Re: [PATCH v7 10/11] arm64: dts: ti: k3-am62p-ti-ipc-firmware: Split r5f memory region

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] Splitting the contiguous firmware memory region into multiple smaller segments breaks ELF loading for existing monolithic firmwares.
--

commit 467b992dbc807cab9f7fc93edfad5b8ea8a2207c
Author: Markus Schneider-Pargmann (TI) <msp@baylibre.com>

arm64: dts: ti: k3-am62p-ti-ipc-firmware: Split r5f memory region

This commit splits the contiguous firmware memory region into more
specific parts. It explicitly describes where to find LPM metadata
and other regions for bootloaders like U-Boot.

> diff --git a/arch/arm64/boot/dts/ti/k3-am62p-ti-ipc-firmware.dtsi b/arch/arm64/boot/dts/ti/k3-am62p-ti-ipc-firmware.dtsi
> index f776511095642..8f7409da83392 100644
> --- a/arch/arm64/boot/dts/ti/k3-am62p-ti-ipc-firmware.dtsi
> +++ b/arch/arm64/boot/dts/ti/k3-am62p-ti-ipc-firmware.dtsi
> @@ -24,9 +24,33 @@ wkup_r5fss0_core0_dma_memory_region: memory@9c800000 {
>  		no-map;
>  	};
>  
> -	wkup_r5fss0_core0_memory_region: memory@9c900000 {
> +	wkup_r5fss0_core0_ipc_region: memory@9c900000 {
>  		compatible = "shared-dma-pool";
> -		reg = <0x00 0x9c900000 0x00 0x01d08000>;
> +		reg = <0x00 0x9c900000 0x00 0x100000>;
> +		no-map;
> +	};

[Severity: High]
Will splitting this contiguous memory region into smaller fragmented regions
break ELF loading for existing monolithic firmwares?

When the remoteproc driver loads the firmware via rproc_elf_load_segments(),
k3_rproc_da_to_va() translates the addresses by checking each DT region
independently. It strictly requires the entire segment to fit within a single
memory region:

drivers/remoteproc/ti_k3_common.c:k3_rproc_da_to_va() {
    ...
    if (da >= dev_addr && ((da + len) <= (dev_addr + size))) {
        offset = da - dev_addr;
    ...
}

If an older firmware has a segment that is larger than the newly defined
boundaries, such as a segment crossing the 1MB ipc region and the 32KB
lpm-stub region, this translation returns NULL and causes an -EINVAL error.

Furthermore, because the driver maps each of these memory regions
independently, the kernel virtual addresses are disjoint:

drivers/remoteproc/ti_k3_common.c:k3_reserved_mem_init() {
    ...
    kproc->rmem[i].cpu_addr = devm_ioremap_resource_wc(dev, &res);
    ...
}

Could this result in a Device Tree ABI regression for existing users with
older monolithic firmwares?

[ ... ]

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260701-topic-am62a-ioddr-dt-v6-19-v7-0-e9db8b16821a@baylibre.com?part=10