Re: [PATCH 7/7] riscv: dts: tenstorrent: Add Atlantis platform

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] Circular clock parent assignment for CLK_RCPU_ROOT creates a circular dependency.
- [Medium] The machine-level IMSIC (`imsic_m`) incorrectly declares guest files (`riscv,guest-index-bits`) and an oversized `reg` mapping.
--

commit 049e98b3d4ab535b3e1fedc8df666805cb9e6805
Author: Joel Stanley <joel@jms.id.au>

riscv: dts: tenstorrent: Add Atlantis platform

Add initial support for the Tenstorrent Atlantis platform, based on the
Atlantis SoC featuring 8x RVA23-compliant Tenstorrent Ascalon-XG cores.

> diff --git a/arch/riscv/boot/dts/tenstorrent/atlantis-soc.dtsi b/arch/riscv/boot/dts/tenstorrent/atlantis-soc.dtsi
> new file mode 100644
> --- /dev/null
> +++ b/arch/riscv/boot/dts/tenstorrent/atlantis-soc.dtsi

[ ... ]

> +	soc {
> +		#address-cells = <2>;
> +		#size-cells = <2>;
> +		compatible = "simple-bus";
> +		interrupt-parent = <&aplic_s>;
> +		ranges;
> +
> +		imsic_m: interrupt-controller@a0000000 {
> +			compatible = "tenstorrent,atlantis-imsics", "riscv,imsics";
> +			riscv,num-ids = <255>;
> +			riscv,guest-index-bits = <6>;
> +			reg = <0x0 0xa0000000 0x0 0x200000>;

[Severity: Medium]
Does this incorrectly declare guest files for the machine-level IMSIC?

The RISC-V Advanced Interrupt Architecture specifies that Guest Interrupt
Files exist exclusively in supervisor-level IMSICs. Since machine-level
IMSICs only contain one interrupt file per hart, would having
riscv,guest-index-bits = <6> and an identically oversized 2MB reg space
violate AIA semantics?

> +			interrupts-extended = <&cpu0_intc 11>, <&cpu1_intc 11>,

[ ... ]

> +		prcm_rcpu: clock-controller@d0000000 {
> +			compatible = "tenstorrent,atlantis-prcm-rcpu";
> +			reg = <0x00 0xd0000000 0x00 0x10000>;
> +			clocks = <&osc_24m>;
> +			#clock-cells = <1>;
> +			#reset-cells = <1>;
> +			assigned-clocks = <&prcm_rcpu CLK_RCPU_ROOT>, <&prcm_rcpu CLK_NOCC_CLK>;
> +			assigned-clock-parents = <&prcm_rcpu CLK_RCPU_ROOT>, <&prcm_rcpu CLK_NOC_PLL>;
> +		};

[Severity: High]
Will this circular clock parent assignment create a dependency issue in
the Common Clock Framework?

The assigned-clocks and assigned-clock-parents map CLK_RCPU_ROOT to itself.
Could this cause of_clk_set_defaults() to fail or silently abandon
reparenting, leaving the system running on an uninitialized default
hardware clock source?

[ ... ]

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260603074222.593243-1-joel@jms.id.au?part=7