On Wed, Sep 1, 2021 at 7:00 AM Anup Patel [off-list ref] wrote:

On Wed, Sep 1, 2021 at 6:59 AM Rob Herring [off-list ref] wrote:

quoted

On Mon, Aug 30, 2021 at 09:47:24AM +0530, Anup Patel wrote:

quoted

The Linux RISC-V now treats IPIs as regular per-CPU IRQs. This means
we have to create a IPI interrupt domain to use CLINT IPI functionality
hence requiring a "interrupt-controller" and "#interrupt-cells" DT
property in CLINT DT nodes.

Impact of this CLINT DT bindings change only affects Linux RISC-V
NoMMU kernel and has no effect of existing M-mode runtime firmwares
(i.e. OpenSBI).

It appears to me you should fix Linux to not need these 2 useless
properties. I say useless because #interrupt-cells being 0 is pretty
useless.

Linux IRQCHIP framework only probes IRQCHIP DT nodes which
have "interrupt-controller" DT property.

Right, I believe I wrote that... So what would it look like to fix
that? The simplest thing is just drop the check for
'interrupt-controller'. That's just a sanity check and we have other
ways to do that now (schemas). Do you need this early? You can always
implement your own initcall.

The "interrupt-cells" DT property
can be removed because as an interrupt controller SiFive CLINT
will only provide IPIs to arch code.

The schema will disagree.

Rob