property.h

Go to the documentation of this file.

#ifndef __LINUX_PROPERTY_H__
#define __LINUX_PROPERTY_H__
 
/*
 * DCL <linux/property.h> -- the software/fwnode property readers, and
 * dev_fwnode().
 *
 * mainline's header is the interface to device tree, ACPI and software
 * nodes: reading "rs485-rts-delay" out of the DT, walking fwnode graphs,
 * fwnode_property_read_u32_array()'s many wrappers.  Xeneva has no firmware
 * description in the kernel at all -- a board's UART is compiled in -- so
 * there is nothing to read, and all three functions answer that.
 *
 * The answers are chosen by looking at what serial_core.c does with them
 * (uart_set_rs485_config, :3533-3559), rather than invented:
 *
 *   dev_fwnode()                NULL, so `if (!dev_fwnode(dev)) return 0;`
 *                                takes its early exit and never claims a
 *                                property exists.
 *   device_property_read_u32_array()  nonzero (-ENODATA), so the `if (!ret)`
 *                                that would apply rs485-rts-delay timings is
 *                                skipped -- and a half-applied pair of
 *                                delays would be worse than none.
 *   device_property_read_bool()  0, so SER_RS485_RX_DURING_TX is only ever
 *                                set by the flags a caller passes in, never
 *                                by a DT node that is not there.
 *
 * dev_fwnode() lives here rather than in device.h because that is where
 * mainline declares it (include/linux/property.h), and device.h includes this
 * header for it.
 *
 *   upstream  include/linux/property.h  (mainline v7.2)
 */
#include <linux/errno.h>    /* -ENODATA */
#include <linux/types.h>    /* u32, u8 */
#include <linux/list.h> /* struct list_head -- fwnode_handle's supplier and consumers */
 
struct device;
 
/*
 * struct fwnode_handle -- mainline's definition, at property.h:66-73 of the
 * same header. It was a forward declaration here until drm_connector.c:3492
 * needed it: that line reads `connector->fwnode->secondary`, and a member
 * access needs the definition, not the promise.
 *
 * All seven members are carried, not only the one that is read. A struct
 * with three of them is indistinguishable from one with seven until somebody
 * writes the fourth, and the write would compile and land at whatever offset
 * the compiler had computed for it -- the failure an ABI pin exists to
 * prevent, arriving in a file that is not pinned.
 *
 * This layout is never persisted and never crosses a boundary: there are no
 * fwnodes in this kernel, so dev_fwnode() below returns NULL for every
 * device and drm_connector.c:3492's comparison is always false.
 */
struct fwnode_handle {
    struct fwnode_handle *secondary;
    const struct fwnode_operations *ops;
    struct device *dev;
    struct list_head supplier;
    struct list_head consumers;
    u8 flags;
};
 
static inline struct fwnode_handle* dev_fwnode(const struct device* dev)
{
    (void)dev;
    return NULL;
}
 
static inline int device_property_read_u32_array(const struct device* dev,
                         const char* propname,
                         u32* values,
                         size_t nvalues)
{
    (void)dev;
    (void)propname;
    (void)values;
    (void)nvalues;
    return -ENODATA;
}
 
static inline int device_property_read_bool(const struct device* dev,
                        const char* propname)
{
    (void)dev;
    (void)propname;
    return 0;
}
 
#endif /* __LINUX_PROPERTY_H__ */