Large ramdisk crashes system with 2.4.7

[Paul Buder]

A large ramdisk will crash the system with kernel 2.4.7.  This is a
128 meg system with a lilo parameter of 'ramdisk=100000', an ext2 fs
on /dev/ram0 created with 80 megs and an attempted dd from /dev/zero
of 60 megs.  I reported this for 2.4.5 and 2.4.6 also.  I notice that
bdflush is looping through the same eight functions about forty times at
the time of the crash which can't be a good thing.  Here is the ksymoops
output for bdflush.

Trace; c012a3fd <__alloc_pages+109/278>
Trace; c012a2f2 <_alloc_pages+16/18>
Trace; c01333c2 <grow_buffers+3e/18c>
Trace; c01315c4 <refill_freelist+1c/50>
Trace; c01315d0 <refill_freelist+28/50>

# Starts looping here

Trace; c0131ab4 <getblk+158/16c>
Trace; c017d5e2 <rd_make_request+72/f4>
Trace; c0175f4c <generic_make_request+110/11c>
Trace; c0175fb2 <submit_bh+5a/74>
Trace; c01761bb <ll_rw_block+1ef/260>
Trace; c01338fa <flush_dirty_buffers+9e/e4>
Trace; c013396e <wakeup_bdflush+2e/34>
Trace; c01315de <refill_freelist+36/50>

# Round two

Trace; c0131ab4 <getblk+158/16c>
Trace; c017d5e2 <rd_make_request+72/f4>
Trace; c0175f4c <generic_make_request+110/11c>
Trace; c0175fb2 <submit_bh+5a/74>
Trace; c01761bb <ll_rw_block+1ef/260>
Trace; c01338fa <flush_dirty_buffers+9e/e4>
Trace; c013396e <wakeup_bdflush+2e/34>
Trace; c01315de <refill_freelist+36/50>