Initialization, configuration, and removal routines

The process for coding driver initialization and configuration in a DDI 8 driver can be summarized as follows. The _load(D2) entry point calls the drv_attach(D3) function. The kernel then calls the config(D2) entry point with the CFG_ADD subfunction once for each board, passing it the key for that board. The driver's config( ) entry point can then call the cm_intr_attach(D3) function or any other cm_xxx( ) or devmem_xxx( ) functions that need a key.

The CFG_ADD subfunction must also allocate a private instance data structure that contains or references state information for the board (possibly including the key, and likely including the interrupt cookie). This is then passed to the kernel as an opaque handle by writing it to (*(void **)idata). All subsequent entry points have this idata value passed back in to the driver so it can get at its per-board state.

The following sections summarize the entry point modifications required for initialization, configuration, and removal when porting drivers from earlier DDI versions or SCO OpenServer 5 to run on DDI 8:

• Remove the init(D2) and start(D2) routines, if present, and code driver initialization in the _load(D2) entry point routine and per-instance initialization for hardware drivers in the CFG_ADD subfunction of the config(D2) routine.

• The _load(D2) entry point routine is required for all DDI 8 drivers. It is called when the driver is dynamically loaded into the kernel or, for statically-configured drivers, during system initialization. _load( ) does general driver initialization.

  If your pre-DDI 8 driver included an optional _load( ) routine, modify it as follows:

  • Remove the prefix from the _load( ) call. For DDI 8 drivers, the _load( ) entry point routine must be unprefixed.

  • Remove any ``wrapper'' macro coding from the routine. The wrapper macros are not required for DDI 8 drivers because all DDI 8 drivers are loadable and their entry points are defined through structures rather than switch tables.

  • Move any hardware-specific configuration information to the CFG_ADD subfunction of the config(D2) entry point routine.

  Code the following in the _load( ) routine whether porting from an earlier DDI version: or SCO OpenServer 5:

  • Populate a drvinfo(D4) structure with information about the driver, including a pointer to the drvops(D4) structure that points to the driver's entry point routine. Any entry point routines that are not used in the driver but have entries in the drvops structure just be set to NULL.

    STREAMS drivers, including MDI drivers, will also need to populate a streamtab(D4str) structure. SDI HBA drivers need to identify additional entry point routines through the hba_info(D4sdi) structure. See the manual pages and sample driver code for more details.

  • Add a call to the drv_attach(D3) function to attach the drvinfo structure.

  • Include any driver-specific initialization that was previously coded in the driver's init(D2) or start(D2) routine.

• Loadable drivers should also include a _unload(D2) entry point routine that calls the drv_detach(D3) function to deregister the driver and frees any general driver resources that were allocated in the _load( ) routine such as any kernel memory and lock structures that were allocated for the driver rather than device instances.

• The config(D2) entry point routine is required for all hardware drivers. config( ) does per-instance initialization and deinitialization. All drivers must include cases for the CFG_ADD subfunction. Hotplug-ready devices (the D_HOT flag is set in the drvinfo(D4) structure), must support additional subfunctions, including CFG_SUSPEND and CFG_REMOVE; see the config(D2) manual page for details.

  • The CFG_ADD subfunction must attach interrupts for hardware drivers by calling the cm_intr_attach(D3) function.

  • The CFG_SUSPEND and CFG_REMOVE subfunctions should call the untimeout(D3) function to cancel any itimeout(D3) or bufcall(D3str) requests and call the cm_intr_detach(D3) function to free up interrupt resources. It should also free all resources that were allocated per device in the CFG_ADD subfunction. The CFG_REMOVE subfunction should free resources only if the CFG_SUSPEND subfunction has not already freed them.

  • ISA drivers must implement the CFG_VERIFY subfunction of the config(D2) entry point routine to check that the configuration parameters are correct for this board. This code replaces the _verify(D2) entry point routine used in DDI drivers prior to version 8.

• Replace calls to the halt(D2) entry point routine with calls to the CFG_SUSPEND subfunction of the config(D2) entry point routine.