Currently the code does things like char _stp_lbuf[NR_CPUS][STP_LOG_BUF_LEN + 1];
Fixed all but io.c a while ago.
_stp_lbuf in io.c and _stp_string in string.h are using huge amounts of BSS space, especially for x86_64 builds.
Fixed. 2007-01-30 Martin Hunt <hunt@redhat.com> * io.c (_stp_vlog): Use dynamic percpu allocations instead of very wasteful static allocations. * print.c (_stp_print_init): Do percpu allocations for io.c. (_stp_print_cleanup): Free percpu allocations.