Bug in "genirq: record trigger type"

Bug in "genirq: record trigger type"

[Benjamin Herrenschmidt]

On Mon, 2008-10-20 at 22:05 +0000, Linux Kernel Mailing List wrote:
> Gitweb:     http://git.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=0c5d1eb77a8be917b638344a22afe1398236482b
> Commit:     0c5d1eb77a8be917b638344a22afe1398236482b
> Parent:     d6d5aeb661fc14655c417f3582ae7ec52985d2a8
> Author:     David Brownell <dbrownell@users.sourceforge.net>
> AuthorDate: Wed Oct 1 14:46:18 2008 -0700
> Committer:  Ingo Molnar <mingo@elte.hu>
> CommitDate: Thu Oct 2 10:24:09 2008 +0200

This one is obviously broken and breaks booting on a whole bunch of
machines (including powermac's and thus my G5, it's never good when my
own machine breaks !).

Nice to see 3 SOB's and one Ack and nobody caught the obvious bug :-)

>  	desc = irq_desc + irq;
> -	if (desc->chip->set_type) {
> -		spin_lock_irqsave(&desc->lock, flags);
> -		ret = desc->chip->set_type(irq, type);
> -		spin_unlock_irqrestore(&desc->lock, flags);
> -	}
> +	if (type == IRQ_TYPE_NONE)
> +		return 0;
> +
> +	spin_lock_irqsave(&desc->lock, flags);
> +	ret = __irq_set_trigger(desc, irq, flags);
                                             ^^^^ type maybe ?

> +	spin_unlock_irqrestore(&desc->lock, flags);
>  	return ret;
>  }

I have to run so no patch until tomorrow unless somebody beats me to it.

Cheers,
Ben.

Re: Bug in "genirq: record trigger type"

[Yinghai Lu]

On Mon, Oct 20, 2008 at 11:32 PM, Benjamin Herrenschmidt
<benh@kernel.crashing.org> wrote:
> On Mon, 2008-10-20 at 22:05 +0000, Linux Kernel Mailing List wrote:
>> Gitweb:     http://git.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=0c5d1eb77a8be917b638344a22afe1398236482b
>> Commit:     0c5d1eb77a8be917b638344a22afe1398236482b
>> Parent:     d6d5aeb661fc14655c417f3582ae7ec52985d2a8
>> Author:     David Brownell <dbrownell@users.sourceforge.net>
>> AuthorDate: Wed Oct 1 14:46:18 2008 -0700
>> Committer:  Ingo Molnar <mingo@elte.hu>
>> CommitDate: Thu Oct 2 10:24:09 2008 +0200
>
> This one is obviously broken and breaks booting on a whole bunch of
> machines (including powermac's and thus my G5, it's never good when my
> own machine breaks !).
>
> Nice to see 3 SOB's and one Ack and nobody caught the obvious bug :-)
>
>>       desc = irq_desc + irq;
>> -     if (desc->chip->set_type) {
>> -             spin_lock_irqsave(&desc->lock, flags);
>> -             ret = desc->chip->set_type(irq, type);
>> -             spin_unlock_irqrestore(&desc->lock, flags);
>> -     }
>> +     if (type == IRQ_TYPE_NONE)
>> +             return 0;
>> +
>> +     spin_lock_irqsave(&desc->lock, flags);
>> +     ret = __irq_set_trigger(desc, irq, flags);
>                                             ^^^^ type maybe ?
>
>> +     spin_unlock_irqrestore(&desc->lock, flags);
>>       return ret;
>>  }
>
> I have to run so no patch until tomorrow unless somebody beats me to it.

there is patch about it, but somehow get lost.

YH

[PATCH] genirq: fix set_irq_type() when recording trigger type

[Ingo Molnar]

* Benjamin Herrenschmidt <benh@kernel.crashing.org> wrote:

> On Mon, 2008-10-20 at 22:05 +0000, Linux Kernel Mailing List wrote:
> > Gitweb:     http://git.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=0c5d1eb77a8be917b638344a22afe1398236482b
> > Commit:     0c5d1eb77a8be917b638344a22afe1398236482b
> > Parent:     d6d5aeb661fc14655c417f3582ae7ec52985d2a8
> > Author:     David Brownell <dbrownell@users.sourceforge.net>
> > AuthorDate: Wed Oct 1 14:46:18 2008 -0700
> > Committer:  Ingo Molnar <mingo@elte.hu>
> > CommitDate: Thu Oct 2 10:24:09 2008 +0200
[...]
> > Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
> > Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
> > Acked-by: Thomas Gleixner <tglx@linutronix.de>
> > Signed-off-by: Ingo Molnar <mingo@elte.hu>
> 
> This one is obviously broken and breaks booting on a whole bunch of 
> machines (including powermac's and thus my G5, it's never good when my 
> own machine breaks !).
> 
> Nice to see 3 SOB's and one Ack and nobody caught the obvious bug :-)

that patch came from -mm towards us shortly before the merge window. 
Once we had it integrated Chris Friesen reported a boot hang on a G5, 
and there's a fix pending for the bug, see it below.

