On Wed, 17 Apr 2013 14:48:19 +0800, Mike <qiudayu@linux.vnet.ibm.com> wrote:
> 在 2012-12-11二的 08:30 +0000，Grant Likely写道：
> > On Tue, 27 Nov 2012 09:41:46 +0800, Mike <qiudayu@linux.vnet.ibm.com> wrote:
> > > 在 2012-11-26一的 20:17 +0000，Grant Likely写道：
> > > > On Mon, 24 Sep 2012 17:37:55 +0800, Mike Qiu <qiudayu@linux.vnet.ibm.com> wrote:
> > > > > Function irq_create_mapping() calls irq_find_mapping(). The later
> > > > > function has checked if the indicated IRQ domain has hw IRQ mapped to
> > > > > virtual IRQ through legacy mode or not and return the value of the
> > > > > legacy irq number by call irq_domain_legacy_revmap(). We needn't
> > > > > to call irq_domain_legacy_revmap() to do same check in
> > > > > irq_create_mapping() again.
> > > > > 
> > > > > The patch removes the duplicate call.
> > > > > 
> > > > > Signed-off-by: Mike Qiu <qiudayu@linux.vnet.ibm.com>
> > > > > ---
> > > > >  kernel/irq/irqdomain.c |    7 +++++--
> > > > >  1 files changed, 5 insertions(+), 2 deletions(-)
> > > > > 
> > > > > diff --git a/kernel/irq/irqdomain.c b/kernel/irq/irqdomain.c
> > > > > index 49a7772..286d672 100644
> > > > > --- a/kernel/irq/irqdomain.c
> > > > > +++ b/kernel/irq/irqdomain.c
> > > > > @@ -547,9 +547,12 @@ unsigned int irq_create_mapping(struct irq_domain *domain,
> > > > >  		return virq;
> > > > >  	}
> > > > >  
> > > > > -	/* Get a virtual interrupt number */
> > > > > +	/*
> > > > > +	 * For IRQ domain with type of IRQ_DOMAIN_MAP_LEGACY, we needn't
> > > > > +	 * create the IRQ mapping for non-existing one, so just return 0.
> > > > > +	 */
> > > > >  	if (domain->revmap_type == IRQ_DOMAIN_MAP_LEGACY)
> > > > > -		return irq_domain_legacy_revmap(domain, hwirq);
> > > > > +		return 0;
> > > > 
> > > > But it does need to return the virq assigned to the hwirq. That is why
> > > > it has to call the revmap function.
> > > > 
> > > Yes, thanks
> > > 
> > > this judgment has been done in
> > >  /*Check if mapping already exists*/
> > > virq = irq_find_mapping(domain, hwirq);
> > > 
> > > if the virq equals none zero, the func irq_create_mapping will 
> > > return the virq value directly(already exists).
> > > 
> > > otherwise, that means virq equals zero, this has two meanings, one is
> > > haven't find the already exist mapping, the other one is in
> > > irq_find_mapping()
> > > 	case IRQ_DOMAIN_MAP_LEGACY:
> > > 		return irq_domain_legacy_revmap(domain, hwirq); 
> > > this may return zero.
> > > 
> > > So, when we check if it is a IRQ_DOMAIN_MAP_LEGACY, we just return zero
> > > is OK. because if it is none zero, it will be return after check if
> > > mapping already exists, and never come here. also we never try to assign
> > > the virq for the legacy map.
> > > 
> > > I don't know if you have understand my explanation.
> > 
> > Yes, your explanation makes sense and you are correct. I'm not going to
> > apply this for v3.8 (I've left it too late before the merge window), but
> > I'll look again after the merge window closes.
> > 
> Hi Grant 
> 
> I don't know if you forget this patch ...
> 
> I haven't see this patch in 3.9-rc7
> 
> just for a reminder

After taking another look at it, I'm not going to apply this patch. It
isn't actively dangerous for the revmap to get called twice, and I've
got a patch pending that removes the legacy map entirely and replaces it
with a pre-populated linear map.

g.