Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Sat, 7 Mar 2020 14:19:46 +0000
From:   Jonathan Cameron <jic23@kernel.org>
To:     Rohit Sarkar <rohitsarkar5398@gmail.com>
Cc:     linux-iio@vger.kernel.org, linux-kernel@vger.kernel.org,
        alexandru.ardelean@analog.com, dragos.bogdan@analog.com
Subject: Re: [PATCH v4] iio: adc: max1363: replace uses of mlock
Message-ID: <20200307141946.79661dd0@archlinux>
In-Reply-To: <5e6355a8.1c69fb81.36f2c.ab37@mx.google.com>
References: <5e6355a8.1c69fb81.36f2c.ab37@mx.google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sat, 7 Mar 2020 13:34:51 +0530
Rohit Sarkar <rohitsarkar5398@gmail.com> wrote:

> Replace usage indio_dev's mlock with either local lock or
> iio_device_claim_direct_mode.
> 
> Signed-off-by: Rohit Sarkar <rohitsarkar5398@gmail.com>

There is a subtlety in here (which is why this one never
got cleaned up before).  We need to protect against:

1) Driver state being accessed from multiple places concurrently.
   That will use your new lock.
2) Doing actions that cannot occur if in buffered mode.  The
   claim_direct_mode stuff is for that.
3) Prevent us enabling monitor mode when in buffered mode, or
   buffered mode when in monitor mode (because it's really
   fiddly to support both given the odd way this hardware
   works and I never bothered).  That requires making sure
   direct mode is claimed before trying to enable the monitor
   mode and also that we don't enable buffered mode. Now interestingly
   there doesn't actually seem to be anything stopping buffered mode
   if monitor mode is on.  Probably a bug, but I'd be nervous about
   fixing that without test hardware.

Ignoring that last subtle bit, we sometimes for 1 and 2 need to take
both locks.  We can't be in buffered mode and we need to edit
local state.

Jonathan

> ---

That's better :)  I should have looked at v4 before v3 I guess.

> Changelog v3 -> v4
> * Fix indentation
> 
> Changelog v2 -> v3
> * use iio_device_claim_direct when switching modes
> * replace mlock usage in max1363_write_event_config
> 
> Changelog v1 -> v2
> * Fix indentation
> 
>  drivers/iio/adc/max1363.c | 24 +++++++++++++++---------
>  1 file changed, 15 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/iio/adc/max1363.c b/drivers/iio/adc/max1363.c
> index 5c2cc61b666e..a1550c0b4c0a 100644
> --- a/drivers/iio/adc/max1363.c
> +++ b/drivers/iio/adc/max1363.c
> @@ -150,6 +150,7 @@ struct max1363_chip_info {
>   * @current_mode:	the scan mode of this chip
>   * @requestedmask:	a valid requested set of channels
>   * @reg:		supply regulator
> + * @lock		lock to ensure state is consistent
>   * @monitor_on:		whether monitor mode is enabled
>   * @monitor_speed:	parameter corresponding to device monitor speed setting
>   * @mask_high:		bitmask for enabled high thresholds
> @@ -169,6 +170,7 @@ struct max1363_state {
>  	const struct max1363_mode	*current_mode;
>  	u32				requestedmask;
>  	struct regulator		*reg;
> +	struct mutex			lock;
>  
>  	/* Using monitor modes and buffer at the same time is
>  	   currently not supported */
> @@ -364,7 +366,9 @@ static int max1363_read_single_chan(struct iio_dev *indio_dev,
>  	struct max1363_state *st = iio_priv(indio_dev);
>  	struct i2c_client *client = st->client;
>  
> -	mutex_lock(&indio_dev->mlock);
> +	ret = iio_device_claim_direct_mode(indio_dev);

So this protects against change of state from polled to buffered mode.
Great.  However, we also use state in here which should be protected
by the local lock.  The solution is to take that local lock as well.

> +	if (ret < 0)
> +		goto error_ret;
>  	/*
>  	 * If monitor mode is enabled, the method for reading a single
>  	 * channel will have to be rather different and has not yet
> @@ -372,7 +376,7 @@ static int max1363_read_single_chan(struct iio_dev *indio_dev,
>  	 *
>  	 * Also, cannot read directly if buffered capture enabled.
>  	 */
> -	if (st->monitor_on || iio_buffer_enabled(indio_dev)) {
> +	if (st->monitor_on) {
>  		ret = -EBUSY;
>  		goto error_ret;
>  	}
> @@ -404,8 +408,9 @@ static int max1363_read_single_chan(struct iio_dev *indio_dev,
>  		data = rxbuf[0];
>  	}
>  	*val = data;
> +
>  error_ret:
> -	mutex_unlock(&indio_dev->mlock);
> +	iio_device_release_direct_mode(indio_dev);
>  	return ret;
>  
>  }
> @@ -705,9 +710,9 @@ static ssize_t max1363_monitor_store_freq(struct device *dev,
>  	if (!found)
>  		return -EINVAL;
>  
> -	mutex_lock(&indio_dev->mlock);
> +	mutex_lock(&st->lock);
>  	st->monitor_speed = i;
> -	mutex_unlock(&indio_dev->mlock);
> +	mutex_unlock(&st->lock);
>  
>  	return 0;
>  }
> @@ -810,12 +815,12 @@ static int max1363_read_event_config(struct iio_dev *indio_dev,
>  	int val;
>  	int number = chan->channel;
>  
> -	mutex_lock(&indio_dev->mlock);
> +	mutex_lock(&st->lock);
>  	if (dir == IIO_EV_DIR_FALLING)
>  		val = (1 << number) & st->mask_low;
>  	else
>  		val = (1 << number) & st->mask_high;
> -	mutex_unlock(&indio_dev->mlock);
> +	mutex_unlock(&st->lock);
>  
>  	return val;
>  }
> @@ -962,7 +967,7 @@ static int max1363_write_event_config(struct iio_dev *indio_dev,
>  	u16 unifiedmask;
>  	int number = chan->channel;
>  
> -	mutex_lock(&indio_dev->mlock);
> +	iio_device_claim_direct_mode(indio_dev);

So the question is whether we are stopping this changing because
we are in buffered mode or because it's local state...

>  	unifiedmask = st->mask_low | st->mask_high;
>  	if (dir == IIO_EV_DIR_FALLING) {
>  
> @@ -989,7 +994,7 @@ static int max1363_write_event_config(struct iio_dev *indio_dev,
>  
>  	max1363_monitor_mode_update(st, !!(st->mask_high | st->mask_low));
>  error_ret:
> -	mutex_unlock(&indio_dev->mlock);
> +	iio_device_release_direct_mode(indio_dev);
>  
>  	return ret;
>  }
> @@ -1587,6 +1592,7 @@ static int max1363_probe(struct i2c_client *client,
>  
>  	st = iio_priv(indio_dev);
>  
> +	mutex_init(&st->lock);
>  	st->reg = devm_regulator_get(&client->dev, "vcc");
>  	if (IS_ERR(st->reg)) {
>  		ret = PTR_ERR(st->reg);