Re: 2.6.15-rc5-rt4: BUG: swapper:0 task might have lost a preemption check!

[Steven Rostedt <rostedt@goodmis.org>]

On Thu, 2005-12-22 at 21:14 -0600, K.R. Foley wrote:
> Lee Revell wrote:
> > Got this on boot.  Same .config as the last one I sent you.
> > 
> > VP_IDE: VIA vt8235 (rev 00) IDE UDMA133 controller on pci0000:00:11.1
> >     ide1: BM-DMA at 0xe008-0xe00f, BIOS settings: hdc:DMA, hdd:pio
> > Probing IDE interface ide1...
> > BUG: swapper:0 task might have lost a preemption check!
> >  [<c010440c>] dump_stack+0x1c/0x20 (20)
> >  [<c01166aa>] preempt_enable_no_resched+0x5a/0x60 (20)
> >  [<c0100dd9>] cpu_idle+0x79/0xb0 (12)
> >  [<c0100280>] _stext+0x40/0x50 (28)
> >  [<c03078e6>] start_kernel+0x176/0x1d0 (20)
> >  [<c0100199>] 0xc0100199 (1086889999)
> > ---------------------------
> > | preempt count: 00000000 ]
> > | 0-level deep critical section nesting:
> > ----------------------------------------
> > 
> > 


OK, I just found an SMP bug, and here's the patch.  Maybe this will help
you kr.  I'm currently running x86_64 SMP with 2.6.15-rc5-rt4 with this
and my softirq-no-hrtimers patch I sent earlier.

-- Steve

Index: linux-2.6.15-rc5-rt4/kernel/workqueue.c
===================================================================
--- linux-2.6.15-rc5-rt4.orig/kernel/workqueue.c	2005-12-23 10:23:25.000000000 -0500
+++ linux-2.6.15-rc5-rt4/kernel/workqueue.c	2005-12-23 10:25:21.000000000 -0500
@@ -370,10 +370,17 @@
 void set_workqueue_thread_prio(struct workqueue_struct *wq, int cpu,
 				int policy, int rt_priority, int nice)
 {
-	struct task_struct *p = wq->cpu_wq[cpu].thread;
+	struct cpu_workqueue_struct *cwq;
+	struct task_struct *p;
 	struct sched_param param = { .sched_priority = rt_priority };
+	unsigned long flags;
 	int ret;
 
+	cwq = per_cpu_ptr(wq->cpu_wq, cpu);
+	spin_lock_irqsave(&cwq->lock, flags);
+	p = cwq->thread;
+	spin_unlock_irqrestore(&cwq->lock, flags);
+
 	set_user_nice(p, nice);
 	ret = sys_sched_setscheduler(p->pid, policy, &param);
 	if (ret)