diff mbox

irq: crossbar: improve allocate_free_irq() complexity

Message ID 1396392259-27437-1-git-send-email-balbi@ti.com (mailing list archive)
State New, archived
Headers show

Commit Message

Felipe Balbi April 1, 2014, 10:44 p.m. UTC
current algorithm in allocate_free_irq() is O(n),
by just keeping track of last allocated IRQ with a
simple unsigned integer, we can find a free IRQ
in O(1).

Signed-off-by: Felipe Balbi <balbi@ti.com>
---

compile-tested only as J6 DTS is currently missing crossbar
altogether :-(

There's a drawback with this patch which I'm not sure if we
should care a lot because I couldn't entirely grasp when is
domain->xlate() called and if we will map/unmap IRQs in runtime
or will this *always* be done only during boot.

If we're talking about runtime IRQ remapping, then this, clearly,
won't work. But if this will be done only during boot up, then we
avoid iterating over the irq_map array each time we try to translate
a new IRQ prior to mapping it.

Comments are highly welcome as I'll probably learn something new
about the IRQ subsystem ;-)

 drivers/irqchip/irq-crossbar.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

Comments

Felipe Balbi April 1, 2014, 10:46 p.m. UTC | #1
On Tue, Apr 01, 2014 at 05:44:19PM -0500, Felipe Balbi wrote:
> current algorithm in allocate_free_irq() is O(n),
> by just keeping track of last allocated IRQ with a
> simple unsigned integer, we can find a free IRQ
> in O(1).
> 
> Signed-off-by: Felipe Balbi <balbi@ti.com>
> ---
> 
> compile-tested only as J6 DTS is currently missing crossbar
> altogether :-(
> 
> There's a drawback with this patch which I'm not sure if we
> should care a lot because I couldn't entirely grasp when is
> domain->xlate() called and if we will map/unmap IRQs in runtime
> or will this *always* be done only during boot.
> 
> If we're talking about runtime IRQ remapping, then this, clearly,
> won't work. But if this will be done only during boot up, then we
> avoid iterating over the irq_map array each time we try to translate
> a new IRQ prior to mapping it.
> 
> Comments are highly welcome as I'll probably learn something new
> about the IRQ subsystem ;-)
> 
>  drivers/irqchip/irq-crossbar.c | 16 ++++++++--------
>  1 file changed, 8 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/irqchip/irq-crossbar.c b/drivers/irqchip/irq-crossbar.c
> index fc817d2..1c4da5a 100644
> --- a/drivers/irqchip/irq-crossbar.c
> +++ b/drivers/irqchip/irq-crossbar.c
> @@ -31,6 +31,7 @@ struct crossbar_device {
>  	void __iomem *crossbar_base;
>  	int *register_offsets;
>  	void (*write) (int, int);
> +	unsigned int current_irq;
>  };
>  
>  static struct crossbar_device *cb;
> @@ -52,16 +53,15 @@ static inline void crossbar_writeb(int irq_no, int cb_no)
>  
>  static inline int allocate_free_irq(int cb_no)
>  {
> -	int i;
> +	int current_irq;
>  
> -	for (i = 0; i < cb->int_max; i++) {
> -		if (cb->irq_map[i] == IRQ_FREE) {
> -			cb->irq_map[i] = cb_no;
> -			return i;
> -		}
> -	}
> +	if (cb->current_irq == cb->int_max)
> +		return -ENODEV;
> +
> +	current_irq = cb->current_irq++;
> +	cb->irq_map[current_irq++] = cb_no;

this increment is bogus, thought I had committed already, please ignore.
Felipe Balbi April 10, 2014, 9:55 p.m. UTC | #2
On Tue, Apr 01, 2014 at 05:44:19PM -0500, Felipe Balbi wrote:
> current algorithm in allocate_free_irq() is O(n),
> by just keeping track of last allocated IRQ with a
> simple unsigned integer, we can find a free IRQ
> in O(1).
> 
> Signed-off-by: Felipe Balbi <balbi@ti.com>
> ---
> 
> compile-tested only as J6 DTS is currently missing crossbar
> altogether :-(
> 
> There's a drawback with this patch which I'm not sure if we
> should care a lot because I couldn't entirely grasp when is
> domain->xlate() called and if we will map/unmap IRQs in runtime
> or will this *always* be done only during boot.
> 
> If we're talking about runtime IRQ remapping, then this, clearly,
> won't work. But if this will be done only during boot up, then we
> avoid iterating over the irq_map array each time we try to translate
> a new IRQ prior to mapping it.
> 
> Comments are highly welcome as I'll probably learn something new
> about the IRQ subsystem ;-)

do not apply this one!!! It won't work in all cases.
diff mbox

Patch

diff --git a/drivers/irqchip/irq-crossbar.c b/drivers/irqchip/irq-crossbar.c
index fc817d2..1c4da5a 100644
--- a/drivers/irqchip/irq-crossbar.c
+++ b/drivers/irqchip/irq-crossbar.c
@@ -31,6 +31,7 @@  struct crossbar_device {
 	void __iomem *crossbar_base;
 	int *register_offsets;
 	void (*write) (int, int);
+	unsigned int current_irq;
 };
 
 static struct crossbar_device *cb;
@@ -52,16 +53,15 @@  static inline void crossbar_writeb(int irq_no, int cb_no)
 
 static inline int allocate_free_irq(int cb_no)
 {
-	int i;
+	int current_irq;
 
-	for (i = 0; i < cb->int_max; i++) {
-		if (cb->irq_map[i] == IRQ_FREE) {
-			cb->irq_map[i] = cb_no;
-			return i;
-		}
-	}
+	if (cb->current_irq == cb->int_max)
+		return -ENODEV;
+
+	current_irq = cb->current_irq++;
+	cb->irq_map[current_irq++] = cb_no;
 
-	return -ENODEV;
+	return current_irq;
 }
 
 static int crossbar_domain_map(struct irq_domain *d, unsigned int irq,