for "BUG: using smp_processor_id() in preemptible" in 2.6.17.rc6.mm2
PATCH 1/1 for "BUG: using smp_processor_id() in preemptible" in 2.6.17.rc6.mm2
Hi all
I saw a stack trace while booting the 2.6.17-rc6-mm2 kernel. Which looks
like this
BUG: using smp_processor_id() in preemptible [00000001] code: init/1 caller is __handle_mm_fault+0x22/0x248 [<c010334c>] show_trace+0x27/0x29 [<c0103486>] dump_stack+0x26/0x28 [<c02e4d6f>] debug_smp_processor_id+0x9f/0xb0 [<c0157a98>] __handle_mm_fault+0x22/0x248 [<c01140b2>] do_page_fault+0x17e/0x605 [<c01030b9>] error_code+0x39/0x40 [<c0198eef>] padzero+0x2c/0x37 [<c0199ff5>] load_elf_binary+0x6e2/0xcfc [<c0174e14>] search_binary_handler+0x96/0x201 [<c017510e>] do_execve+0x18f/0x243 [<c010152b>] sys_execve+0x3c/0x81 [<c04a9c93>] syscall_call+0x7/0xb
Cause for the bug is a set of functions count_vm_event(),
count_vm_events() and get_cpu_vm_events() calling smp_processor_id() in preemptible context without disabling irq's
So I developed patch for this problem,attaching the patch here
include/linux/page-flags.h +11 -1
@@ -141,17 +141,27 @@ DECLARE_PER_CPU(struct vm_event_state, vstatic inline unsigned long get_cpu_vm_events (enum vm_event_item item) { -return __get_cpu_var (vm_event_states).event[item]; +unsigned long flags, res; +local_irq_save (flags); + res =__get_cpu_var (vm_event_states).event[item]; +local_irq_restore (flags); +return res; }static inline void count_vm_event (enum vm_event_item item) { +unsigned long flags; +local_irq_save (flags);__get_cpu_var (vm_event_states).event[item]++; +local_irq_restore (flags); }static inline void count_vm_events (enum vm_event_item item,long delta) { +unsigned long flags; +local_irq_save (flags);__get_cpu_var (vm_event_states).event[item] += delta; +local_irq_restore (flags); }extern void all_vm_events (unsigned long *);