When building the kernel with Clang ThinLTO enabled, the compiler
can mangle static variable names by appending suffixes such as
".llvm.<hash>" to prevent naming collisions across translation units.
This name mangling breaks the section mismatch whitelisting in modpost.
modpost relies on glob patterns (e.g., "*_ops" or "*_probe") to identify
safe references between permanent data and initialization code. Because
the LTO suffix modifies the end of the symbol name, legitimately
whitelisted structures fail the match, resulting in false positive
warnings.
For example, a static pernet_operations struct triggers the following:
WARNING: modpost: vmlinux: section mismatch in reference: \
ping_v4_net_ops.llvm.5641696707737373282 (section: .data) -> \
ping_v4_proc_init_net (section: .init.text)
Fix this by ignoring "*_ops.llvm.*" in "from" symbol names (the same
as "*_ops").
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202606111233.kM8oo8Df-lkp@intel.com/
Signed-off-by: Rong Xu <xur@google.com>
Link: https://patch.msgid.link/20260617224623.1346309-1-xur@google.com
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Modpost emits hundreds of warnings when using Clang to build for ARCH=um
and CONFIG_GCOV=y. e.g.:
vmlinux (__llvm_covfun): unexpected non-allocatable section.
Did you forget to use "ax"/"aw" in a .S file?
Note that for example <linux/init.h> contains
section definitions for use in .S files.
For example, when we use LLVM for a kunit user mode build with coverage:
python3 tools/testing/kunit/kunit.py build --make_options LLVM=1 \
--kunitconfig=tools/testing/kunit/configs/default.config \
--kunitconfig=tools/testing/kunit/configs/coverage_uml.config
The behaviour occurs when building the kernel for ARCH=um with code
coverage enabled. The warnings come from modpost's check_sec_ref
function, which ensures no sections reference others that will be
discarded. covfun and covmap sections must reference __init and __exit
sections to collect coverage data, triggering the modpost warning.
To suppress these warnings, these section names have been added to
modpost's whitelist. This is unlikely to suppress legitimate warnings as
Clang will only insert these sections when building with coverage, and
can be assumed to manage these references safely.
Signed-off-by: James Lee <james@codeconstruct.com.au>
Link: https://patch.msgid.link/20260604-dev-coverage-patch-v1-1-9f9368253cb4@codeconstruct.com.au
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Add distributed ThinLTO build support for the Linux kernel.
This new mode offers several advantages: (1) Increased
flexibility in handling user-specified build options.
(2) Improved user-friendliness for developers. (3) Greater
convenience for integrating with objtool and livepatch.
Note that "distributed" in this context refers to a term
that differentiates in-process ThinLTO builds by invoking
backend compilation through the linker, not necessarily
building in distributed environments.
Distributed ThinLTO is enabled via the
`CONFIG_LTO_CLANG_THIN_DIST` Kconfig option. For example:
> make LLVM=1 defconfig
> scripts/config -e LTO_CLANG_THIN_DIST
> make LLVM=1 oldconfig
> make LLVM=1 vmlinux -j <..>
The build flow proceeds in four stages:
1. Perform FE compilation, mirroring the in-process ThinLTO mode.
2. Thin-link the generated IR files and object files.
3. Find all IR files and perform BE compilation, using the flags
stored in the .*.o.cmd files.
4. Link the BE results to generate the final vmlinux.o.
NOTE: This patch currently implements the build for the main kernel
image (vmlinux) only. Kernel module support is planned for a
subsequent patch.
Tested on the following arch: x86, arm64, loongarch, and
riscv.
The earlier implementation details can be found here:
https://discourse.llvm.org/t/rfc-distributed-thinlto-build-for-kernel/85934
Signed-off-by: Rong Xu <xur@google.com>
Co-developed-by: Masahiro Yamada <masahiroy@kernel.org>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Tested-by: Piotr Gorski <piotrgorski@cachyos.org>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Link: https://patch.msgid.link/20260529185347.2418373-4-xur@google.com
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
buf_printf() uses a fixed-size stack buffer. vsnprintf() returns the
number of bytes that *would* have been written to that buffer, which can
be larger than the size of said buffer if the formatted string is too
long.
The problem is that whenever this happens buf_printf() currently passes
this length, unchecked, to buf_write(), which silently reads past the
stack buffer and copies invalid data into the output buffer.
Fix this by detecting vsnprintf() failures and truncations before
appending to the output buffer, and report a fatal error instead of
producing corrupt symbol names.
