the "enable_gic=0" cannot work on PI4

[figozhang]

hi guys, i am writing a Bare Metal app on PI4, and i want to use the legacy interrupt controller, so i add "enable_gic=0" in the config.txt, but it seem the interrupt cannot route to legacy interrupt controller.

here is my config.txt: [pi4] kernel=myos4.bin enable_gic=0

[pi3] kernel=myos3.bin

[all] arm_64bit=1 #enable 64bit image enable_uart=1 #enable pl uart uart_2ndstage=1 #enable FW debug info

what is the problem?

[pelwell]

The Device Tree and the firmware configuration have to agree on interrupt usage. If you are using a .dtb file that includes a top-level interrupt-parent property referring to a node with a compatible string of "arm,gic-400" then it will force enable_gic to be 1.

[figozhang]

if i delete the dtb file in the SD card, it canot boot to my app. so any suggestion that if i want to run my bare metal app with legacy interrupt controller?

[pelwell]

I think you have two options:

1. Create a version of the DTB that uses the legacy interrupt controller. If you aren't going to use Device Tree at all then you could use sed or a hex editor to change "arm,gic-400" to something else the same length, e.g. "not,gic-400".
2. Embrace the GIC - as I understand it, it's not significantly harder to use than the legacy controller, and the performance is going to be better. @dp111 might be able to comment on this.

[MartinDupont]

I just lost 3 days of work because of this issue.

When you read the documentation for enable_gic, it says:

On the Raspberry Pi 4B, if this value is set to 0 then the interrupts will be routed to the ARM cores using the legacy interrupt controller, rather than via the GIC-400. The default value is 1.

It makes absolutely no mention of the device tree. In fact, it implies that setting enable_gic=0 is enough to enable the legacy controller. Which is why I spent 3 days messing around with different registers trying to figure out why my IRQ's weren't firing.

At the very least, the documentation needs to be updated.

I had a different workaround to the one proposed by pelwell though. I found the armstub firmware and simply removed all the lines that deal with the GIC and then compiled it myself. Worked like a charm.