Messages in this thread | | | Date | Fri, 25 Apr 2014 13:50:25 +0200 | Subject | Re: [PATCH v2 3/6] ARM: dts: Device tree for AXM55xx. | From | Linus Walleij <> |
| |
On Fri, Apr 25, 2014 at 11:43 AM, Anders Berg <anders.berg@lsi.com> wrote: > On Fri, Apr 25, 2014 at 11:16:18AM +0200, Linus Walleij wrote: >> On Thu, Apr 24, 2014 at 7:47 PM, Anders Berg <anders.berg@lsi.com> wrote: >> > On Thu, Apr 24, 2014 at 03:24:14PM +0200, Linus Walleij wrote: >> >> >> One interrupt per CPU core? >> >> >> >> The drivers for these blocks will really just grab the first IRQ and >> >> then I guess they >> >> will only be able to execute on CPU0. >> >> >> >> It's definately correct to list all the IRQs here, but how do you envision >> >> the drivers making use of them in the long run? >> > >> > It's one interrupt line per input pin (so with the current driver only the first pin >> > is usable as interrupt source). >> >> Hm I'm not sure I understand what a "pin" is in this concept ... >> being maintainer of the pin control subsystem and all that really >> triggers my interest. >> > > Ok, maybe should replace "pin" with "GPIO" in my previous comment. > > So, a clarification. In one of the PL061 blocks (named gpio0 in the dts) there > is a separate interrupt per GPIO (I assume the motivation here is to enable to > control irq affinity per GPIO), where as the other block has a more standard > configuration with a single interrupt for all 8 GPIOs in that block.
OK! I get it. Thus this is not a stock PL061, but a version modified to generate a separate IRQ line per GPIO line.
This means that you can only get a proper, working IRQ from the first line on gpio0 right? All other IRQs will be ignored.
I see no problem in augmenting the driver to handle this if #irqs == 8, the big change needs to happen in drivers/amba/bus.c that has no way to pass any more than two IRQs atm.
I guess the best is to augment struct amba_device with a field struct resource *resource and num_resources like platform_device instead of the hard-coded single resource for iomem and two hardcoded IRQ placeholders, so that an arbitrary amount of resources can be added to an amba_device as well. But that may be quite a lot of work. (Russell will know which direction to take here.)
A simpler, local approach is to add some custom DT parsing code to handle this in the pl061 driver, but that feels a bit hacky.
Yours, Linus Walleij
| |