[PATCH] free_irq(): Actually handle DEBUG_SHIRQ

[PATCH] free_irq(): Actually handle DEBUG_SHIRQ

[Maciej W. Rozycki]

 Code used for DEBUG_SHIRQ in free_irq() is unreachable -- the for() loop 
within never terminates otherwise than by return.  This is an obvious fix.

Signed-off-by: Maciej W. Rozycki <macro@linux-mips.org>
---
 Please apply.

 While at it, I have a question about the complementary code within 
request_irq(): when the option in question is enabled, a spurious IRQ is 
posted before internal setup is done by setup_irq().  This includes the 
->depth counter.  Now if the interrupt handler of some driver calls
disable_irq_nosync(), then code in setup_irq() will mess up the state of 
->depth afterwards if this is the first handler being installed (i.e. 
within the "if (!shared)" block).  This is reported later on when 
enable_irq() called by the driver, by means of a message like this:

Unbalanced enable for IRQ 34
WARNING: at kernel/irq/manage.c:158 enable_irq()

Now given DEBUG_SHIRQ is a debug facility, not to be normally used for 
production, is the phenomenon as described considered a design limitation 
we can live with, or is it a bug that qualifies for a fix?

 I can reproduce this problem easily with phylib.

  Maciej

patch-mips-2.6.23-rc5-20070904-debug-shirq-0
diff -up --recursive --new-file linux-mips-2.6.23-rc5-20070904.macro/kernel/irq/manage.c linux-mips-2.6.23-rc5-20070904/kernel/irq/manage.c
--- linux-mips-2.6.23-rc5-20070904.macro/kernel/irq/manage.c	2007-09-04 04:56:21.000000000 +0000
+++ linux-mips-2.6.23-rc5-20070904/kernel/irq/manage.c	2007-09-11 23:58:59.000000000 +0000
@@ -448,7 +448,7 @@ void free_irq(unsigned int irq, void *de
 			if (action->flags & IRQF_SHARED)
 				handler = action->handler;
 			kfree(action);
-			return;
+			break;
 		}
 		printk(KERN_ERR "Trying to free already-free IRQ %d\n", irq);
 		spin_unlock_irqrestore(&desc->lock, flags);

Re: [PATCH] free_irq(): Actually handle DEBUG_SHIRQ

[David Woodhouse]

On Wed, 2007-09-12 at 12:17 +0100, Maciej W. Rozycki wrote:
> Code used for DEBUG_SHIRQ in free_irq() is unreachable -- the for()
> loop within never terminates otherwise than by return.  This is an
> obvious fix.

Yeah... except that it enables new debugging code which (due to my
stupidity) wasn't enabled before. A fix for this is already in the -mm
tree, to be merged (along with a corresponding lockdep-related fix)
after 2.6.23 release.

-- 
dwmw2