Working in the matt-nb5-mips64 branch (based on 5.0_STABLE), I ran into a couple MP problems in subr_autoconf.c:
since configure2() starts secondary processors before kicking off interrupt config threads, config_interrupts_thread() (and everything it calls) need to me made MP safe. first, the deferred_config_queue and interrupt_config_queue need locking to prevent the tail queues from getting munged. second, some (all?) of the dc_func calls go to (driver) functions that are not MP safe. I witnessed these failures on a RMI XLS-616 based MIPS MP system. I committed fixes to subr_autoconf.c that seem to fix those problems: http://mail-index.netbsd.org/source-changes/2011/02/08/msg018238.html Has anyone else run into this type of problem on other MP platforms? Does the fix look reasonable? regards, -cliff-
