Re: [PATCH v1 07/27] xen/riscv: implement make_timer_node()

[Oleksii Kurochko <oleksii.kurochko@gmail.com>]

On 4/1/26 4:24 PM, Jan Beulich wrote:
> On 10.03.2026 18:08, Oleksii Kurochko wrote:
>> The make_timer_node() function is implemented to return 0,
>> as RISC-V does not require the creation of a timer node.
>>
>> Signed-off-by: Oleksii Kurochko <oleksii.kurochko@gmail.com>
> 
> Acked-by: Jan Beulich <jbeulich@suse.com>
> 
> Yet: Why is this needed then in the first place? If this node may not
> be required, why would the function need providing? Shouldn't the call
> site be conditional then?

Generally, in DT for RISC-V there is a document which describes timer 
node (riscv,timer.yaml or sifive,clint.yaml), but the driver 
(drivers/clocksource/timer-riscv.c:244) is declared with 
TIMER_OF_DECLARE(riscv_timer, "riscv", ...). It matches the CPU node 
(compatible "riscv"), not the timer node itself. It then does 
of_find_compatible_node(NULL, NULL, "riscv,timer") only to read the 
optional riscv,timer-cannot-wake-cpu flag. But I don't see too much 
sense at the moment to enable timer-cannot-wake-cpu option.

Regarding, sifive,clint.yaml basically it has almost the same as 
riscv,timer node but also additionally provide some MMIO which are 
acessible only in M-mode (what isn't the case for virtualization as 
guest isn't ran in M-mode, otherwise emulation of M-mode register will 
be needed).

So potentially some will want to  add riscv,timer node to enable 
timer-cannot-wake-cpu option, so then some changes additionally should 
be done in this function.

Considering that it isn't needed now and timebase-frequency property is 
always part of cpu node not a timer one, I think we can go with an empty 
implementation of make_timer_node().
I can add some extra information to commit message.

~ Oleksii