[PATCH fixes] riscv: Properly export reserved regions in /proc/iomem

[Alexandre Ghiti]

Hi Bjorn,

On 09/04/2025 20:21, Björn Töpel wrote:
> From: Björn Töpel <bjorn at rivosinc.com>
>
> The /proc/iomem represents the kernel's memory map. Regions marked
> with "Reserved" tells the user that the range should not be tampered
> with. Kexec-tools, when using the older kexec_load syscall relies on
> the "Reserved" regions to build the memory segments, that will be the
> target of the new kexec'd kernel.
>
> The RISC-V port tries to expose all reserved regions to userland, but
> some regions were not properly exposed: Regions that resided in both
> the "regular" and reserved memory block, e.g. the EFI Memory Map. A
> missing entry could result in reserved memory being overwritten.
>
> It turns out, that arm64, and loongarch had a similar issue a while
> back:
>
>    commit d91680e687f4 ("arm64: Fix /proc/iomem for reserved but not memory regions")
>    commit 50d7ba36b916 ("arm64: export memblock_reserve()d regions via /proc/iomem")
>
> Similar to the other ports, resolve the issue by splitting the regions
> in an arch initcall, since we need a working allocator.
>
> Fixes: ffe0e5261268 ("RISC-V: Improve init_resources()")
> Signed-off-by: Björn Töpel <bjorn at rivosinc.com>
> ---
>   arch/riscv/kernel/setup.c | 36 +++++++++++++++++++++++++++++++++++-
>   1 file changed, 35 insertions(+), 1 deletion(-)
>
> diff --git a/arch/riscv/kernel/setup.c b/arch/riscv/kernel/setup.c
> index c174544eefc8..f7c9a1caa83e 100644
> --- a/arch/riscv/kernel/setup.c
> +++ b/arch/riscv/kernel/setup.c
> @@ -66,6 +66,9 @@ static struct resource bss_res = { .name = "Kernel bss", };
>   static struct resource elfcorehdr_res = { .name = "ELF Core hdr", };
>   #endif
>   
> +static int num_standard_resources;
> +static struct resource *standard_resources;
> +
>   static int __init add_resource(struct resource *parent,
>   				struct resource *res)
>   {
> @@ -139,7 +142,7 @@ static void __init init_resources(void)
>   	struct resource *res = NULL;
>   	struct resource *mem_res = NULL;
>   	size_t mem_res_sz = 0;
> -	int num_resources = 0, res_idx = 0;
> +	int num_resources = 0, res_idx = 0, non_resv_res = 0;
>   	int ret = 0;
>   
>   	/* + 1 as memblock_alloc() might increase memblock.reserved.cnt */
> @@ -193,6 +196,7 @@ static void __init init_resources(void)
>   	/* Add /memory regions to the resource tree */
>   	for_each_mem_region(region) {
>   		res = &mem_res[res_idx--];
> +		non_resv_res++;
>   
>   		if (unlikely(memblock_is_nomap(region))) {
>   			res->name = "Reserved";
> @@ -210,6 +214,9 @@ static void __init init_resources(void)
>   			goto error;
>   	}
>   
> +	num_standard_resources = non_resv_res;
> +	standard_resources = &mem_res[res_idx + 1];
> +
>   	/* Clean-up any unused pre-allocated resources */
>   	if (res_idx >= 0)
>   		memblock_free(mem_res, (res_idx + 1) * sizeof(*mem_res));
> @@ -221,6 +228,33 @@ static void __init init_resources(void)
>   	memblock_free(mem_res, mem_res_sz);
>   }
>   
> +static int __init reserve_memblock_reserved_regions(void)
> +{
> +	u64 i, j;
> +
> +	for (i = 0; i < num_standard_resources; i++) {
> +		struct resource *mem = &standard_resources[i];
> +		phys_addr_t r_start, r_end, mem_size = resource_size(mem);
> +
> +		if (!memblock_is_region_reserved(mem->start, mem_size))
> +			continue;
> +
> +		for_each_reserved_mem_range(j, &r_start, &r_end) {
> +			resource_size_t start, end;
> +
> +			start = max(PFN_PHYS(PFN_DOWN(r_start)), mem->start);
> +			end = min(PFN_PHYS(PFN_UP(r_end)) - 1, mem->end);
> +
> +			if (start > mem->end || end < mem->start)
> +				continue;
> +
> +			reserve_region_with_split(mem, start, end, "Reserved");
> +		}
> +	}
> +
> +	return 0;
> +}
> +arch_initcall(reserve_memblock_reserved_regions);
>   
>   static void __init parse_dtb(void)
>   {
>
> base-commit: a24588245776dafc227243a01bfbeb8a59bafba9


Reviewed-by: Alexandre Ghiti <alexghiti at rivosinc.com>

It will be merged with other fixes in the next rc-ish!

Thanks,

Alex