Re: [PATCH 1/1] Revert "genirq: Remove the second parameter from handle_irq_event_percpu()"

[zhuyj]

Hi, all

I made tests for this patch. To now, I can not find any similar problem.

Best Regards!
Zhu Yanjun

On 01/14/2016 09:29 AM, Huang Shijie wrote:
> On Wed, Jan 13, 2016 at 02:07:25PM +0100, Thomas Gleixner wrote:
> > On Wed, 13 Jan 2016, zyjzyj2000@xxxxxxxxx wrote:
> >
> > > After this commit 71f64340fc0e ("genirq: Remove the second parameter
> > > from handle_irq_event_percpu()") is applied, the variable action is
> > > not protected by raw_spin_lock. The following calltrace will pop up.
> > Thanks, for the report. I missed that detail when merging the patch!
> >
> > Just for correctness sake: You miss to explain why this can happen.
> >
> > It's not about the variable action, it's about desc->action not being
> > protected anymore. So the reason why this oopses is that the action is being
> > removed concurrently.
> >
> > CPU 0			CPU 1
> >
> > free_irq()		lock(desc)
> > lock(desc)		handle_edge_irq()
> > 			  handle_irq_event(desc)
> > 			    unlock(desc)
> > desc->action = NULL	    handle_irq_event_percpu(desc)
> > 	       		      action = desc->action
> >
> > While the original code did:
> >
> > free_irq()		lock(desc)
> > lock(desc)		handle_edge_irq()
> > 			  handle_irq_event()
> > 	       		    action = desc->action
> > 			    unlock(desc)
> > desc->action = NULL	    handle_irq_event_percpu(desc, action)
> > 	       		
> > So now the question is whether we revert that patch or simply change
> > handle_irq_event_percpu() to deal with that. Patch below.
> >
> > That preserves us the code size reduction of commit 71f64340fc0e. This is safe
> > because we either see a valid desc->action or NULL. If the action is about to
> > be removed it is still valid as free_irq() is blocked on synchronize_irq().
> >
> > free_irq()		lock(desc)
> > lock(desc)		handle_edge_irq()
> > 			  handle_irq_event(desc)
> > 			    set(INPROGRESS)
> > 			    unlock(desc)
> > 			      handle_irq_event_percpu(desc)
> > 	       		        action = desc->action
> > desc->action = NULL
> > sychronize_irq()
> >    while(INPROGRESS);	   lock(desc)
> > 			   clr(INPROGRESS)
> > free(action)
> >
> > That's basically the same mechanism as we have for shared
> > interrupts. action->next can become NULL while handle_irq_event_percpu()
> > runs. Either it sees the action or NULL. It does not matter, because action
> > itself cannot go away.
> >
> > Thanks,
> >
> > 	tglx
> >
> > 8<-------------
> >
> > --- a/kernel/irq/handle.c
> > +++ b/kernel/irq/handle.c
> > @@ -136,9 +136,15 @@ irqreturn_t handle_irq_event_percpu(stru
> >   {
> >   	irqreturn_t retval = IRQ_NONE;
> >   	unsigned int flags = 0, irq = desc->irq_data.irq;
> > -	struct irqaction *action = desc->action;
> > +	struct irqaction *action;
> >   
> > -	do {
> > +	/*
> > +	 * READ_ONCE is not required here. The compiler cannot reload action
> > +	 * because it'll be action->next for the second iteration of the loop.
> > +	 */
> > +	action = desc->action;
> > +
> > +	while (action) {
> >   		irqreturn_t res;
> >   
> >   		trace_irq_handler_entry(irq, action);
> > @@ -173,7 +179,7 @@ irqreturn_t handle_irq_event_percpu(stru
> >   
> >   		retval |= res;
> >   		action = action->next;
> > -	} while (action);
> > +	}
> >   
> >   	add_interrupt_randomness(irq, flags);
> I prefer to this patch, revert the old the patch is not a good solution.
>
> thanks
> Huang Shijie