Will send it to Linus expressly. Fortunately only a minority of Linux 
users will ever run a box that uses set_irq_type() - but yes it needs to 
be fixed.

	Ingo

------------------------------>
>From aac4ddd11a8d0e402ddc3fbc75204cb64efa0aac Mon Sep 17 00:00:00 2001
From: Chris Friesen <cfriesen@nortel.com>
Date: Mon, 20 Oct 2008 12:41:58 -0600
Subject: [PATCH] genirq: fix set_irq_type() when recording trigger type

In set_irq_type() we want to pass the type rather than the current
interrupt state.

Signed-off-by: Chris Friesen <cfriesen@nortel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 kernel/irq/chip.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/kernel/irq/chip.c b/kernel/irq/chip.c
index 4895fde..3de6ea3 100644
--- a/kernel/irq/chip.c
+++ b/kernel/irq/chip.c
@@ -127,7 +127,7 @@ int set_irq_type(unsigned int irq, unsigned int type)
 		return 0;
 
 	spin_lock_irqsave(&desc->lock, flags);
-	ret = __irq_set_trigger(desc, irq, flags);
+	ret = __irq_set_trigger(desc, irq, type);
 	spin_unlock_irqrestore(&desc->lock, flags);
 	return ret;
 }

Re: Bug in "genirq: record trigger type"

[Ingo Molnar]

* Yinghai Lu <yinghai@kernel.org> wrote:

> On Mon, Oct 20, 2008 at 11:32 PM, Benjamin Herrenschmidt
> <benh@kernel.crashing.org> wrote:
> > On Mon, 2008-10-20 at 22:05 +0000, Linux Kernel Mailing List wrote:
> >> Gitweb:     http://git.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=0c5d1eb77a8be917b638344a22afe1398236482b
> >> Commit:     0c5d1eb77a8be917b638344a22afe1398236482b
> >> Parent:     d6d5aeb661fc14655c417f3582ae7ec52985d2a8
> >> Author:     David Brownell <dbrownell@users.sourceforge.net>
> >> AuthorDate: Wed Oct 1 14:46:18 2008 -0700
> >> Committer:  Ingo Molnar <mingo@elte.hu>
> >> CommitDate: Thu Oct 2 10:24:09 2008 +0200
> >
> > This one is obviously broken and breaks booting on a whole bunch of
> > machines (including powermac's and thus my G5, it's never good when my
> > own machine breaks !).
> >
> > Nice to see 3 SOB's and one Ack and nobody caught the obvious bug :-)
> >
> >>       desc = irq_desc + irq;
> >> -     if (desc->chip->set_type) {
> >> -             spin_lock_irqsave(&desc->lock, flags);
> >> -             ret = desc->chip->set_type(irq, type);
> >> -             spin_unlock_irqrestore(&desc->lock, flags);
> >> -     }
> >> +     if (type == IRQ_TYPE_NONE)
> >> +             return 0;
> >> +
> >> +     spin_lock_irqsave(&desc->lock, flags);
> >> +     ret = __irq_set_trigger(desc, irq, flags);
> >                                             ^^^^ type maybe ?
> >
> >> +     spin_unlock_irqrestore(&desc->lock, flags);
> >>       return ret;
> >>  }
> >
> > I have to run so no patch until tomorrow unless somebody beats me to it.
> 
> there is patch about it, but somehow get lost.

should be all sorted now, the fix is below.

	Ingo

------------------>
>From aac4ddd11a8d0e402ddc3fbc75204cb64efa0aac Mon Sep 17 00:00:00 2001
From: Chris Friesen <cfriesen@nortel.com>
Date: Mon, 20 Oct 2008 12:41:58 -0600
Subject: [PATCH] genirq: fix set_irq_type() when recording trigger type

In set_irq_type() we want to pass the type rather than the current
interrupt state.

