[head tinderbox] failure on i386/i386

Adrian Chadd adrian at freebsd.org
Thu Nov 10 17:15:16 UTC 2011


The problem right now is that when the code is compiled as a module,
it sucks in -all- of the chipset support rather than only a subset.
So there are two problems:

* AH_SUPPORT_AR5416 is still needed to build the driver and HAL, and
* the module builds all chipset support, whether or not
AH_SUPPORT_AR5416 is defined.

I notice that some modules do check whether VIMAGE is defined (.if
defined(VIMAGE)) - is there some equally evil way that I can propagate
some build time options to the module build to specify which modules
are built, based on the existance or not of AH_SUPPORT_AR5416?

(It gets messier, as at least two wireless SoC modules just don't need
to be built unless the user definitely wants them to be.)

I still think the correct thing to do at the present time is to
explicitly not build the ath modules unless the architecture can use
it, and if it can use it, build them with explicit net80211/ath
configuration. Otherwise you end up with modules/kernel being out of
sync in terms of configuration. For example, a user reported (and this
is why I "fixed" sys/modules/ath/Makefile) that they had added
ATH_DEBUG and AH_DEBUG - but because the Makefile overrides opt_ah.h
regardless of whether it exists or not, it didn't include the
debugging code. Similar hilarity occurs with/without IEEE80211_TDMA
and IEEE80211_DEBUG for example.



More information about the freebsd-current mailing list