PATCH to fix irq sharing and SA_INTERRUPT on x86. please review
further to my email yesterday (to which i've had no response :) i propose the attached patch to arch/i386/kernel/irq.c. it corrects what i see as a bug in interrupt handling.
currently if a driver requests SA_INTERRUPT in an interrupt handler, it is only called with interrupts disabled if it is the first handler in the list.
my patch modifies setup_irq to put any interrupt with SA_INTERRUPT in the front of the handler queue (eg before any handlers without the flag).
and also modifies handle_IRQ_event to only enable interrupts when it hits the first handler with SA_INTERRUPT not set.
the patch is against 2.5.62 and greatly improves stability on my SMP system.
james
arch/i386/kernel/irq.c +34 -17
@@ -201,11 +201,13 @@int handle_IRQ_event (unsigned int irq,struct pt_regs * regs,struct irqaction * action) {int status =1 ;/* Force the "do bottom halves" bit */ - -if (!(action->flags &SA_INTERRUPT )) -local_irq_enable (); +int enabled =SA_INTERRUPT ; do { + if (~action->flags & enabled) { + local_irq_enable(); + enabled = 0; + } status |= action->flags; action->handler(irq, action->dev_id, regs); action = action->next;
@@ -761,23 +763,38 @@* The following block of code has to be executed atomically */ spin_lock_irqsave (&desc->lock,flags); - p = &desc->action; -if ((old = *p) !=NULL ) { -/* Can't share interrupts unless both agree to */ -if (!(old->flags & new->flags &SA_SHIRQ )) { -spin_unlock_irqrestore (&desc->lock,flags); -return -EBUSY ; +if (new->flags &SA_INTERRUPT ) { + p = &desc->action; + old = *p; +/* Add SA_INTERRUPT interrupts to the front of the queue +* 'cos it's faster than adding them in the middle */ +if (old) { +if (!(old->flags & new->flags &SA_SHIRQ )) { +spin_unlock_irqrestore (&desc->lock,flags); +return -EBUSY ; + } + new->next = old; + shared =1 ; } + *p = new; + }else { + p = &desc->action; +if ((old = *p) !=NULL ) { +/* Can't share interrupts unless both agree to */ +if (!(old->flags & new->flags &SA_SHIRQ )) { +spin_unlock_irqrestore (&desc->lock,flags); +return -EBUSY ; + } -/* add new interrupt at end of irq queue */ -do { - p = &old->next; - old = *p; - }while (old); - shared =1 ; +/* add new interrupt at end of irq queue */ +do { + p = &old->next; + old = *p; + }while (old); + shared =1 ; + } + *p = new; } - - *p = new;if (!shared) { desc->depth =0 ;
Quoted 8 lines · James Harper
further to my email yesterday (to which i've had no response :) i propose the attached patch to
arch/i386/kernel/irq.c. it corrects what i see as a bug in interrupt handling.currently if a driver requests
SA_INTERRUPTin an interrupt handler, it is only called with interrupts disabled if it is the first handler in the list.
ewww..
Quoted 6 lines
my patch modifies
setup_irqto put any interrupt withSA_INTERRUPTin the front of the handler queue (eg before any handlers without the flag).and also modifies
handle_IRQ_eventto only enable interrupts when it hits the first handler withSA_INTERRUPTnot set.
Yes, that's a nice fix, thanks.
Other architectures appear to have inherited this bug.