Re: [PATCH] genirq: Introduce irq_read_line()

[Marc Zyngier <marc.zyngier@arm.com>]

Hi Bjorn,

On 24/10/14 18:31, Bjorn Andersson wrote:
> On Tue, Oct 21, 2014 at 2:34 AM, Marc Zyngier <marc.zyngier@arm.com> wrote:
>> Hi all,
>>
>> On 21/10/14 10:22, Thomas Gleixner wrote:
>>> On Thu, 4 Sep 2014, Bjorn Andersson wrote:
>>>
>>>> On Tue, Aug 19, 2014 at 1:23 PM, Bjorn Andersson
>>>> <bjorn.andersson@sonymobile.com> wrote:
>>>>> Introduce the irq_read_line() function to allow device drivers to read
>>>>> the current logical state of an input when the hardware only exposes
>>>>> this through status bits in the interrupt controller.
>>>>>
>>>>> The new function is backed by a new callback function in the irq_chip -
>>>>> irq_read_line() - that can be implemented by irq_chips that owns such
>>>>> status bits.
>>>>>
>>>>> Based on rfc patch from April 2011 by Abhijeet.
>>>>>
>>>>> Cc: Abhijeet Dharmapurikar <adharmap@codeaurora.org>
>>>>> Signed-off-by: Bjorn Andersson <bjorn.andersson@sonymobile.com>
>>>>
>>>> ping?
>>>
>>> Sorry, slipped through the cracks. I was talking about this to Marc
>>> last week and he needs it for yet another reason. He had some thoughts
>>> about the state representation, so I wait for him to comment.
>>
>> Thanks for putting me in the loop. For the record, here's the RFC I
>> posted back in June:
>>
> 
> Thanks for having a similar problem as me :)
> 
>> http://lists.infradead.org/pipermail/linux-arm-kernel/2014-June/266328.html
>>
>> and the patch implementing a similar concept:
>> http://lists.infradead.org/pipermail/linux-arm-kernel/2014-June/266331.html
>>
> 
> I would prefer to see that you had explicit functions for the various things
> that you would like to set and get. It would in my eyes give cleaner client
> code, at the cost of a few extra functions.

I don't really like the idea of having a gazillion of accessors for each
individual property. I'd rather have an extensible API.

>> Basic idea is that you can read (and possibly write back) various
>> low-level attributes (pending, masked, active) that an interrupt
>> controller may implement. Given your use case, we should loose the
>> checks on the interrupt being forwarded, as this makes little sense
>> outside of virtualization.
>>
> 
> Relaxing the forwarding requirement seems to solve my use cases. May I ask why
> your accessors would only be made available for forwarded IRQs - at a framework
> level?

This was never intended for wider use, and restricting it to forwarded
interrupts was a concious design decision, as I find the idea of a
device driver messing with the internal state of the interrupt
controller positively repulsive... But hey, the more the merrier.

> Stephen Boyd talked about the need to be able to mask/unmask interrupts from
> client code in the Qualcomm platform as well - most likely to block wakeup
> sources(?)

What's wrong with irq_disable?

>> I'm planning to respin the series this week, as I have a number of
>> changes (there is hardly any need for the various states to be reported
>> atomically, for example), and a number of bugs have been found.
>>
> 
> Sounds good, for clarity I would like them to be explicit separate functions.
> But as long as it's not limited to forwarded IRQs we should be able to use it.

I just pushed out a branch:
git://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms.git
irq/irqchip_state

Please let me know if that's useful for you.

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...