* Re: [PATCH] of: reserved_mem: mark the early_memremap() array helpers __init
2026-11-02 22:45 ` [PATCH] of: reserved_mem: mark the early_memremap() array helpers __init Sasha Levin
@ 2026-11-03 9:44 ` Marek Szyprowski
2026-11-06 21:32 ` Rob Herring (Arm)
1 sibling, 0 replies; 3+ messages in thread
From: Marek Szyprowski @ 2026-11-03 9:44 UTC (permalink / raw)
To: Sasha Levin, Rob Herring
Cc: Saravana Kannan, Mike Rapoport, Nathan Chancellor, devicetree,
llvm, linux-kernel
On 02.11.2026 23:45, Sasha Levin wrote:
> early_array_map() and early_array_unmap() wrap early_memremap() and
> early_memunmap(), which are __init, but the helpers themselves are only
> static inline. If the compiler decides not to inline them, the out of
> line copy ends up in .text and modpost complains. clang 21 does this for
> a riscv allnoconfig build:
>
> WARNING: modpost: vmlinux: section mismatch in reference: early_array_map+0x30 (section: .text) -> early_memremap (section: .init.text)
> ERROR: modpost: Section mismatches detected.
>
> Both helpers are only called from fdt_scan_reserved_mem(), which is
> __init, so mark them __init too.
>
> Fixes: aa05faf81583 ("of: reserved_mem: allocate and map the reserved_mem array early")
> Signed-off-by: Sasha Levin <sashal@kernel.org>
I wasn't aware that 'static inline' functions might need __init
annotations.
Acked-by: Marek Szyprowski <m.szyprowski@samsung.com>
> ---
> KernelCI's clang-21 riscv builds fail on this once dt/for-next is
> merged, so it will probably show up in tomorrow's linux-next as well.
>
> Tested with a riscv LLVM=1 allnoconfig build (fails at modpost before,
> links after), plus W=1 builds of of_reserved_mem.o for riscv and arm64
> defconfig with clang, and arm64 defconfig with gcc.
>
> drivers/of/of_reserved_mem.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/of/of_reserved_mem.c b/drivers/of/of_reserved_mem.c
> index 0fd1683cba011..630ad65466385 100644
> --- a/drivers/of/of_reserved_mem.c
> +++ b/drivers/of/of_reserved_mem.c
> @@ -66,11 +66,11 @@ static int __init early_init_dt_alloc_reserved_memory_arch(phys_addr_t size,
> */
> #if defined(CONFIG_GENERIC_EARLY_IOREMAP) && !defined(CONFIG_PPC64)
> #include <asm/early_ioremap.h>
> -static inline void *early_array_map(void *array, size_t size)
> +static inline void * __init early_array_map(void *array, size_t size)
> {
> return early_memremap(__pa(array), size);
> }
> -static inline void early_array_unmap(void *map, size_t size)
> +static inline void __init early_array_unmap(void *map, size_t size)
> {
> early_memunmap(map, size);
> }
Best regards
--
Marek Szyprowski, PhD
Samsung R&D Institute Poland
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [PATCH] of: reserved_mem: mark the early_memremap() array helpers __init
2026-11-02 22:45 ` [PATCH] of: reserved_mem: mark the early_memremap() array helpers __init Sasha Levin
2026-11-03 9:44 ` Marek Szyprowski
@ 2026-11-06 21:32 ` Rob Herring (Arm)
1 sibling, 0 replies; 3+ messages in thread
From: Rob Herring (Arm) @ 2026-11-06 21:32 UTC (permalink / raw)
To: Sasha Levin
Cc: linux-kernel, Marek Szyprowski, Saravana Kannan, llvm,
devicetree, Mike Rapoport, Nathan Chancellor
On Mon, 02 Nov 2026 17:45:28 -0500, Sasha Levin wrote:
> early_array_map() and early_array_unmap() wrap early_memremap() and
> early_memunmap(), which are __init, but the helpers themselves are only
> static inline. If the compiler decides not to inline them, the out of
> line copy ends up in .text and modpost complains. clang 21 does this for
> a riscv allnoconfig build:
>
> WARNING: modpost: vmlinux: section mismatch in reference: early_array_map+0x30 (section: .text) -> early_memremap (section: .init.text)
> ERROR: modpost: Section mismatches detected.
>
> Both helpers are only called from fdt_scan_reserved_mem(), which is
> __init, so mark them __init too.
>
> Fixes: aa05faf81583 ("of: reserved_mem: allocate and map the reserved_mem array early")
> Signed-off-by: Sasha Levin <sashal@kernel.org>
> ---
> KernelCI's clang-21 riscv builds fail on this once dt/for-next is
> merged, so it will probably show up in tomorrow's linux-next as well.
>
> Tested with a riscv LLVM=1 allnoconfig build (fails at modpost before,
> links after), plus W=1 builds of of_reserved_mem.o for riscv and arm64
> defconfig with clang, and arm64 defconfig with gcc.
>
> drivers/of/of_reserved_mem.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
Applied, thanks!
^ permalink raw reply [flat|nested] 3+ messages in thread