lkml.org 
[lkml]   [2005]   [Aug]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: 2.6.13-rc4-mm1: Divide by zero in find_idlest_group
Date
On Fri, 12 Aug 2005 01:21 am, Dave Kleikamp wrote:
> I encounted this trap on a 2-way i386 box running 2.6.13-rc4-mm1:
>
> [70347.743727] divide error: 0000 [#2]
> [70347.752979] PREEMPT SMP DEBUG_PAGEALLOC
> [70347.773060] last sysfs file: /devices/pnp0/00:11/id
>
> Program received signal SIGTRAP, Trace/breakpoint trap.
> 0xc0119556 in find_idlest_group (sd=0xc1084dc0, p=0xc3b96030, this_cpu=0)
> at sched.c:1033
> 1033 target_load = target_load * rq->prio_bias /
> rq->nr_running;
>
> (gdb) where
> #0 0xc0119556 in find_idlest_group (sd=0xc1084dc0, p=0xc3b96030,
> this_cpu=0) at sched.c:1033
> #1 0xc0119744 in sched_balance_self (cpu=0, flag=4) at sched.c:1151
> #2 0xc011a63f in sched_exec () at sched.c:1840
> #3 0xc016cacb in do_execve (filename=0xc3e35000 "/bin/sh",
> argv=0xbfff512c, envp=0xbfff6010, regs=0xa00) at exec.c:1162
> #4 0xc0101b5a in sys_execve (regs=
> {ebx = -1208173323, ecx = -1073786580, edx = -1073782768, esi =
> -1073782768, edi = -1208107392, ebp = -1073786620, eax = 11, xds = 123, xes
> = -1072693125, orig_eax = 11, eip = -1208664822, xcs = 115, eflags = 658,
> esp = -1073786628, xss = 123}) at process.c:787 #5 0xc0102fdd in
> syscall_call () at current.h:9
> #6 0xb7fcbcf5 in ?? ()
>
> print *(struct runqueue *)0xc108c400
> $1 = {lock = {raw_lock = {slock = 1}, break_lock = 0}, nr_running = 0,
> prio_bias = 0, cpu_load = {0, 0, 0}, nr_switches = 8877527,
> nr_uninterruptible = 132930, expired_timestamp = 0,
> timestamp_last_tick = 70347783539837, curr = 0xc1103550, idle =
> 0xc1103550, prev_mm = 0x0, active = 0xc108c8c0, expired = 0xc108c448,
> arrays = {{ nr_active = 0, bitmap = {0, 0, 0, 0, 4096}, queue = {{next =
> 0xc108c460, prev = 0xc108c460}, {next = 0xc108c468, prev = 0xc108c468}, {
> next = 0xc108c470, prev = 0xc108c470}, {next = 0xc108c478, prev =
> 0xc108c478}, {next = 0xc108c480, prev = 0xc108c480}, { next = 0xc108c488,
> prev = 0xc108c488}, {next = 0xc108c490,
>
> in __source_load, it looks like rq_nr_running must have changed between
> the if statement and the divide:
>
> if (rq->nr_running > 1 || (idle == NOT_IDLE && rq->nr_running))
> /*
> * If we are busy rebalancing the load is biased by
> * priority to create 'nice' support across cpus. When
> * idle rebalancing we should only bias the source_load if
> * there is more than one task running on that queue to
> * prevent idle rebalance from trying to pull tasks from a
> * queue with only one running task.
> */
> source_load = source_load * rq->prio_bias / rq->nr_running;
>
> Should there be any locking around this? Or should the value of
> rq->nr_running be saved to a local variable as in this untested patch?

Very sneaky..

On initial inspection your patch makes complete sense. I see no point in
adding locking to this function as the accuracy is not critical. Want to give
your patch a run and then push it to akpm? Thanks!

Cheers,
Con
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-08-11 17:36    [W:0.048 / U:0.776 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site