On Mon, 10 Mar 2025 at 06:29, Sourojeet Adhikari
<s23ad...@csclub.uwaterloo.ca> wrote:
>
> Hello,
>
> So far I've figured out that I need to add a member called SplitIRQ
> in the BCMSocPeripheralBaseState struct, which is of size defined
> by `BCM2835_NUM_SPLITTERS`. Then from what I can tell through
> reading through the codebase, I should do something similar to
> what happens in `exynos4210.c` in the `exynos4210_init_board_irqs`
> function?

QEMU qdev devices have a multiple-stage initialization process:
 (1) 'init', which in this case means calling object_initialize_child().
     In this case you should do this in raspi_peripherals_base_init(),
     where we do this for the other child devices of the common
     peripheral base object.
 (2) set properties (in this case the "num-lines" property, which
     needs to be set to 2). You do this in the parent object's 'realize'
     method, in this case bcm_soc_peripherals_common_realize().
     There are several different equivalent APIs to set a simple
     integer property; using qdev_prop_set_uint16() like the exynos
     code does is fine.
 (3) realize the device by calling qdev_realize(); this has to be
     done after all properties are set.
 (4) wire up the input and output GPIO lines (you want to wire up
     a GPIO only once the devices on both ends have been realized).
     For this case, in the bcm_soc_peripherals_common_realize()
     function we're going to wire up the the device's outbound
     IRQ to the input of the appropriate splitter, and connect
     output 0 of that splitter to the BCM2835_IC correct interrupt
     controller input.
 (5) The last bit of wiring, where we connect splitter output 1 to
     the correct GIC input, we do in bcm2838.c, and only after
     we've called 'realize' for both the bcm2835_peripherals object
     and the GIC object.

> Sorry for taking a while to get back and still having quite a few
> questions about this. Thank you for helping me out with writting
> this patch.

No worries; this stuff is all quite complicated and we don't
really document it very well.

thanks
-- PMM

Reply via email to