Device hierarchy

[John-Mark Gurney]

Leonardo Passos wrote this message on Sat, Sep 14, 2013 at 14:09 -0400:
> Hi all,
> 
> I am checking some code from the FreeBSD repository, and I came across a
> situation that I could not explain, at least from the existing
> documentation.
> 
> Looking at sys/dev/ata/ata-raid.c in the 9.1 release, I see the following
> module declaration (line 4616):
> 
> static moduledata_t ata_raid_moduledata =
>     { "ataraid", ata_raid_module_event_handler, NULL };
> DECLARE_MODULE(ata, ata_raid_moduledata, SI_SUB_RAID, SI_ORDER_FIRST);
> MODULE_VERSION(ataraid, 1);
> MODULE_DEPEND(ataraid, ata, 1, 1, 1);
> MODULE_DEPEND(ataraid, ad, 1, 1, 1);

Read below, it looks like DECLARE_MODULE's name is the same:
#define DECLARE_MODULE_WITH_MAXVER(name, data, sub, order, maxver)      \
        MODULE_DEPEND(name, kernel, __FreeBSD_version,                  \
            __FreeBSD_version, maxver);                 \
        MODULE_METADATA(_md_##name, MDT_MODULE, &data, #name);          \
        SYSINIT(name##module, sub, order, module_register_init, &data); \
        struct __hack

#define DECLARE_MODULE(name, data, sub, order)                          \
        DECLARE_MODULE_WITH_MAXVER(name, data, sub, order, MODULE_KERNEL_MAXVER)

and the first arg to SYSINIT is uniquifier...

> Another weird situation appears in sys/dev/ata-isa.c (line 201):
> 
> static driver_t ata_isa_driver = {
>     "ata",
>     ata_isa_methods,
>     sizeof(struct ata_channel),
> };
> 
> DRIVER_MODULE(ata, isa, ata_isa_driver, ata_devclass, 0, 0);
> MODULE_DEPEND(ata, ata, 1, 1, 1);
> 
> Here, the driver module name is set to "ata", and later, it is stated that
> ata depends on ata. Is the dependency referring to another ata module, at a
> different level in the driver stacking? How are these names and
> dependencies resolved?

Here, MODULE_DEPEND should probably be:
MODULE_DEPEND(isa/ata, ata, 1, 1, 1);

but I'm not sure how that would/will work...  Here DRIVER_MODULE does
something special and names the module w/ the bus/name instead of just
the first one, and why if you run kldstat -v, you'll see something like:
                45 isa/ata

in the output...  But after looking at the MODULE_DEPEND, what really
matters is that the first arg is unique per file since the macro only
uses it to name symbols:
#define MODULE_METADATA(uniquifier, type, data, cval)                   \
        static struct mod_metadata _mod_metadata##uniquifier = {        \
                MDT_STRUCT_VERSION,                                     \
                type,                                                   \
                data,                                                   \
                cval                                                    \
        };                                                              \
        DATA_SET(modmetadata_set, _mod_metadata##uniquifier)

#define MODULE_DEPEND(module, mdepend, vmin, vpref, vmax)               \
        static struct mod_depend _##module##_depend_on_##mdepend = {    \
                vmin,                                                   \
                vpref,                                                  \
                vmax                                                    \
        };                                                              \
        MODULE_METADATA(_md_##module##_on_##mdepend, MDT_DEPEND,        \
            &_##module##_depend_on_##mdepend, #mdepend)


So, adding the slash is probably not useful...  It should probably be
made clear in MODULE_DEPEND that name is only used to uniquify the
symbols and have no useful info...  Maybe we should switch to using
__LINE__ instead, and make name option/unused?

We should probably change the first arg for MODULE_DEPEND and
DECLARE_MODULE to be like SYSINIT's and name them uniquifier..

P.S. The macros I pasted came from sys/module.h.

-- 
  John-Mark Gurney				Voice: +1 415 225 5579

     "All that I will do, has been done, All that I have, has not."