When you compile in the ESP driver in 2.4.7+ the kernel will hang right
before it spawns init. The last line if output I see is the "Freeing
unused kernel memory: " message.
Using the Magic SysReq keys, I see that the kernel seems to be spinning
on c011b47f. I also see c01c6a9d every once in a while. Here's the
appropriate System.map lines:
c011ae20 T do_sys_settimeofday
c011aee0 T sys_settimeofday
c011afe0 T do_adjtimex
c011b3e0 T sys_adjtimex
c011b490 T do_softirq
c011b560 T raise_softirq
c011b5b0 T open_softirq
c011b5d0 T tasklet_schedule
c011b650 T tasklet_hi_schedule
c011b6c0 t tasklet_action
c011b7b0 t tasklet_hi_action
c011b8a0 T tasklet_init
c011b8d0 T tasklet_kill
c011b960 t bh_action
....
c01c6740 t alloc_tty_struct
c01c6770 t _tty_make_name
c01c67c0 T tty_name
c01c67e0 t check_tty_count
c01c6880 T tty_register_ldisc
c01c68e0 t tty_set_ldisc
c01c6a80 T get_tty_driver
c01c6ad0 T tty_check_change
c01c6b40 t hung_up_tty_read
c01c6b60 t hung_up_tty_write
c01c6b80 t hung_up_tty_poll
c01c6b90 t hung_up_tty_ioctl
c01c6bb0 t tty_lseek
v2.4.6 works fine with the same config. v2.4.7 works fine too if I don't
include the ESP driver. Note: The hang occurs whether you have an ESP card
or not.
Since there hasn't been any ESP changes, I assume it's a bug in the ESP driver
triggered by the softirq changes. I would try to do binary searches for the
specific section of code that makes it break, but there are softirq changes
all over the place in 2.4.7 and I didn't feel I was up to the task.
Attached is the config for the broken v2.4.7.
-kf
This archive was generated by hypermail 2b29 : Sat Sep 15 2001 - 21:00:48 EST