On 04/03/2016 11:52 AM, Peter Rosin wrote:
> From: Peter Rosin <p...@axentia.se>
>
> Allocate an explicit i2c mux core to handle parent and child adapters
> etc. Update the select/deselect ops to be in terms of the i2c mux core
> instead of the child adapter.
>
> --- a/drivers/iio/imu/inv_mpu6050/inv_mpu_acpi.c
> +++ b/drivers/iio/imu/inv_mpu6050/inv_mpu_acpi.c
> @@ -136,16 +133,15 @@ static int inv_mpu_probe(struct i2c_client *client,
> return result;
>
> st = iio_priv(dev_get_drvdata(&client->dev));
> - st->mux_adapter = i2c_add_mux_adapter(client->adapter,
> - &client->dev,
> - client,
> - 0, 0, 0,
> - inv_mpu6050_select_bypass,
> - inv_mpu6050_deselect_bypass);
> - if (!st->mux_adapter) {
> - result = -ENODEV;
> + st->muxc = i2c_mux_one_adapter(client->adapter, &client->dev, 0, 0,
> + 0, 0, 0,
> + inv_mpu6050_select_bypass,
> + inv_mpu6050_deselect_bypass);
> + if (IS_ERR(st->muxc)) {
> + result = PTR_ERR(st->muxc);
> goto out_unreg_device;
> }
> + st->muxc->priv = client;
I just tested this patch on mpu9150 (combo mpu6050+ak8975) and I got a
crash on probe which looks sort of like this:
[ 5.565374] RIP: 0010:[<ffffffff81481aed>] [<ffffffff81481aed>]
mutex_lock+0xd/0x30
[ 5.565374] Call Trace:
[ 5.565374] [<ffffffff813be34c>] inv_mpu6050_select_bypass+0x1c/0xa0
...
[ 5.565374] [<ffffffff81392141>] i2c_transfer+0x51/0x90
...
[ 5.565374] [<ffffffff81392cb5>] i2c_smbus_read_i2c_block_data+0x45/0xd0
[ 5.565374] [<ffffffff813bec5a>] ak8975_probe+0x14a/0x5c0
...
[ 5.565374] [<ffffffff813933d8>] i2c_new_device+0x178/0x1e0
[ 5.565374] [<ffffffff8139362d>] of_i2c_register_device+0xdd/0x1a0
[ 5.565374] [<ffffffff8139372b>] of_i2c_register_devices+0x3b/0x60
[ 5.565374] [<ffffffff81393e88>] i2c_register_adapter+0x178/0x410
[ 5.565374] [<ffffffff81394203>] i2c_add_adapter+0x73/0x90
[ 5.565374] [<ffffffff81395f3d>] i2c_mux_add_adapter+0x2ed/0x400
[ 5.565374] [<ffffffff81396091>] i2c_mux_one_adapter+0x41/0x70
[ 5.565374] [<ffffffff813be48a>] inv_mpu_probe+0xba/0x1a0
This happens because muxc->priv is not initialized when probing devices
behind the mux as described by devicetree. This can be easily fixed by
using muxc->dev instead of muxc->priv, or perhaps by calling
i2c_mux_alloc, initializing priv and later doing i2c_mux_add_adapter.
These fixes are driver-specific and other drivers might experience
similar issues. Perhaps i2c_mux_one_adapter should take void *priv just
like old i2c_add_mux_adapter?
After I fix this locally the deadlock when reading from both devices no
longer happens.
--
Regards,
Leonard
