sched while atomic

Latest 2.6.11rc* give me this interesting message at boot:
Mount-cache hash table entries: 512 (order: 0, 4096 bytes)
scheduling while atomic: swapper/0x00000002/0
Call trace:
 [c01c7580] schedule+0x640/0x6bc
 [c0004698] syscall_exit_work+0x120/0x124
 [c00a5414] proc_device_tree_init+0x7c/0x98
 [c02be9b4] proc_root_init+0x14c/0x158
 [c02a660c] start_kernel+0x178/0x1b0
 [00003a5c] 0x3a5c

Since it doesn't happen on x86, I first thought it was because of the ppc 
specific init of the device_tree, but commenting this out just lets it 
happen somewhere else. It seems schedule is called when it returns from a 
syscall, which seems normal behaviour to me, but the recent 
preempt_disable() in start_kernel makes the scheduler give these warnings.
So what's happening? 

Also, I think this is already know for a while:
/lib/modules/2.6.11-rc3/kernel/drivers/video/vga16fb.ko needs unknown 
symbol vgacon_remap_base

regards,

danny