Re: [patch 1/6] hardirq: Make hardirq bits generic

From: Thomas Gleixner
Date: Wed Nov 06 2013 - 12:23:33 EST


Geert,

On Mon, 4 Nov 2013, Geert Uytterhoeven wrote:
> But only if you look at "[m68k] IRQ: add handle_polled_irq() for timer
> based soft interrupt" (http://www.spinics.net/lists/linux-m68k/msg05889.html)
> first ;-)

Done. Thanks for the reminder!

> Below is a patch with some fixups, on top of your two patches.
>
> Unfortunately it still hangs somewhere after mounting the root filesystem.
>
> Using this debug code for do_IRQ():
>
> diff --git a/arch/m68k/kernel/irq.c b/arch/m68k/kernel/irq.c
> index aaf7b15fad41..da9687803d98 100644
> --- a/arch/m68k/kernel/irq.c
> +++ b/arch/m68k/kernel/irq.c
> @@ -22,11 +22,21 @@ asmlinkage int do_IRQ(int irq, struct pt_regs *regs)
> struct pt_regs *oldregs = set_irq_regs(regs);
> int nested = regs->sr & ~ALLOWINT;
>
> +static int nesting;
> +const char prefix[] = " ";
> +unsigned long flags;
> +local_irq_save(flags);
> +nesting++;
> +printk("# %sirq %d nested %d\n", &prefix[16-2*nesting], irq, nested);
> +local_irq_restore(flags);
> irq_enter();
> generic_handle_irq(irq);
> irq_exit_nested(nested);
>
> set_irq_regs(oldregs);
> +local_irq_save(flags);
> +nesting--;
> +local_irq_restore(flags);
> return nested;
> }
>
> I get output like
>
> # irq 15 nested 0
> # irq 15 nested 1024
>
> irq 15 while irq 15 in progress??

Huch, that's odd.

> With similar debug code on the old working do_IRQ(), I get
> - slightly less deep nesting,
> - do_IRQ() is never re-entered with the same irq number.
>
> Also note that the value of "nested" doesn't match the indentation level,
> which depends on my own bookkeeping using "nesting".

Well, nested is just an indicator. It's not the nest level.

nested = pt->sr & ~ALLOWINT;
i.e.:
nested = pt->sr & 0x0700;

So in the case above nested is 0x400

> Anyone with an idea where it's going wrong?

The original code does:

add_preempt_count(HARDIRQ_OFFSET);

do_IRQ()
irq_enter();
add_preempt_count(HARDIRQ_OFFSET);

handle_irq();

irq_exit();
local_irq_disable();
sub_preempt_count(HARDIRQ_OFFSET);

sub_preempt_count(HARDIRQ_OFFSET);

/* Check for nested irq */
if (in_hardirq())
reti();

/* Check for nested irq again */
if (pt->sr & ~ALLOWINT != 0)
reti();

do_softirq();
....
ret_from_exception();

With the patches in place it looks like this:

do_IRQ()
nested = pt->sr & ~ALLOWINT;

irq_enter();
add_preempt_count(HARDIRQ_OFFSET);

handle_irq();

irq_exit_nested(nested);
local_irq_disable();
sub_preempt_count(HARDIRQ_OFFSET);
if (!nested && !in_hardirq())
do_softirq()

return nested;

if (nested)
reti();

ret_from_exception();

So all it does essentially is to move the softirq invocation in the
non nested case a tad earlier. I'm really puzzled as I can't spot the
point where this change makes a real difference.

Thanks,

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