2018-12-07 10:16:58

by James Morse

[permalink] [raw]
Subject: [PATCH arm64/kexec] arm64: kexec_file: forbid kdump via kexec_file_load()

Now that kexec_walk_memblock() can do the crash-kernel placement itself
architectures that don't support kdump via kexe_file_load() need to
explicitly forbid it.

We don't support this on arm64 until the kernel can add the elfcorehdr
and usable-memory-range fields to the DT. Without these the crash-kernel
overwrites the previous kernel's memory during startup.

Add a check to refuse crash image loading.

Signed-off-by: James Morse <[email protected]>
---

Context: http://lore.kernel.org/r/[email protected]

arch/arm64/kernel/kexec_image.c | 4 ++++
1 file changed, 4 insertions(+)

diff --git a/arch/arm64/kernel/kexec_image.c b/arch/arm64/kernel/kexec_image.c
index 1ad1d5f8f024..07bf740bea91 100644
--- a/arch/arm64/kernel/kexec_image.c
+++ b/arch/arm64/kernel/kexec_image.c
@@ -47,6 +47,10 @@ static void *image_load(struct kimage *image,
struct kexec_segment *kernel_segment;
int ret;

+ /* We don't support crash kernels yet. */
+ if (image->type == KEXEC_TYPE_CRASH)
+ return ERR_PTR(-EOPNOTSUPP);
+
/*
* We require a kernel with an unambiguous Image header. Per
* Documentation/booting.txt, this is the case when image_size
--
2.19.2



2018-12-07 15:28:11

by Bhupesh Sharma

[permalink] [raw]
Subject: Re: [PATCH arm64/kexec] arm64: kexec_file: forbid kdump via kexec_file_load()

Hi James,

On Fri, Dec 7, 2018 at 3:44 PM James Morse <[email protected]> wrote:
>
> Now that kexec_walk_memblock() can do the crash-kernel placement itself
> architectures that don't support kdump via kexe_file_load() need to
> explicitly forbid it.
>
> We don't support this on arm64 until the kernel can add the elfcorehdr
> and usable-memory-range fields to the DT. Without these the crash-kernel
> overwrites the previous kernel's memory during startup.
>
> Add a check to refuse crash image loading.
>
> Signed-off-by: James Morse <[email protected]>
> ---
>
> Context: http://lore.kernel.org/r/[email protected]
>
> arch/arm64/kernel/kexec_image.c | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/arch/arm64/kernel/kexec_image.c b/arch/arm64/kernel/kexec_image.c
> index 1ad1d5f8f024..07bf740bea91 100644
> --- a/arch/arm64/kernel/kexec_image.c
> +++ b/arch/arm64/kernel/kexec_image.c
> @@ -47,6 +47,10 @@ static void *image_load(struct kimage *image,
> struct kexec_segment *kernel_segment;
> int ret;
>
> + /* We don't support crash kernels yet. */
> + if (image->type == KEXEC_TYPE_CRASH)
> + return ERR_PTR(-EOPNOTSUPP);
> +
> /*
> * We require a kernel with an unambiguous Image header. Per
> * Documentation/booting.txt, this is the case when image_size
> --
> 2.19.2
>

TBH, I would have loved the complete 'kexec_file_load' support to have
made way upstream for arm64 as well, but I understand that it is stuck
till we have an agreement on the DT side of things. So, refusing a
crash image loading in such a case makes sense for now.

So, please feel free to add:
Reviewed-by: Bhupesh Sharma <[email protected]>

Thanks,
Bhupesh