> ipw2100_impl.h: > need to include <net/if.h> somewhere? seems like IFF_PROMISC is used > in that driver.
I will add it (right now it is included indirectly ipw2100_impl.h <- mac.h <- dld.h <- net/if.h). > - mac_setprop(): > in the (!init_link) case, we only want to commit the > changes via mac_prop_update if the mc_setprop callback itself > succeeds Hmm, I thought I fixed that but evidently I didn't. Thanks for noticing. > - what is the expected typical/atypical driver usage? I think > we had discussed possible cases where the driver caches the mph, and > talked about how to deal with cached prop handles by using > refcounts. Is that deferred for a separate wad of changes? No, this is all supported in this wad. proplists are refcounted, they won't be freed until it reaches 0. Drivers increment/decrement it via mac_prop_init/fini. The framework does this via mac_prop_load/unload - which are called from mac_start/stop. So all cases are covered: when only the framework uses cached properties, when only the driver uses them, and when both. > - nit: the XXX at line 2424 should be fixed in the gate by now, > after Samant's putback, so this comment can be removed. OK. > - Not an urgent priority, more an RFE: > would be nice to have an mdb walker for the mac property list. What's the advantage compared to ::list piped into ::print ? > This changes the interfaces in Section 4 of the design doc. > If you have any edits for that section (maybe its better to leave > specifics out, and ARC the details later for the Brussels-persistence > case), please send them my way. I'll look into this. thanks, -Artem
