kexec crash kernel boot failure on arm64

[Mark Rutland]

On Fri, Jul 24, 2015 at 03:07:24AM +0100, Anurup m wrote:
> Hi All,
> 
> There is a problem observed with crash kernel boot in kdump on arm64.

With which kernel? Mainline doesn't have kexec or kdump support for
arm64.

> On arm64 hardware board, when I enable the purgatory segment, the crash kernel doesnot boot.
> When checked with trace32, it is observed that the control comes to purgatory_start routine, 
> but the instructions are seen as UNDEF and the boot hangs. But when I took the memory dump, the
> contents were seen as proper(matching with the purgatory_start code).
> 
> I did some experiments to analyze this issue. Tried changing the Load order of kexec segments and
> observed results as below
> ------------------------------------------------------------------------------
>     Segments Load order						crash kernel boot status
>     --------------------			            -------------------------
> 1) crash kernel, initrd, dtb. Elfcorehdr		-	 Boot Success - without purgatory
> 2) crash kernel, initrd, dtb. Purgatory, elfcorehdr -  HUNG as control does not reach purgatory segment.
> 3) crash kernel, elfcorehdr, purgatory, dtb, initrd -	 Boot Success
> 4) crash kernel, initrd, dtb, purgatory, elfcorehdr, - Boot Success
>    an extra segment(~20M)).
> 
> From this I could infer that If I load a larger segment after purgatory (in the load order), the crash
> Kernel boots. i.e. memory sync is taking some time.
> 
> So to clarify if memory sync is the Issue, I tried flush the data cache after writing the kexec segments.
> 
> kernel/kexec.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/kernel/kexec.c b/kernel/kexec.c index 7bb25f0..ca36aa0 100644
> --- a/kernel/kexec.c
> +++ b/kernel/kexec.c
> @@ -1176,6 +1176,10 @@ static int kimage_load_crash_segment(struct kimage *image,
>  		else
>  			result = copy_from_user(ptr, buf, uchunk);
>  		kexec_flush_icache_page(page);
> +		/* Flush Dcache to make sure it is push to DRAM
> +		 * This is added as workaround for crash kernel
> +		 * boot failure */
> +		__flush_dcache_area((__force void *)ptr, uchunk);
>  		kunmap(page);
>  		if (result) {
>  			result = -EFAULT;
> 
> With the above change, control could reach purgatory_start, but this time it loops due to sha256_digest
> Verify failure. It is able to boot to crash kernel (after comment verify_sha256_digest)
> 
> What could be the possible reasons for this issue? Please share your comments.

The only verify_sha256_digest I can see in the kernel tree is under
arch/x86. Without knowing what kernel you're running, it's not possible
to answer your question.

Mark.