver rather than as a one off patch.
> 
> Another question that I have is why are we writing inside a read
> function(ade7754_spi_read_reg_24)?
>
It's a register read (sort of) hence the reg in the name.
It's telling it which register to read by first writing that.
 
> static int ade7754_spi_read_reg_24(struct device *dev,
>                                    u8 reg_address, u32 *val)
> {
>         struct iio_dev *indio_dev = dev_to_iio_dev(dev);
>         struct ade7754_state *st = iio_priv(indio_dev);
>         int ret;
>         struct spi_transfer xfers[] = {
>                 {
>                         .tx_buf = st->tx,
>                         .rx_buf = st->rx,
>                         .bits_per_word = 8,
>                         .len = 4,
>                 },
>         };
> 
>         mutex_lock(&st->buf_lock);
>         st->tx[0] = ADE7754_READ_REG(reg_address);
>         st->tx[1] = 0;
>         st->tx[2] = 0;
>         st->tx[3] = 0;
> 
>         ret = spi_sync_transfer(st->us, xfers, ARRAY_SIZE(xfers));
>         if (ret) {
>                 dev_err(&st->us->dev, "problem when reading 24 bit
> register 0x%02X",
>                         reg_address);
>                 goto error_ret;
>         }
>         *val = (st->rx[1] << 16) | (st->rx[2] << 8) | st->rx[3];
> 
> error_ret:
>         mutex_unlock(&st->buf_lock);
>         return ret;
> }
> 
> Thanks!
> Gargi
> 
>>
>> Jonathan
>>
>>
>>
>>
>>
>>
>>
>>>
>>>
>>>>
>>>> Looking through the driver there seem to be other places as well that do
>>>> read-modify-write that should be protected by a lock, but currently are not.
>>>> This might be a good task.
>>>
>>> Am I right in understanding that we want to introduce mutex lock for
>>> writes in other drivers as well?
>>>
>>> Thanks,
>>> Gargi
>>>>
>>>>>   **/
>>>>>  struct ade7753_state {
>>>>> -         struct spi_device   *us;
>>>>> -                 struct mutex        buf_lock;
>>>>> -                         u8          tx[ADE7753_MAX_TX] ____cacheline_aligned;
>>>>> -                                 u8          rx[ADE7753_MAX_RX];
>>>>> +     struct spi_device   *us;
>>>>> +     struct mutex        buf_lock;
>>>>> +     struct mutex        lock;       /* protect sensor state */
>>>>> +     u8          tx[ADE7753_MAX_TX] ____cacheline_aligned;
>>>>> +     u8          rx[ADE7753_MAX_RX];
>>>>>  };
>>>>>
>>>>>  static int ade7753_spi_write_reg_8(struct device *dev,
>>>>> @@ -484,7 +486,7 @@ static ssize_t ade7753_write_frequency(struct device *dev,
>>>>>       if (!val)
>>>>>               return -EINVAL;
>>>>>
>>>>> -     mutex_lock(&indio_dev->mlock);
>>>>> +     mutex_lock(&st->lock);
>>>>>
>>>>>       t = 27900 / val;
>>>>>       if (t > 0)
>>>>> @@ -505,7 +507,7 @@ static ssize_t ade7753_write_frequency(struct device *dev,
>>>>>       ret = ade7753_spi_write_reg_16(dev, ADE7753_MODE, reg);
>>>>>
>>>>>  out:
>>>>> -     mutex_unlock(&indio_dev->mlock);
>>>>> +     mutex_unlock(&st->lock);
>>>>>
>>>>>       return ret ? ret : len;
>>>>>  }
>>>>>
>>>>
>>>> --
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-iio" in
>>> the body of a message to majordomo@xxxxxxxxxxxxxxx
>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>
>>