Re: bcm43xx-mac80211: a fix for the shared interrupt problem

[Michael Buesch <mb@bu3sch.de>]

On Saturday 28 July 2007 05:36:04 Larry Finger wrote:
> Michael,
> 
> I was in the right area with my previous fix to the "irq 11: nobody cared"
> message. I think for shared interrupts, skipping out if the status is not
> BCM43xx_STAT_STARTED is too severe. I added a message printing out the status
> whenever it skipped out, and found many thousands of interrupts with the
> status equal to BCM43xx_STAT_INITIALIZED. The patch below makes the necessary
> change and modifies a couple of asserts that depend on it.
> 
> This patch has been tested on BCM4306 and BCM4318 with shared interrupts,
> and BCM4311 without.
> 
> Thanks,
> 
> Larry
> ----
> 
> 
> Index: wireless-mb/drivers/net/wireless/bcm43xx-mac80211/bcm43xx_main.c
> ===================================================================
> --- wireless-mb.orig/drivers/net/wireless/bcm43xx-mac80211/bcm43xx_main.c
> +++ wireless-mb/drivers/net/wireless/bcm43xx-mac80211/bcm43xx_main.c
> @@ -1374,7 +1374,7 @@ static void bcm43xx_interrupt_tasklet(st
>  
>  	spin_lock_irqsave(&dev->wl->irq_lock, flags);
>  
> -	assert(bcm43xx_status(dev) == BCM43xx_STAT_STARTED);
> +	assert(bcm43xx_status(dev) >= BCM43xx_STAT_INITIALIZED);
>  
>  	reason = dev->irq_reason;
>  	for (i = 0; i < ARRAY_SIZE(dma_reason); i++) {
> @@ -1512,7 +1512,7 @@ static irqreturn_t bcm43xx_interrupt_han
>  
>  	spin_lock(&dev->wl->irq_lock);
>  
> -	if (bcm43xx_status(dev) < BCM43xx_STAT_STARTED)
> +	if (bcm43xx_status(dev) < BCM43xx_STAT_INITIALIZED)
>  		goto out;
>  	reason = bcm43xx_read32(dev, BCM43xx_MMIO_GEN_IRQ_REASON);
>  	if (reason == 0xffffffff) /* shared IRQ */
> @@ -1522,7 +1522,7 @@ static irqreturn_t bcm43xx_interrupt_han
>  	if (!reason)
>  		goto out;
>  
> -	assert(bcm43xx_status(dev) == BCM43xx_STAT_STARTED);
> +	assert(bcm43xx_status(dev) >= BCM43xx_STAT_INITIALIZED);
>  
>  	dev->dma_reason[0] = bcm43xx_read32(dev, BCM43xx_MMIO_DMA0_REASON)
>  			     & 0x0001DC00;
> 
> 

That's not the right bugfix.
If the interface is not STAT_STARTED, it should _not_ generate IRQs.
The bug is elsewhere in the init routines.
Please find out which IRQ bits are generated for these spurious IRQs.

-- 
Greetings Michael.