pcmcia's use of IRQ_NOAUTOEN

pcmcia's use of IRQ_NOAUTOEN

[Jan Beulich]

Commit 635416ef393e8cec5a89fc6c1de710ee9596a51e introduced
this, but I can't see how it can take effect - the argument passed to
request_irq() only affects action->flags (IRQF_*), whereas the flag
in question is a desc->status one (IRQ_*). Am I overlooking
something? If not, while currently this just should not work as expected,
it's latently more significant (as soon as the IRQF_* definitions make
use of the bit used by IRQ_NOAUTOEN).

Thanks, Jan

pcmcia's use of IRQ_NOAUTOEN

[Jan Beulich]

(re-sending to correct Alan's email address)

Commit 635416ef393e8cec5a89fc6c1de710ee9596a51e introduced
this, but I can't see how it can take effect - the argument passed to
request_irq() only affects action->flags (IRQF_*), whereas the flag
in question is a desc->status one (IRQ_*). Am I overlooking
something? If not, while currently this just should not work as expected,
it's latently more significant (as soon as the IRQF_* definitions make
use of the bit used by IRQ_NOAUTOEN).

Thanks, Jan

Re: pcmcia's use of IRQ_NOAUTOEN

[Dominik Brodowski]

Hey,

On Mon, Mar 08, 2010 at 10:17:11AM +0000, Jan Beulich wrote:
> (re-sending to correct Alan's email address)
> 
> Commit 635416ef393e8cec5a89fc6c1de710ee9596a51e introduced
> this, but I can't see how it can take effect - the argument passed to
> request_irq() only affects action->flags (IRQF_*), whereas the flag
> in question is a desc->status one (IRQ_*). Am I overlooking
> something? If not, while currently this just should not work as expected,
> it's latently more significant (as soon as the IRQF_* definitions make
> use of the bit used by IRQ_NOAUTOEN).

You're right, it seems to make no sense. Or less than that.

Alan, what do you think?

Best,
	Dominik

Re: pcmcia's use of IRQ_NOAUTOEN

[Alan Cox]

On Mon, 8 Mar 2010 20:47:57 +0100
Dominik Brodowski <linux@dominikbrodowski.net> wrote:

> Hey,
> 
> On Mon, Mar 08, 2010 at 10:17:11AM +0000, Jan Beulich wrote:
> > (re-sending to correct Alan's email address)
> > 
> > Commit 635416ef393e8cec5a89fc6c1de710ee9596a51e introduced
> > this, but I can't see how it can take effect - the argument passed
> > to request_irq() only affects action->flags (IRQF_*), whereas the
> > flag in question is a desc->status one (IRQ_*). Am I overlooking
> > something? If not, while currently this just should not work as
> > expected, it's latently more significant (as soon as the IRQF_*
> > definitions make use of the bit used by IRQ_NOAUTOEN).
> 
> You're right, it seems to make no sense. Or less than that.
> 
> Alan, what do you think?

It should be setting the correct NOAUTOEN bits so that the IRQ isn't
randomly enabled as some hardware then crashes. How it ended up wrong
upstream I don't know as it was tested and fixed the bug, so presumably
the right version got tested or it happened to work back on the old irq
code by luck.

I don't think you are overlooking anything - there is no deep magic
involved.

Alan

Re: pcmcia's use of IRQ_NOAUTOEN

[Dominik Brodowski]

Hey,

On Mon, Mar 15, 2010 at 05:43:59PM +0000, Alan Cox wrote:
> > You're right, it seems to make no sense. Or less than that.
> > 
> > Alan, what do you think?
> 
> It should be setting the correct NOAUTOEN bits so that the IRQ isn't
> randomly enabled as some hardware then crashes. How it ended up wrong
> upstream I don't know as it was tested and fixed the bug, so presumably
> the right version got tested or it happened to work back on the old irq
> code by luck.

http://lkml.org/lkml/2007/4/5/127 and then
http://lkml.org/lkml/2007/4/26/485 talk a different language :(

> I don't think you are overlooking anything - there is no deep magic
> involved.

I'll just revert it.

Best,
	Dominik