David > Christopher Horne wrote: > >> o OBP sd(7D)/st(7D) "stub" (wildcard) nodes are an implementation >> artifact that should be hidden. These nodes do not represent >> devices, they have no addressing information. With a facility >> to mark these nodes as hidden in a driver's >> DDI_CTLOPS_INITCHILD, confusion surrounding their existence in >> the Solaris device tree can be avoided. > > I'm not sure you can hide those nodes without making > some other changes to the glue code that's used to > convert OS dev_t's and/or device pathnames to OBP > pathnames used by the reboot command and by the installers, > for example. > > They really shouldn't be a problem. They don't have > unit addresses and are effectively equivalent to what > you are proposing to do, except there's only 1 per > target device type per scsi-like bus. > > Will that glue code still see the obp wildcard nodes? > I'm pretty sure it needs them to get the correct target > device name. Maybe they are only needed for legacy > non-generic target device names, but I don't know if > the code distinguishes between those cases or not. > > Either way, the "compatible" propval is still needed > from them for mounting root, I believe. > > -David
The issue of how Solaris deals with OBP 'wilcard' nodes draws attention away from the primary focus of this proposal. I have removed the above paragraph, and placed an updated proposal text here: http://sac.eng/PSARC/2009/203/materials/ndi_hide.fasttrack.txt -Chris
