Rework device API: one device in many classes

[Nicolas Pouillon]

We currently have a problem with devices that are in more than one driver class at the same time. We have to split them in subdevices, even if it is incorrect.

Some examples:

• CPU nodes are also ICUs
• soclib Xicu is timer and ICU

There once was a discussion about allowing devices to implement more than one class at the same time. How to do this ?

[Nicolas Pouillon]

Could we do:

enum device_class_e
{
    DEVICE_CLASS_FOO,
#if defined(CONFIG_FEATURE_BAR)
    DEVICE_CLASS_BAR,
#endif
    DEVICE_CLASS_COUNT,
};

#define DEVICE_CLASS_UNAVAILABLE 0xff

struct driver_funcs_s
{
    union {
        struct {
            void (*op1)(struct device_s *dev, … );
            void (*op2)(struct device_s *dev, … );
        } foo;
#if defined(CONFIG_FEATURE_BAR)
        struct {
            void (*op1)(struct device_s *dev, … );
            void (*op2)(struct device_s *dev, … );
        } bar;
#endif
    };
}

struct driver_s
{
    error_t (*init)(struct device_s *dev, void *param);
    error_t (*cleanup)(struct device_s *dev);

#if defined(CONFIG_DRIVER_MULTICLASS)
    uint8_t funcs_offset[DEVICE_CLASS_COUNT];
#endif

    struct driver_funcs_s funcs[0];
};

Then to access funcs of type BAR:

const struct driver_funcs_s *driver_class_funcs(
    const struct driver_s *drv,
    enum device_class_e class)
{
#if defined(CONFIG_DRIVER_MULTICLASS)
    const uint8_t offset = drv->funcs_offset[class];

    assert(offset != DEVICE_CLASS_UNAVAILABLE);
#else
    const uint8_t offset = 0;
#endif

    return &drv->funcs[offset];
}

void device_bar_op1(struct device_s *dev, …)
{
    const struct driver_funcs_s *funcs =
        driver_class_funcs(dev->drv, DEVICE_CLASS_BAR);

    return funcs->bar.op1( dev, … );
}

Added from current code, there is a new indirection through drv->funcs_offset.

• There is a cost for a byte load in addition to current implementation
  • but as it is constant, this is cached data.
• There is a cost for a table of byte
  • but as it is an index in a char, we loose at most DEVICE_CLASS_COUNT bytes per driver.
• We can still make driver_class_funcs return &funcs[0] if there is no support for multiple drivers per device.

Is this acceptable ?

[Nicolas Pouillon]

PoC