On 2025/07/01 22:55, Dave Martin wrote:
This series aims to clean up an aspect of coredump generation:ELF coredumps contain a set of notes describing the state of machine registers and other information about the dumped process. Notes are identified by a numeric identifier n_type and a "name" string, although this terminology is somewhat misleading. Officially, the "name" of a note is really an "originator" or namespace identifier that indicates how to interpret n_type [1], although in practice it is often used more loosely. Either way, each kind of note needs _both_ a specific "name" string and a specific n_type to identify it robustly. To centralise this knowledge in one place and avoid the need for ad-hoc code to guess the correct name for a given note, commit 7da8e4ad4df0 ("elf: Define note name macros") [2] added an explicit NN_<foo> #define in elf.h to give the name corresponding to each named note type NT_<foo>. Now that the note name for each note is specified explicitly, the remaining guesswork for determining the note name for common and arch-specific regsets in ELF core dumps can be eliminated. This series aims to do just that: * Patch 2 adds a user_regset field to specify the note name, and a helper macro to populate it correctly alongside the note type. * Patch 3 ports away the ad-hoc note names in the common coredump code. * Patches 4-22 make the arch-specific changes. (This is pretty mechanical for most arches.) * The final patch adds a WARN() when no note name is specified, and simplifies the fallback guess. This should only be applied when all arches have ported across. See the individual patches for details. Testing: * x86, arm64: Booted in a VM and triggered a core dump with no WARN(), and verified that the dumped notes are the same. * arm: Build-tested only (for now). * Other arches: not tested yet Any help with testing is appreciated. If the following generates the same notes (as dumped by readelf -n core) and doesn't trigger a WARN, then we are probably good. $ sleep 60 & $ kill -QUIT $! (Register content might differ between runs, but it should be safe to ignore that -- this series only deals with the note names and types.) Cheers ---Dave [1] System V Application Binary Interface, Edition 4.1, Section 5 (Program Loading and Dynamic Linking) -> "Note Section" https://refspecs.linuxfoundation.org/elf/gabi41.pdf [2] elf: Define note name macros https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/include/uapi/linux/elf.h?id=7da8e4ad4df0dd12f37357af62ce1b63e75ae2e6 Dave Martin (23): regset: Fix kerneldoc for struct regset_get() in user_regset regset: Add explicit core note name in struct user_regset binfmt_elf: Dump non-arch notes with strictly matching name and type ARC: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names ARM: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names arm64: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names csky: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names hexagon: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names LoongArch: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names m68k: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names MIPS: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names nios2: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names openrisc: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names parisc: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names powerpc/ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names riscv: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names s390/ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names sh: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names sparc: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names x86/ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names um: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names xtensa: ptrace: Use USER_REGSET_NOTE_TYPE() to specify regset note names binfmt_elf: Warn on missing or suspicious regset note names arch/arc/kernel/ptrace.c | 4 +- arch/arm/kernel/ptrace.c | 6 +- arch/arm64/kernel/ptrace.c | 52 ++++++++--------- arch/csky/kernel/ptrace.c | 4 +- arch/hexagon/kernel/ptrace.c | 2 +- arch/loongarch/kernel/ptrace.c | 16 ++--- arch/m68k/kernel/ptrace.c | 4 +- arch/mips/kernel/ptrace.c | 20 +++---- arch/nios2/kernel/ptrace.c | 2 +- arch/openrisc/kernel/ptrace.c | 4 +- arch/parisc/kernel/ptrace.c | 8 +-- arch/powerpc/kernel/ptrace/ptrace-view.c | 74 ++++++++++++------------ arch/riscv/kernel/ptrace.c | 12 ++-- arch/s390/kernel/ptrace.c | 42 +++++++------- arch/sh/kernel/ptrace_32.c | 4 +- arch/sparc/kernel/ptrace_32.c | 4 +- arch/sparc/kernel/ptrace_64.c | 8 +-- arch/x86/kernel/ptrace.c | 22 +++---- arch/x86/um/ptrace.c | 10 ++-- arch/xtensa/kernel/ptrace.c | 4 +- fs/binfmt_elf.c | 36 +++++++----- fs/binfmt_elf_fdpic.c | 17 +++--- include/linux/regset.h | 12 +++- 23 files changed, 194 insertions(+), 173 deletions(-) base-commit: 86731a2a651e58953fc949573895f2fa6d456841
For the whole series: Reviewed-by: Akihiko Odaki <od...@rsg.ci.i.u-tokyo.ac.jp> Regards, Akihiko Odaki _______________________________________________ linux-snps-arc mailing list linux-snps-arc@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-snps-arc