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
Link: https://patch.msgid.link/20260527-nova-exports-v2-1-06de4c556d55@nvidia.com
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Pull module updates from Sami Tolvanen:
"Kernel symbol flags:
- Replace the separate *_gpl symbol sections (__ksymtab_gpl and
__kcrctab_gpl) with a unified symbol table and a new __kflagstab
section.
This section stores symbol flags, such as the GPL-only flag, as an
8-bit bitset for each exported symbol. This is a cleanup that
simplifies symbol lookup in the module loader by avoiding table
fragmentation and will allow a cleaner way to add more flags later
if needed.
Module signature UAPI:
- Move struct module_signature to the UAPI headers to allow reuse by
tools outside the kernel proper, such as kmod and
scripts/sign-file.
This also renames a few constants for clarity and drops unused
signature types as preparation for hash-based module integrity
checking work that's in progress.
Sysfs:
- Add a /sys/module/<module>/import_ns sysfs attribute to show the
symbol namespaces imported by loaded modules.
This makes it easier to verify driver API access at runtime on
systems that care about such things (e.g. Android).
Cleanups and fixes:
- Force sh_addr to 0 for all sections in module.lds. This prevents
non-zero section addresses when linking modules with 'ld.bfd -r',
which confused elfutils.
- Fix a memory leak of charp module parameters on module unload when
the kernel is configured with CONFIG_SYSFS=n.
- Override the -EEXIST error code returned by module_init() to
userspace. This prevents confusion with the errno reserved by the
module loader to indicate that a module is already loaded.
- Simplify the warning message and drop the stack dump on positive
returns from module_init().
- Drop unnecessary extern keywords from function declarations and
synchronize parse_args() arguments with their implementation"
* tag 'modules-7.1-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/modules/linux: (23 commits)
module: Simplify warning on positive returns from module_init()
module: Override -EEXIST module return
documentation: remove references to *_gpl sections
module: remove *_gpl sections from vmlinux and modules
module: deprecate usage of *_gpl sections in module loader
module: use kflagstab instead of *_gpl sections
module: populate kflagstab in modpost
module: add kflagstab section to vmlinux and modules
module: define ksym_flags enumeration to represent kernel symbol flags
selftests/bpf: verify_pkcs7_sig: Use 'struct module_signature' from the UAPI headers
sign-file: use 'struct module_signature' from the UAPI headers
tools uapi headers: add linux/module_signature.h
module: Move 'struct module_signature' to UAPI
module: Give MODULE_SIG_STRING a more descriptive name
module: Give 'enum pkey_id_type' a more specific name
module: Drop unused signature types
extract-cert: drop unused definition of PKEY_ID_PKCS7
docs: symbol-namespaces: mention sysfs attribute
module: expose imported namespaces via sysfs
module: Remove extern keyword from param prototypes
...
Read kflagstab section for vmlinux and modules to determine whether
kernel symbols are GPL only.
This patch eliminates the need for fragmenting the ksymtab for infering
the value of GPL-only symbol flag, henceforth stop populating *_gpl
versions of the ksymtab and kcrctab in modpost.
Signed-off-by: Siddharth Nayyar <sidnayyar@google.com>
Reviewed-by: Petr Pavlu <petr.pavlu@suse.com>
Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
This patch adds the ability to create entries for kernel symbol flag
bitsets in kflagstab. Modpost populates only the GPL-only flag for now.
Signed-off-by: Siddharth Nayyar <sidnayyar@google.com>
Reviewed-by: Petr Pavlu <petr.pavlu@suse.com>
Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
A recent strengthening of -Wunused-but-set-variable (enabled with -Wall)
in clang under a new subwarning, -Wunused-but-set-global, points out an
unused static global variable in scripts/mod/modpost.c:
scripts/mod/modpost.c:59:13: error: variable 'extra_warn' set but not used [-Werror,-Wunused-but-set-global]
59 | static bool extra_warn;
| ^
This variable has been unused since commit 6c6c1fc09d ("modpost:
require a MODULE_DESCRIPTION()") but that is expected, as there are
currently no extra warnings at W=1 right now. Declare the variable with
the unused attribute to make it clear to the compiler that this variable
may be unused.
Cc: stable@vger.kernel.org
Fixes: 6c6c1fc09d ("modpost: require a MODULE_DESCRIPTION()")
Link: https://patch.msgid.link/20260325-modpost-extra_warn-unused-but-set-global-v1-1-2e84003b7e81@kernel.org
Reviewed-by: Nicolas Schier <nsc@kernel.org>
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Several symbol patterns used to be whitelisted to allow drivers to refer
to functions annotated with __devinit and __devexit, which have since
been removed.
