|
Description
|
The putback for this CR:
6647542 POSIX scheduling should be compatible with Solaris scheduling classes
introduced a severe failure mode for machines with separate
root (/) and /usr disk partitions.
The problem is caused by libc's initialization section. It interrogates
the real-time class interface to determine the real-time class id,
in this line of code:
self->ul_rtclassid = get_info_by_policy(SCHED_FIFO)->pcc_info.pc_cid;
Juergen Keil figured it out. He said:
Root cause is that libc_init needs to be able to load the sched/RT
kernel module, but the RT module lives in the /usr filesystem. When
init(1M) starts, /usr isn't mounted yet (if your system is setup
to use a separate /usr filesystem). For this reason
get_info_by_policy(SCHED_FIFO) returns NULL, and we're crashing in
libc_init. 8-/
The net result is that init fails at boot and the kernel continues
fruitlessly to restart it. You get a string of these messages:
WARNING: init(1M) exited on fatal signal 11: restarting automatically
WARNING: init(1M) exited on fatal signal 11: restarting automatically
WARNING: init(1M) exited on fatal signal 11: restarting automatically
WARNING: init(1M) exited on fatal signal 11: restarting automatically
...
See the suggested fix.
|
