Hi,
First, thanks for fixing this!
So, I've been trying to find a failure beyond the printouts being wrong
for the past couple days, and largely failed to duplicate a functional
issue.
That said, this does fix a couple printouts I checked on 32-bit 686
debian 12. Generally I think the code is ok as well and have reviewed it
but i'm going to withhold the review by tag because I still think the
commit messages could use some cleanup.
Tested-by: Jeremy Linton <[email protected]>
Thanks again.
On 8/19/25 4:30 PM, Askar Safin wrote:
TL;DR: this patchset fixes regression, introduced in aecc554e7b.
This patchset should be backported to all distributions, which packaged
v2.0.31, otherwise
kexec doesn't work at all at my laptop with pretty common setup with v2.0.31.
v2.0.31 is broken without this patchset.
See details of this bug on my laptop in first commit message.
Okay, why the bug happens? I suspect this is because of "%lux"
string specifiers, which are totally wrong. The author meant "print (and scan) in
hexademical"
here, but this specifier prints (and scans) number in decimal, followed by literal
"x". Oops.
And this seems to break kexec.
The bug reproduces on kexec-tools aecc554e7ba , but
doesn't reproduce on kexec-tools 6aecc32c6db .
I. e. it is regression, introduced by aecc554e7ba .
Okay, how to fix this? Well, this is not easy. In 07821da7cf and d2f4297166
Andy Shevchenko
observed compilation warnings, when %lx is used with uint64_t, so he replaced
%lx with %llx.
Then in aecc554e7b Jeremy Linton observed warnings with %llx and replaced it
with %lux.
(Yes, C is nightmare.)
So, uint64_t is sometimes defined as long unsigned, and thus needs %lx, and
sometimes as
long long unsigned and thus needs %llx.
How to fix this once and for all?
I see three ways.
1. uint64_t a; printf ("%llx", (unsigned long long)a);
2. uint64_t a; printf ("%" PRIx64, a);
3. uint64_t a; printf ("%w64x", a);
I think that %w64x is beautiful, but it causes compilation warnings on clang.
(Facepalm.)
Also it was introduced in C23, which is too young.
"(unsigned long long)a" is the best in my opinion. This is what I used in v1.
But Andy said to me that POD conversions are evil.
PRIx64 is ugly, but this is the only option left. So this is what I used in
this version.
Also this patchset fixes other misc things. See commit messages for details.
I tested on my laptop that this patchset actually fixes the bug.
v1:
https://lore.kernel.org/kexec/[email protected]/
v2: https://lore.kernel.org/kexec/[email protected]/
v3:
https://lore.kernel.org/kexec/[email protected]/
Changes since v1:
* Addressed Andy's comments
* I reproduced the bug (and tested the fix) on slightly different versions of
Linux
and linux-firmware (see first commit message)
Changes since v2:
* Commit messages only
Changes since v3:
* Commit messages only
Askar Safin (6):
Fix printf string specifiers, otherwise kexec doesn't work on my
laptop
kexec/kexec-elf-exec.c: Replace %lux with %lx
kexec/arch/i386/x86-linux-setup.c: replace %d with %u
util_lib/elf_info.c: fix typo: prink -> printk
kexec/arch/i386/kexec-x86-common.c: remove duplicate <stdio.h>
kexec/arch/arm64/crashdump-arm64.c: remove extra whitespace
kexec/arch/arm64/crashdump-arm64.c | 2 +-
kexec/arch/i386/crashdump-x86.c | 3 ++-
kexec/arch/i386/kexec-x86-common.c | 4 ++--
kexec/arch/i386/x86-linux-setup.c | 3 ++-
kexec/kexec-elf-exec.c | 2 +-
util_lib/elf_info.c | 9 +++++----
6 files changed, 13 insertions(+), 10 deletions(-)
