On Wed, Feb 27, 2013 at 7:28 PM, Paul Mundt <[email protected]> wrote:
> On Wed, Feb 27, 2013 at 06:52:51PM +0900, Magnus Damm wrote:
>> As you know, the INTC code that you are referring to is a full
>> interrupt controller designed to work directly with CPU cores like SH
>> and ARM. Newer ARM cores like Cortex-A9 all include the GIC both for
>> IPI purpose in case of SMP and they also implement SPI functionality
>> for I/O device interrupts. So the need for vendor-specific interrupt
>> controller IP is almost down to nothing these days.
>>
> Yes, I'm aware of that.
Good!
>> With that in mind I do really doubt that we end up reimplementing
>> sh_intc. The upcoming designs that I am aware of do not change their
>> external IRQ pin hardware to force us to update this driver. What
>> happens after that I'm afraid I can't say. =)
>>
> While I don't expect you would end up with a full reimplmentation, my
> concern is more that it would just be reproducing stuff that's already in
> place. IOW, perhaps it's less work to put sh_intc on a diet, and tie the
> core functionality that you need for external IRQ pins in to an irqchip
> there -- with the core of the old code adapted in to some sort of common
> base library, rather than coming up with a new lightweight irqchip driver
> and having to incrementally pile stuff on top of it later.
I'm afraid that I can't really see how we would want to pile that much
stuff on top of this driver in the future. Maybe we need to adjust it
slightly, but that should be fairly easy.
>From my point of view it all boils down to this for our ARM SoCs:
1) The driver is only suitable for older GIC-based designs coming from
the Renesas side.
2) GIC-based SoCs like EMEV2 coming from ex-NEC is using different IP
and different driver.
3) Older SoCs that do not make use of GIC are using the full blown
INTC implementation.
>> I mainly wrote this driver to support the r8a7779 SoC which can't be
>> driven by the existing INTC code base. During development I decided to
>> try to share the driver between multiple GIC-based SoCs like r8a7779
>> and sh73a0. The reason behind trying to share this driver between
>> multiple SoCs is to remove SoC-specific hacks that can't be dealt with
>> by the existing INTC code.
>>
> Ok, fair enough.
>
>> I don't really understand why you would want to use a generic table
>> driven driver when you have the possibility of using a
>> hardware-specific driver.
>
> For the same reason sh_intc was written in the first place, every CPU
> subtype more or less had a similar set of interrupt controllers with
> minor deviations. Those deviations were sufficient enough to make the
> code pretty hairy in places, and what we have now is really more of a
> subsystem than an irqchip driver.
>
> Having to reproduce 90% of the code when you're adding new CPUs at the
> rate of 2 a month hardly makes an SoC-specific driver realistic,
> especially as you just end up with identical features being broken in
> subtly differnt ways on different SoCs. That being said, a lot of that is
> legacy stuff and a result of no CPU team talking to any other CPU team
> ever, and it seems like things have improved enough in that regard that
> perhaps a hardware-specific driver won't be a complete throw-away
> disaster like it was before. It's a risk either way, I just hope your
> lightweight solution remains lightweight and consistent long enough that
> we don't have multiple copies of slightly different drivers doing exactly
> same thing spiralling out of control and turning in to a maintenance
> nightmare.
Yes, I agree about the history with zillions of different SoCs, and
the INTC and PFC design choice.
But regarding the external pins in case of GIC-based designs, this
seems to be something we can reuse as a regular driver.
> If sh_intc doesn't deal with the needs of the new GIC-backed SoCs then
> that's of course something that has to be addressed regardless (whether
> that be hacking up sh_intc or adding new hardware-specific irqchips).
Ignoring things won't work, agreed.
>> I suppose you are wondering why no one seems to be working on INTC DT
>> support at this point. The truth is that we got very little feedback
>> and development support with interrupts in general last autumn and on
>> top of that our developers went their own way instead of following
>> advice.
>>
> I assumed it was either being rewritten or had already been merged, so I
> was simply surprised to hear otherwise. I will dig through the archives a
> bit later and see about getting caught up.
Sounds good.
Thanks,
/ magnus