Commit e1dc1bfe5b ("modpost: remove more symbol patterns from the
section check whitelist") removed most of these patterns but left
'*_probe' after a reported warning in an irqchip driver.
Turns out that was indeed an incorrect reference which has now been
fixed by commit 9b685058ca ("irqchip/qcom-irq-combiner: Fix section
mismatch").
A recently added clocksource driver also relies on this suffix to
suppress another valid warning, and that is being fixed separately. [1]
Note that drivers with valid reasons for suppressing the warnings can
use the __ref macros.
Link: https://lore.kernel.org/lkml/20251017054943.7195-1-johan@kernel.org/ [1]
Signed-off-by: Johan Hovold <johan@kernel.org>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Nathan Chancellor <nathan@kernel.org>
Link: https://patch.msgid.link/20251020091613.22562-1-johan@kernel.org
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
In preparation for klp-build livepatch module creation tooling,
suppress warnings for unresolved references to linker-generated
__start_* and __stop_* section bounds symbols.
These symbols are expected to be undefined when modpost runs, as they're
created later by the linker.
Cc: Masahiro Yamada <masahiroy@kernel.org>
Acked-by: Petr Mladek <pmladek@suse.com>
Tested-by: Joe Lawrence <joe.lawrence@redhat.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org>
For some modules, modalias is generated using the modpost utility and
the section is added to the module file.
When a module is added inside vmlinux, modpost does not generate
modalias for such modules and the information is lost.
As a result kmod (which uses modules.builtin.modinfo in userspace)
cannot determine that modalias is handled by a builtin kernel module.
$ cat /sys/devices/pci0000:00/0000:00:14.0/modalias
pci:v00008086d0000A36Dsv00001043sd00008694bc0Csc03i30
$ modinfo xhci_pci
name: xhci_pci
filename: (builtin)
license: GPL
file: drivers/usb/host/xhci-pci
description: xHCI PCI Host Controller Driver
Missing modalias "pci:v*d*sv*sd*bc0Csc03i30*" which will be generated by
modpost if the module is built separately.
To fix this it is necessary to generate the same modalias for vmlinux as
for the individual modules. Fortunately '.vmlinux.export.o' is already
generated from which '.modinfo' can be extracted in the same way as for
vmlinux.o.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Signed-off-by: Alexey Gladkov <legion@kernel.org>
Tested-by: Stephen Rothwell <sfr@canb.auug.org.au>
Reviewed-by: Nicolas Schier <nsc@kernel.org>
Link: https://patch.msgid.link/28d4da3b0e3fc8474142746bcf469e03752c3208.1758182101.git.legion@kernel.org
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Explicitly adding MODULE_IMPORT_NS("module:...") is not allowed.
Currently, this is only checked at run time. That is, when such a
module is loaded, an error message like the following is shown:
foo: module tries to import module namespace: module:bar
Obviously, checking this at compile time improves usability.
In such a case, modpost will report the following error at compile time:
ERROR: modpost: foo: explicitly importing namespace "module:bar" is not allowed.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Instead of only accepting "module:${name}", extend it with a comma
separated list of module names and add tail glob support.
That is, something like: "module:foo-*,bar" is now possible.
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Reviewed-by: Petr Pavlu <petr.pavlu@suse.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Designate the "module:${modname}" symbol namespace to mean: 'only
export to the named module'.
Notably, explicit imports of anything in the "module:" space is
forbidden.
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Reviewed-by: Petr Pavlu <petr.pavlu@suse.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Slight cleanup by using a for() loop instead of while(). This makes it
clearer what is the iteration and what is the actual work done.
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
... and don't error out so hard on missing module descriptions.
Before commit 6c6c1fc09d ("modpost: require a MODULE_DESCRIPTION()")
we used to warn about missing module descriptions, but only when
building with extra warnigns (ie 'W=1').
After that commit the warning became an unconditional hard error.
And it turns out not all modules have been converted despite the claims
to the contrary. As reported by Damian Tometzki, the slub KUnit test
didn't have a module description, and apparently nobody ever really
noticed.
The reason nobody noticed seems to be that the slub KUnit tests get
disabled by SLUB_TINY, which also ends up disabling a lot of other code,
both in tests and in slub itself. And so anybody doing full build tests
didn't actually see this failre.
