Re: [PATCH] net/cadence/macb: clear interrupts simply and correctly

[Jongsung Kim]

On 06/17/2014 12:50 PM, SÃren Brinkmann wrote:
> On Tue, 2014-06-17 at 11:38AM +0900, Jongsung Kim wrote:
>> On 06/16/2014 11:56 PM, SÃren Brinkmann wrote:
>>> On Mon, 2014-06-16 at 02:00PM +0900, Jongsung Kim wrote:
>>>> On 06/13/2014 12:44 AM, SÃren Brinkmann wrote:
>>>>> This is now clearing all IRQ flags which is probably not what we want
>>>>> here. This is handling RX only. We still want the non-RX interrupts to go to
>>>>> the actual interrupt service routing.
>>>>
>>>> The ISR(Interrupt Status Register) is read only in the interrupt service
>>>> routine, macb_interrupt. But is partially cleared here and there. Further
>>>> handler-functions decide jobs to be done by reading/checking other status
>>>> registers. (e.g., TSR, RSR) So, clearing the ISR after reading looks not
>>>> a bad idea.
>>>
>>> But you are clearing _all_ interrupt flags in the RX NAPI handler.
>>> Doesn't that mean we might miss certain events?
>>
>> Please inspect my patch again. What I did in the macb_poll is removing
>> statements clearing the Rx-complete interrupt, not clearing all the
>> interrupts.
> 
> Why is clearing those bits removed? It's probably not a big hit, but it might
> result in a pointless interrupt which could be avoided. But it should
> probably clear all RX interrupts - MACB_RX_INT_FLAGS - instead of just RCOMP.
> For clear-on-read implementations it shouldn't make a difference.

I agree. But I removed it because I think stepping the same procedure
regardless of the "gem_irq_clear_read" implementation is better than
implementation-specific optimization.

> And in the if-condition in that new helper, I'd add '&& status' to
> avoid writing back zeros.

Good point. I'll add it when I resend v2.

Jongsung
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/