[v5,3/4] i2c: core: Allow drivers to disable i2c-core irq mapping

By default the i2c-core will try to get an irq with index 0 on ACPI / of
instantiated devices. This is troublesome on some ACPI systems where the
irq info at index 0 in the CRS table may contain nonsense and/or point
to an irqchip for which there is no Linux driver.

If this happens then before this commit the driver's probe method would
never get called because i2c_device_probe will try to get an irq by
calling acpi_dev_gpio_irq_get which will always return -EPROBE in this
case, as it waits for a matching irqchip driver to load. Thus causing
the driver to not get a chance to bind.

This commit adds a new disable_i2c_core_irq_mapping flag to struct
i2c_driver which a driver can set to tell the core to skip irq mapping.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
---
Changes in v2:
-Actually also use the irq_index for of interrupts
Changes in v3:
-Add kernel doc for new i2c_driver irq_index member
-Remove duplicate assignment of driver in i2c_device_probe
Changes in v4:
-Add a disable_i2c_core_irq_mapping flag to i2c_driver instead of
 an irq_index member (effectively a rewrite of the patch, dropped
 the Reviewed-by-s)
---
 drivers/i2c/i2c-core.c | 6 +++---
 include/linux/i2c.h    | 3 +++
 2 files changed, 6 insertions(+), 3 deletions(-)

On Wed, 2017-04-05 at 00:03 +0200, Hans de Goede wrote:
> By default the i2c-core will try to get an irq with index 0 on ACPI /
> of
> instantiated devices. This is troublesome on some ACPI systems where
> the
> irq info at index 0 in the CRS table may contain nonsense and/or point
> to an irqchip for which there is no Linux driver.
> 
> If this happens then before this commit the driver's probe method
> would
> never get called because i2c_device_probe will try to get an irq by
> calling acpi_dev_gpio_irq_get which will always return -EPROBE in this
> case, as it waits for a matching irqchip driver to load. Thus causing
> the driver to not get a chance to bind.
> 
> This commit adds a new disable_i2c_core_irq_mapping flag to struct
> i2c_driver which a driver can set to tell the core to skip irq
> mapping.

FWIW:
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>

> 
> Signed-off-by: Hans de Goede <hdegoede@redhat.com>
> ---
> Changes in v2:
> -Actually also use the irq_index for of interrupts
> Changes in v3:
> -Add kernel doc for new i2c_driver irq_index member
> -Remove duplicate assignment of driver in i2c_device_probe
> Changes in v4:
> -Add a disable_i2c_core_irq_mapping flag to i2c_driver instead of
>  an irq_index member (effectively a rewrite of the patch, dropped
>  the Reviewed-by-s)
> ---
>  drivers/i2c/i2c-core.c | 6 +++---
>  include/linux/i2c.h    | 3 +++
>  2 files changed, 6 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/i2c/i2c-core.c b/drivers/i2c/i2c-core.c
> index 00c4cef..7a065c4 100644
> --- a/drivers/i2c/i2c-core.c
> +++ b/drivers/i2c/i2c-core.c
> @@ -985,7 +985,9 @@ static int i2c_device_probe(struct device *dev)
>  	if (!client)
>  		return 0;
>  
> -	if (!client->irq) {
> +	driver = to_i2c_driver(dev->driver);
> +
> +	if (!client->irq && !driver->disable_i2c_core_irq_mapping) {
>  		int irq = -ENOENT;
>  
>  		if (client->flags & I2C_CLIENT_HOST_NOTIFY) {
> @@ -1007,8 +1009,6 @@ static int i2c_device_probe(struct device *dev)
>  		client->irq = irq;
>  	}
>  
> -	driver = to_i2c_driver(dev->driver);
> -
>  	/*
>  	 * An I2C ID table is not mandatory, if and only if, a
> suitable Device
>  	 * Tree match table entry is supplied for the probing device.
> diff --git a/include/linux/i2c.h b/include/linux/i2c.h
> index 53fa50f..3a57e3d 100644
> --- a/include/linux/i2c.h
> +++ b/include/linux/i2c.h
> @@ -149,6 +149,7 @@ enum i2c_alert_protocol {
>   * @detect: Callback for device detection
>   * @address_list: The I2C addresses to probe (for detect)
>   * @clients: List of detected clients we created (for i2c-core use
> only)
> + * @disable_i2c_core_irq_mapping: Tell the i2c-core to not do irq-
> mapping
>   *
>   * The driver.owner field should be set to the module owner of this
> driver.
>   * The driver.name field should be set to the name of this driver.
> @@ -212,6 +213,8 @@ struct i2c_driver {
>  	int (*detect)(struct i2c_client *, struct i2c_board_info *);
>  	const unsigned short *address_list;
>  	struct list_head clients;
> +
> +	bool disable_i2c_core_irq_mapping;
>  };
>  #define to_i2c_driver(d) container_of(d, struct i2c_driver, driver)
>

On Wed, Apr 05, 2017 at 12:03:34AM +0200, Hans de Goede wrote:
> By default the i2c-core will try to get an irq with index 0 on ACPI / of
> instantiated devices. This is troublesome on some ACPI systems where the
> irq info at index 0 in the CRS table may contain nonsense and/or point
> to an irqchip for which there is no Linux driver.
> 
> If this happens then before this commit the driver's probe method would
> never get called because i2c_device_probe will try to get an irq by
> calling acpi_dev_gpio_irq_get which will always return -EPROBE in this
> case, as it waits for a matching irqchip driver to load. Thus causing
> the driver to not get a chance to bind.
> 
> This commit adds a new disable_i2c_core_irq_mapping flag to struct
> i2c_driver which a driver can set to tell the core to skip irq mapping.
> 
> Signed-off-by: Hans de Goede <hdegoede@redhat.com>

Applied to for-next, thanks!