Signed-off-by: Chris Friesen <cfriesen@nortel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 kernel/irq/chip.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/kernel/irq/chip.c b/kernel/irq/chip.c
index 4895fde..3de6ea3 100644
--- a/kernel/irq/chip.c
+++ b/kernel/irq/chip.c
@@ -127,7 +127,7 @@ int set_irq_type(unsigned int irq, unsigned int type)
 		return 0;
 
 	spin_lock_irqsave(&desc->lock, flags);
-	ret = __irq_set_trigger(desc, irq, flags);
+	ret = __irq_set_trigger(desc, irq, type);
 	spin_unlock_irqrestore(&desc->lock, flags);
 	return ret;
 }

Re: Bug in "genirq: record trigger type"

[Benjamin Herrenschmidt]

On Tue, 2008-10-21 at 09:28 +0200, Ingo Molnar wrote:

> From: Chris Friesen <cfriesen@nortel.com>
> Date: Mon, 20 Oct 2008 12:41:58 -0600
> Subject: [PATCH] genirq: fix set_irq_type() when recording trigger type
> 
> In set_irq_type() we want to pass the type rather than the current
> interrupt state.
> 
> Signed-off-by: Chris Friesen <cfriesen@nortel.com>
> Signed-off-by: Ingo Molnar <mingo@elte.hu>

Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

> ---
>  kernel/irq/chip.c |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)
> 
> diff --git a/kernel/irq/chip.c b/kernel/irq/chip.c
> index 4895fde..3de6ea3 100644
> --- a/kernel/irq/chip.c
> +++ b/kernel/irq/chip.c
> @@ -127,7 +127,7 @@ int set_irq_type(unsigned int irq, unsigned int type)
>  		return 0;
>  
>  	spin_lock_irqsave(&desc->lock, flags);
> -	ret = __irq_set_trigger(desc, irq, flags);
> +	ret = __irq_set_trigger(desc, irq, type);
>  	spin_unlock_irqrestore(&desc->lock, flags);
>  	return ret;
>  }

Re: Bug in "genirq: record trigger type"

[Ingo Molnar]

* Benjamin Herrenschmidt <benh@kernel.crashing.org> wrote:

> On Tue, 2008-10-21 at 09:28 +0200, Ingo Molnar wrote:
> 
> > From: Chris Friesen <cfriesen@nortel.com>
> > Date: Mon, 20 Oct 2008 12:41:58 -0600
> > Subject: [PATCH] genirq: fix set_irq_type() when recording trigger type
> > 
> > In set_irq_type() we want to pass the type rather than the current
> > interrupt state.
> > 
> > Signed-off-by: Chris Friesen <cfriesen@nortel.com>
> > Signed-off-by: Ingo Molnar <mingo@elte.hu>
> 
> Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

thx, added your ack to the commit as well.

	Ingo

Re: Bug in "genirq: record trigger type"

[David Brownell]

On Monday 20 October 2008, Benjamin Herrenschmidt wrote:
> This one is obviously broken and breaks booting on a whole bunch of
> machines (including powermac's and thus my G5, it's never good when my
> own machine breaks !).
> 
> Nice to see 3 SOB's and one Ack and nobody caught the obvious bug :-)

As you saw, that one's fixed.  Chris' patch unfortunately didn't
get integrated right away.


I'm a bit more curious about another potential issue though ... as
described in the patch comment:

     - Make set_irq_type() usage match request_irq() usage:
        * IRQ_TYPE_NONE should be a NOP; succeed, so irq_chip methods
          won't have to handle that case any more (many do it wrong).

It might be a bit more accurate to say irq_chip.set_type() methods
are *inconsistent* in handling IRQ_TYPE_NONE.  Previously the
set_irq_type() method would pass that down to irq_chip code. 

I had observed two behaviors, but I thought I observed a third one
in some of the PowerPC code:

 (1) ignore it ... matching request_irq() usage
 (2) return an error ... nasty
 (3) assign some irq_chip-specific trigger mode

That third behavior might cause a bit of trouble, but I think
it was only used during platform init.  Someone more attuned
to PowerPC might want to check ...

- Dave

Re: Bug in "genirq: record trigger type"

[Benjamin Herrenschmidt]

On Tue, 2008-10-21 at 01:01 -0700, David Brownell wrote:
> 
> I'm a bit more curious about another potential issue though ... as
> described in the patch comment:
> 
>      - Make set_irq_type() usage match request_irq() usage:
>         * IRQ_TYPE_NONE should be a NOP; succeed, so irq_chip methods
>           won't have to handle that case any more (many do it wrong).
> 
> It might be a bit more accurate to say irq_chip.set_type() methods
> are *inconsistent* in handling IRQ_TYPE_NONE.  Previously the
> set_irq_type() method would pass that down to irq_chip code. 
> 
> I had observed two behaviors, but I thought I observed a third one
> in some of the PowerPC code:
> 
>  (1) ignore it ... matching request_irq() usage
>  (2) return an error ... nasty
>  (3) assign some irq_chip-specific trigger mode
> 
> That third behavior might cause a bit of trouble, but I think
> it was only used during platform init.  Someone more attuned
> to PowerPC might want to check ...

Ok, I wrote a lot of the port of powerpc stuff to genirq so I supposed
I'm the person to do that sweep :-) I'll have a look tomorrow.

Thanks for the heads up,

Cheers,
Ben.