[PATCH] gianfar: irq_of_parse_and_map() error unnoticed

[PATCH] gianfar: irq_of_parse_and_map() error unnoticed

[Roel Kluin]

Not sure which irq_of_parse_and_map() is used, but I found definitions here:

vi arch/microblaze/kernel/irq.c +23
vi arch/powerpc/kernel/irq.c +727
vi arch/sparc/kernel/of_device_32.c +33
vi arch/sparc/kernel/of_device_64.c +59

They either return 0 or NO_IRQ - either defined 0, -1, 255 or INT_MAX.
------------------------------>8-------------8<---------------------------------
priv->interruptTransmit, -Receive and -Error are unsigned, so the error path
wasn't taken when irq_of_parse_and_map() returned an incorrect irq.

Signed-off-by: Roel Kluin <roel.kluin@gmail.com> 
---
diff --git a/drivers/net/gianfar.c b/drivers/net/gianfar.c
index b2c4967..e30d158 100644
--- a/drivers/net/gianfar.c
+++ b/drivers/net/gianfar.c
@@ -203,9 +203,12 @@ static int gfar_of_init(struct net_device *dev)
 
 		priv->interruptError = irq_of_parse_and_map(np, 2);
 
-		if (priv->interruptTransmit < 0 ||
-				priv->interruptReceive < 0 ||
-				priv->interruptError < 0) {
+		if (priv->interruptTransmit == 0 ||
+				priv->interruptTransmit > 254 ||
+				priv->interruptReceive == 0 ||
+				priv->interruptReceive > 254 ||
+				priv->interruptError == 0 ||
+				priv->interruptReceive > 254) {
 			err = -EINVAL;
 			goto err_out;
 		}

Re: [PATCH] gianfar: irq_of_parse_and_map() error unnoticed

[David Miller]

From: Roel Kluin <roel.kluin@gmail.com>
Date: Thu, 23 Apr 2009 14:52:44 +0200

> Not sure which irq_of_parse_and_map() is used, but I found definitions here:
> 
> vi arch/microblaze/kernel/irq.c +23
> vi arch/powerpc/kernel/irq.c +727
> vi arch/sparc/kernel/of_device_32.c +33
> vi arch/sparc/kernel/of_device_64.c +59
> 
> They either return 0 or NO_IRQ - either defined 0, -1, 255 or INT_MAX.
> ------------------------------>8-------------8<---------------------------------
> priv->interruptTransmit, -Receive and -Error are unsigned, so the error path
> wasn't taken when irq_of_parse_and_map() returned an incorrect irq.
> 
> Signed-off-by: Roel Kluin <roel.kluin@gmail.com> 

This test is way too convoluted for my taste.

I would rather that the implementations of this interface
use a consistent value for errors.  Preferrably NO_IRQ.

Not applied.

[PATCH 2/2] gianfar: irq_of_parse_and_map() error unnoticed

[Roel Kluin]

priv->interruptTransmit, -Receive and -Error are unsigned, so the error path
wasn't taken when irq_of_parse_and_map() returned an incorrect irq.

Signed-off-by: Roel Kluin <roel.kluin@gmail.com> 
---
diff --git a/drivers/net/gianfar.c b/drivers/net/gianfar.c
index b2c4967..55c2ce8 100644
--- a/drivers/net/gianfar.c
+++ b/drivers/net/gianfar.c
@@ -203,9 +203,9 @@ static int gfar_of_init(struct net_device *dev)
 
 		priv->interruptError = irq_of_parse_and_map(np, 2);
 
-		if (priv->interruptTransmit < 0 ||
-				priv->interruptReceive < 0 ||
-				priv->interruptError < 0) {
+		if (priv->interruptTransmit == NO_IRQ ||
+				priv->interruptReceive == NO_IRQ ||
+				priv->interruptError == NO_IRQ) {
 			err = -EINVAL;
 			goto err_out;
 		}

Re: [PATCH 2/2] gianfar: irq_of_parse_and_map() error unnoticed

[Andrew Morton]

On Mon, 27 Apr 2009 13:05:30 +0200
Roel Kluin <roel.kluin@gmail.com> wrote:

> priv->interruptTransmit, -Receive and -Error are unsigned, so the error path
> wasn't taken when irq_of_parse_and_map() returned an incorrect irq.
> 
> Signed-off-by: Roel Kluin <roel.kluin@gmail.com> 
> ---
> diff --git a/drivers/net/gianfar.c b/drivers/net/gianfar.c
> index b2c4967..55c2ce8 100644
> --- a/drivers/net/gianfar.c
> +++ b/drivers/net/gianfar.c
> @@ -203,9 +203,9 @@ static int gfar_of_init(struct net_device *dev)
>  
>  		priv->interruptError = irq_of_parse_and_map(np, 2);
>  
> -		if (priv->interruptTransmit < 0 ||
> -				priv->interruptReceive < 0 ||
> -				priv->interruptError < 0) {
> +		if (priv->interruptTransmit == NO_IRQ ||
> +				priv->interruptReceive == NO_IRQ ||
> +				priv->interruptError == NO_IRQ) {
>  			err = -EINVAL;
>  			goto err_out;
>  		}

I guess that's OK, as gianfar.c appears to be a ppc-only thing, and ppc
does implement NO_IRQ.

However what a lot (all) rtc drivers do is

	if (irq <= 0)

which (afaik) will dtrt in all cases: NO_IRQ==0, NO_IRQ<0,
!defined(NO_IRQ).

Then again, it will dtwt if the platform allows irq numbers of zero.

argh, how did we get into this pickle?