[PATCH 1/2] rt/threadirqs: don't need to save irqs in do_hardirq()

[PATCH 1/2] rt/threadirqs: don't need to save irqs in do_hardirq()

[Frederic Weisbecker]

do_hardirq() has only one caller do_irqd() in a path where irq are already
disabled. So we don't need to save irqs while holding desc->lock

Replace spin_lock_irqsave by spin_lock.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
---
 kernel/irq/manage.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
index ed7c5e3..6e9baf8 100644
--- a/kernel/irq/manage.c
+++ b/kernel/irq/manage.c
@@ -905,7 +905,7 @@ static void do_hardirq(struct irq_desc *desc)
 {
 	unsigned long flags;
 
-	spin_lock_irqsave(&desc->lock, flags);
+	spin_lock(&desc->lock);
 
 	if (!(desc->status & IRQ_INPROGRESS))
 		goto out;
@@ -921,7 +921,7 @@ static void do_hardirq(struct irq_desc *desc)
 	else
 		thread_do_irq(desc);
  out:
-	spin_unlock_irqrestore(&desc->lock, flags);
+	spin_unlock(&desc->lock);
 
 	if (waitqueue_active(&desc->wait_for_handler))
 		wake_up(&desc->wait_for_handler);
-- 
1.6.1

Re: [PATCH 1/2] rt/threadirqs: don't need to save irqs in do_hardirq()

[Arnaldo Carvalho de Melo]

Em Thu, Feb 12, 2009 at 07:27:08PM +0100, Frederic Weisbecker escreveu:
> do_hardirq() has only one caller do_irqd() in a path where irq are already
> disabled. So we don't need to save irqs while holding desc->lock
> 
> Replace spin_lock_irqsave by spin_lock.
> 
> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
> ---
>  kernel/irq/manage.c |    4 ++--
>  1 files changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
> index ed7c5e3..6e9baf8 100644
> --- a/kernel/irq/manage.c
> +++ b/kernel/irq/manage.c
> @@ -905,7 +905,7 @@ static void do_hardirq(struct irq_desc *desc)
>  {
>  	unsigned long flags;

Can flags be removed too?
  
> -	spin_lock_irqsave(&desc->lock, flags);
> +	spin_lock(&desc->lock);
>  
>  	if (!(desc->status & IRQ_INPROGRESS))
>  		goto out;
> @@ -921,7 +921,7 @@ static void do_hardirq(struct irq_desc *desc)
>  	else
>  		thread_do_irq(desc);
>   out:
> -	spin_unlock_irqrestore(&desc->lock, flags);
> +	spin_unlock(&desc->lock);
>  
>  	if (waitqueue_active(&desc->wait_for_handler))
>  		wake_up(&desc->wait_for_handler);

Re: [PATCH 1/2] rt/threadirqs: don't need to save irqs in do_hardirq()

[Frederic Weisbecker]

On Thu, Feb 12, 2009 at 09:24:59PM -0200, Arnaldo Carvalho de Melo wrote:
> Em Thu, Feb 12, 2009 at 07:27:08PM +0100, Frederic Weisbecker escreveu:
> > do_hardirq() has only one caller do_irqd() in a path where irq are already
> > disabled. So we don't need to save irqs while holding desc->lock
> > 
> > Replace spin_lock_irqsave by spin_lock.
> > 
> > Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
> > ---
> >  kernel/irq/manage.c |    4 ++--
> >  1 files changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
> > index ed7c5e3..6e9baf8 100644
> > --- a/kernel/irq/manage.c
> > +++ b/kernel/irq/manage.c
> > @@ -905,7 +905,7 @@ static void do_hardirq(struct irq_desc *desc)
> >  {
> >  	unsigned long flags;
> 
> Can flags be removed too?


Actually... I cheated.
Once the two patches were done, I saw the unused variable warning. So I remade the second patch
to remove flags but..yes it should be on the first patch, I must confess...
   
> > -	spin_lock_irqsave(&desc->lock, flags);
> > +	spin_lock(&desc->lock);
> >  
> >  	if (!(desc->status & IRQ_INPROGRESS))
> >  		goto out;
> > @@ -921,7 +921,7 @@ static void do_hardirq(struct irq_desc *desc)
> >  	else
> >  		thread_do_irq(desc);
> >   out:
> > -	spin_unlock_irqrestore(&desc->lock, flags);
> > +	spin_unlock(&desc->lock);
> >  
> >  	if (waitqueue_active(&desc->wait_for_handler))
> >  		wake_up(&desc->wait_for_handler);

Re: [PATCH 1/2] rt/threadirqs: don't need to save irqs in do_hardirq()

[Steven Rostedt]

On Fri, 13 Feb 2009, Frederic Weisbecker wrote:

> On Thu, Feb 12, 2009 at 09:24:59PM -0200, Arnaldo Carvalho de Melo wrote:
> > Em Thu, Feb 12, 2009 at 07:27:08PM +0100, Frederic Weisbecker escreveu:
> > > do_hardirq() has only one caller do_irqd() in a path where irq are already
> > > disabled. So we don't need to save irqs while holding desc->lock
> > > 
> > > Replace spin_lock_irqsave by spin_lock.
> > > 
> > > Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
> > > ---
> > >  kernel/irq/manage.c |    4 ++--
> > >  1 files changed, 2 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
> > > index ed7c5e3..6e9baf8 100644
> > > --- a/kernel/irq/manage.c
> > > +++ b/kernel/irq/manage.c
> > > @@ -905,7 +905,7 @@ static void do_hardirq(struct irq_desc *desc)
> > >  {
> > >  	unsigned long flags;
> > 
> > Can flags be removed too?
> 
> 
> Actually... I cheated.
> Once the two patches were done, I saw the unused variable warning. So I remade the second patch
> to remove flags but..yes it should be on the first patch, I must confess...

That's fine, I'll pull both of them in.

Thanks Frederic!

-- Steve

Re: [PATCH 1/2] rt/threadirqs: don't need to save irqs in do_hardirq()

[Arnaldo Carvalho de Melo]

Em Thu, Feb 12, 2009 at 06:37:16PM -0500, Steven Rostedt escreveu:
> 
> On Fri, 13 Feb 2009, Frederic Weisbecker wrote:
> 
> > On Thu, Feb 12, 2009 at 09:24:59PM -0200, Arnaldo Carvalho de Melo wrote:
> > > Em Thu, Feb 12, 2009 at 07:27:08PM +0100, Frederic Weisbecker escreveu:
> > > > do_hardirq() has only one caller do_irqd() in a path where irq are already
> > > > disabled. So we don't need to save irqs while holding desc->lock
> > > > 
> > > > Replace spin_lock_irqsave by spin_lock.
> > > > 
> > > > Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
> > > > ---
> > > >  kernel/irq/manage.c |    4 ++--
> > > >  1 files changed, 2 insertions(+), 2 deletions(-)
> > > > 
> > > > diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
> > > > index ed7c5e3..6e9baf8 100644
> > > > --- a/kernel/irq/manage.c
> > > > +++ b/kernel/irq/manage.c
> > > > @@ -905,7 +905,7 @@ static void do_hardirq(struct irq_desc *desc)
> > > >  {
> > > >  	unsigned long flags;
> > > 
> > > Can flags be removed too?
> > 
> > 
> > Actually... I cheated.
> > Once the two patches were done, I saw the unused variable warning. So I remade the second patch
> > to remove flags but..yes it should be on the first patch, I must confess...
> 
> That's fine, I'll pull both of them in.
> 
> Thanks Frederic!

Ditto, that was just a follow up thingy, nothing against the change :-)

- Arnaldo