These routines were originally developed so that device drivers could manage 64-bit pointers on devices that save space only for 32-bit pointers.
Many device drivers need to pass a 32-bit value to the hardware when attempting I/O. Later, when that I/O completes, the only way the driver has to identify the request that generated that I/O is via a "token". When the I/O is initiated, the driver passes this token to the hardware. When the I/O
completes the hardware passes back this 32-bit token.
Before Solaris supported 64-bit pointers, device drivers just passed a raw 32-bit pointer to the hardware. When pointers grew to be 64 bits this was no longer possible. The id32_*() routines were created to help drivers translate between 64-bit pointers and a 32-bit token.
Given a 32- or 64-bit pointer, the routine id32_alloc() allocates a 32-bit token, returning 0 if KM_NOSLEEP was specified and memory could not be allocated. The allocated token is passed back to id32_lookup() to obtain the original 32- or
64-bit pointer.
The routine id32_free() is used to free an allocated token. Once id32_free() is called, the supplied token is no longer valid.
Note that these routines have some degree of error checking. This is done so that an invalid token passed to id32_lookup() will not be accepted as valid. When id32_lookup() detects an invalid token it returns NULL. Calling routines should check for this return
value so that they do not try to dereference a NULL pointer.
|