[PATCH v1] kexec-tools/s390: Improve initrd memory reservation

[Ilya Leoshkevich]

On 8/28/26 17:22, Joseph Cathcart wrote:
> Improve memory reservation for initrd. Remove unnecessary initrd
> range from memory_ranges[]. Initrd currently falls between
> "System RAM" ranges, and is therefore already reserved.


What if this changes in the future? Isn't the current approach better then?


> Refactor so that arch_reuse_initrd() is responsible for populating
> retained_initrd_base and retained_initrd_size, rather than
> get_memory_ranges_s390().
>
> Add die() case for when user uses --reuseinitrd while booting a
> crash kernel.
>
> Signed-off-by: Joseph Cathcart <cathcartj08 at gmail.com>
> ---
>   kexec/arch/s390/kexec-image.c |  8 ++++++--
>   kexec/arch/s390/kexec-s390.c  | 19 ++++++++++---------
>   kexec/arch/s390/kexec-s390.h  |  2 +-
>   3 files changed, 17 insertions(+), 12 deletions(-)
>
> diff --git a/kexec/arch/s390/kexec-image.c b/kexec/arch/s390/kexec-image.c
> index 1a59a32..a9e5442 100644
> --- a/kexec/arch/s390/kexec-image.c
> +++ b/kexec/arch/s390/kexec-image.c
> @@ -79,7 +79,8 @@ int image_s390_load_file(int argc, char **argv, struct kexec_info *info)
>   	}
>   
>   	if (reuse_initrd)
> -		die("--reuseinitrd not supported with --kexec-file-syscall. Please use --kexec-syscall\n");
> +		die("--reuseinitrd not supported with --kexec-file-syscall. "
> +			"Please use --kexec-syscall\n");


Not sure what the kexec-tools rules there are (I see some exceptions in e.g.
fs2dt.c), but in the kernel string literals must not be split for ease of
grepping.


>   
>   	if (ramdisk) {
>   		info->initrd_fd = open(ramdisk, O_RDONLY);
> @@ -143,6 +144,9 @@ image_s390_load(int argc, char **argv, const char *kernel_buf,
>   	if (ramdisk && reuse_initrd)
>   		die("Can't specify --ramdisk or --initrd with --reuseinitrd\n");
>   
> +	if (reuse_initrd && (info->kexec_flags & KEXEC_ON_CRASH))
> +		die("Can't specify --reuseinitrd while booting a crash kernel\n");
> +

I agree that trying to reuse initrd for a crash kernel makes little sense on
popular distros, but in theory one can create a universal initrd. So is 
there a
technical reason to prohibit this?

>   	if (info->kexec_flags & KEXEC_ON_CRASH) {
>   		if (parse_iomem_single("Crash kernel\n", &crash_base,
>   				       &crash_end))
> @@ -173,7 +177,7 @@ image_s390_load(int argc, char **argv, const char *kernel_buf,
>   				  ramdisk_origin, ramdisk_len);
>   	} else if (reuse_initrd) {
>   		ramdisk_origin = retained_initrd_base;
> -		ramdisk_len = retained_initrd_size;
> +		ramdisk_len = (off_t)retained_initrd_size;

I guess this is to placate -Wsign-conversion?
It would be good to have this mentioned in the commit message.

>   	}
>   	if (info->kexec_flags & KEXEC_ON_CRASH) {
>   		if (load_crashdump_segments(info, crash_base, crash_end))
> diff --git a/kexec/arch/s390/kexec-s390.c b/kexec/arch/s390/kexec-s390.c
> index 4e77fc4..fd0d87e 100644
> --- a/kexec/arch/s390/kexec-s390.c
> +++ b/kexec/arch/s390/kexec-s390.c
> @@ -26,10 +26,17 @@
>   
>   static struct memory_range memory_range[MAX_MEMORY_RANGES];
>   unsigned long long retained_initrd_base, retained_initrd_size;
> -unsigned int reuse_initrd = 0;
> +unsigned char reuse_initrd;
>   
>   void arch_reuse_initrd(void)
>   {
> +	uint64_t start, end;
> +
> +	if (parse_iomem_single("initrd\n", &start, &end))
> +		die("Couldn't find initrd entry in /proc/iomem\n");
> +
> +	retained_initrd_base = (unsigned long long)start;
> +	retained_initrd_size = (unsigned long long)end - retained_initrd_base + 1;
>   	reuse_initrd = 1;
>   }


What is the benefit of moving this here? I see that there is no 
consensus among
the other architectures.

[...]