Re: [PATCH v7 1/5] usb: dwc3: omap: use request_threaded_irq()

[Felipe Balbi <balbi@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 2232 bytes --]


Hi,

Roger Quadros <rogerq@ti.com> writes:
>>> @@ -497,8 +503,8 @@ static int dwc3_omap_probe(struct platform_device *pdev)
>>>  	/* check the DMA Status */
>>>  	reg = dwc3_omap_readl(omap->base, USBOTGSS_SYSCONFIG);
>>>  
>>> -	ret = devm_request_irq(dev, omap->irq, dwc3_omap_interrupt, 0,
>>> -			"dwc3-omap", omap);
>>> +	ret = devm_request_threaded_irq(dev, omap->irq, dwc3_omap_interrupt,
>>> +					NULL, 0, "dwc3-omap", omap);
>> 
>> if you're using threaded_irq, it's better to have a NULL top half and
>> valid bottom half.
>
> But in this case we don't need a bottom half as there is nothing to do :).
>
>> 
>> In fact, since this will be shared, you could do a proper preparation
>> and on top half check if $this device generated the IRQ and
>> conditionally schedule the bottom half. Don't forget to mask device's
>> interrupts from top half so you can run without IRQF_ONESHOT.
>> 
>
> Why do this at all if there is nothing to do in the bottom half?

oh, but there is :-)

The whole idea of threaded IRQs is that you spend as little time as
possible on top half and the (strong) recommendation is that you *only*
check if $this device generated the interrupt. Note that "checking if
$this device generated the interrupt" will be mandatory as soon as you
mark the IRQ line as shared ;-)

So here's how this should look like:

static irqreturn_t dwc3_omap_interrupt(int irq, void *_omap)
{
        struct dwc3_omap *omap = _omap;
        u32 reg;

        reg = readl(IRQSTATUS)
        if (reg) {
                mask_interrupts(omap);
        	return IRQ_WAKE_THREAD;
        }

	return IRQ_HANDLED;
}

static irqreturn_t dwc3_omap_threaded_interrupt(int irq, void *_omap)
{
        struct dwc3_omap *omap = _omap;
        u32 reg;

        spin_lock(&omap->lock);
        reg = readl(IRQSTATUS);

        if (reg & BIT0)
        	handle_bit_0(omap);

	if (reg & BIT1)
        	handle_bit_1(omap);

	unmask_interrupts(omap);
	spin_unlock(&omap->lock);

	return IRQ_HANDLED;
}

this will *always* behave well with RT and non-RT kernels. It also
allows for the user to change priorities on these interrupt handlers if
necessary.

-- 
balbi

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 818 bytes --]