Re: pic/arm lost blocked irqs

[Adam Hoka <adam.hoka%gmail.com@localhost>]

On Mon, 30 Aug 2010 22:31:43 +0900 (JST)
KIYOHARA Takashi <kiyohara%kk.iij4u.or.jp@localhost> wrote:

> Hi! Adam,
> 
> 
> From: KIYOHARA Takashi <kiyohara%kk.iij4u.or.jp@localhost>
> Date: Mon, 30 Aug 2010 21:25:31 +0900 (JST)
> 
> > From: Adam Hoka <adam.hoka%gmail.com@localhost>
> > Date: Mon, 30 Aug 2010 13:48:10 +0200
> > 
> > > i saw hangups in interrupt handling too with the dm9000 driver
> > > i have tried to port. maybe it wasnt my mistake then?
> > 
> > That patch of me fixes a bug.  However, I happen another bug, too.
> > For instance, can you type your keyboard on tip(1)(or cu(1)) of Host
> > toward the console of your devkit9000?
> > 
> > My smsh(4) works again with typing keyboard.
> 
> 
> Maybe, it is this.  ;-)
> 
> arm/pic/pic.c:pic_deliver_irqs() --
> 
>         for (;;) {
>                 pending_irqs = pic_find_pending_irqs_by_ipl(pic, irq_base,
>                     *ipending, ipl);
>                 :
>                 if (pending_irqs == 0) {
>                         irq_count += 32;
>                         if (__predict_true(irq_count >= pic->pic_maxsources))
>                                 break;
>                         :
>                 }
>                 :
>                 do {
>                         :
>                         pending_irqs = pic_find_pending_irqs_by_ipl(pic,
>                             irq_base, *ipending, ipl);
>                 } while (pending_irqs);
>                 :
>          }
>          :
>         if (atomic_and_32_nv(&pic->pic_pending_ipls, ~ipl_mask) == 0)
>                 atomic_and_32(&pic_pending_pics, ~__BIT(pic->pic_id));
> 
> 
> com uses intr-74 and gpio uses intr between 29 and 34, both IPL_HIGH
> (IPL_SERIAL).  If gpio interrupts in comintr(intr 74), pic_pending_pics
> is cleared, and ipending not cleared.
> 
> 
> Index: pic.c
> ===================================================================
> RCS file: /cvsroot/src/sys/arch/arm/pic/pic.c,v
> retrieving revision 1.4
> diff -u -r1.4 pic.c
> --- pic.c       30 Dec 2008 05:43:14 -0000      1.4
> +++ pic.c       30 Aug 2010 13:29:23 -0000
> @@ -196,7 +196,7 @@
>  #endif
>         uint32_t pending_irqs;
>         uint32_t blocked_irqs;
> -       int irq;
> +       int irq, dispatched;
>         bool progress = false;
>         
>         KASSERT(pic->pic_pending_ipls & ipl_mask);
> @@ -206,6 +206,7 @@
>         irq_count = 0;
>  #endif
>  
> +       dispatched = 0;
>         for (;;) {
>                 pending_irqs = pic_find_pending_irqs_by_ipl(pic, irq_base,
>                     *ipending, ipl);
> @@ -214,14 +215,19 @@
>                 if (pending_irqs == 0) {
>  #if PIC_MAXSOURCES > 32
>                         irq_count += 32;
> -                       if (__predict_true(irq_count >= pic->pic_maxsources))
> -                               break;
> -                       irq_base += 32;
> -                       ipending++;
> -                       iblocked++;
> -                       if (irq_base >= pic->pic_maxsources) {
> +                       if (__predict_true(irq_count >= pic->pic_maxsources)) 
> {
> +                               if (!dispatched)
> +                                       break;
> +                               irq_base = 0;
> +                               irq_count = 0;
> +                               dispatched = 0;
>                                 ipending = pic->pic_pending_irqs;
>                                 iblocked = pic->pic_blocked_irqs;
> +                       } else {
> +                               irq_base += 32;
> +                               ipending++;
> +                               iblocked++;
> +                               KASSERT(irq_base <= pic->pic_maxsources);
>                         }
>                         continue;
>  #else
> @@ -251,6 +257,7 @@
>                         atomic_or_32(iblocked, blocked_irqs);
>                         atomic_or_32(&pic_blocked_pics, __BIT(pic->pic_id));
>                 }
> +               dispatched = 1;
>         }
>  
>         KASSERT(progress);
> 
> 
> In my case, this fix clear that problem.  ;-)
> 
> Thanks,
> --
> kiyohara

yes it does fix the hangup with my dme driver port
will you commit?

-- 
NetBSD - Simplicity is prerequisite for reliability