So let's disable SLUB_TINY for build-only tests, since it clearly ends
up limiting build coverage. Also turn the missing module descriptions
error back into a warning, but let's keep it around for non-'W=1'
builds.
Reported-by: Damian Tometzki <damian@riscv-rocks.de>
Link: https://lore.kernel.org/all/01070196099fd059-e8463438-7b1b-4ec8-816d-173874be9966-000000@eu-central-1.amazonses.com/
Cc: Masahiro Yamada <masahiroy@kernel.org>
Cc: Jeff Johnson <jeff.johnson@oss.qualcomm.com>
Fixes: 6c6c1fc09d ("modpost: require a MODULE_DESCRIPTION()")
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Since commit 1fffe7a34c ("script: modpost: emit a warning when the
description is missing"), a module without a MODULE_DESCRIPTION() has
resulted in a warning with make W=1. Since that time, all known
instances of this issue have been fixed. Therefore, now make it an
error if a MODULE_DESCRIPTION() is not present.
Signed-off-by: Jeff Johnson <jeff.johnson@oss.qualcomm.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
The logic to retrieve the basename appears multiple times.
Factor out the common pattern into a helper function.
I copied kbasename() from include/linux/string.h and renamed it
to get_basename().
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nicolas Schier <nicolas@fjasle.eu>
Linus observed that the symbol_request(utf8_data_table) call fails when
CONFIG_UNICODE=y and CONFIG_TRIM_UNUSED_KSYMS=y.
symbol_get() relies on the symbol data being present in the ksymtab for
symbol lookups. However, EXPORT_SYMBOL_GPL(utf8_data_table) is dropped
due to CONFIG_TRIM_UNUSED_KSYMS, as no module references it in this case.
Probably, this has been broken since commit dbacb0ef67 ("kconfig option
for TRIM_UNUSED_KSYMS").
This commit addresses the issue by leveraging modpost. Symbol names
passed to symbol_get() are recorded in the special .no_trim_symbol
section, which is then parsed by modpost to forcibly keep such symbols.
The .no_trim_symbol section is discarded by the linker scripts, so there
is no impact on the size of the final vmlinux or modules.
This commit cannot resolve the issue for direct calls to __symbol_get()
because the symbol name is not known at compile-time.
Although symbol_get() may eventually be deprecated, this workaround
should be good enough meanwhile.
Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
If you know that your kernel modules will only ever be loaded by a newer
kernel, you can disable BASIC_MODVERSIONS to save space. This also
allows easy creation of test modules to see how tooling will respond to
modules that only have the new format.
Signed-off-by: Matthew Maurer <mmaurer@google.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Generate both the existing modversions format and the new extended one
when running modpost. Presence of this metadata in the final .ko is
guarded by CONFIG_EXTENDED_MODVERSIONS.
We no longer generate an error on long symbols in modpost if
CONFIG_EXTENDED_MODVERSIONS is set, as they can now be appropriately
encoded in the extended section. These symbols will be skipped in the
previous encoding. An error will still be generated if
CONFIG_EXTENDED_MODVERSIONS is not set.
Reviewed-by: Sami Tolvanen <samitolvanen@google.com>
Signed-off-by: Matthew Maurer <mmaurer@google.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
I do not think the '#' flag is useful here because adding the explicit
'0x' is clearer. Add the '0' flag to zero-pad the CRC values.
This change gives better alignment in the generated *.mod.c files.
There is no impact to the compiled modules.
[Before]
$ grep -A5 modversion_info fs/efivarfs/efivarfs.mod.c
static const struct modversion_info ____versions[]
__used __section("__versions") = {
{ 0x907d14d, "blocking_notifier_chain_register" },
{ 0x53d3b64, "simple_inode_init_ts" },
{ 0x65487097, "__x86_indirect_thunk_rax" },
{ 0x122c3a7e, "_printk" },
[After]
$ grep -A5 modversion_info fs/efivarfs/efivarfs.mod.c
static const struct modversion_info ____versions[]
__used __section("__versions") = {
{ 0x0907d14d, "blocking_notifier_chain_register" },
{ 0x053d3b64, "simple_inode_init_ts" },
{ 0x65487097, "__x86_indirect_thunk_rax" },
{ 0x122c3a7e, "_printk" },
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
With the latest binutils, modpost fails with a bus error on some
architectures such as ARM and sparc64.
Since binutils commit 1f1b5e506bf0 ("bfd/ELF: restrict file alignment
for object files"), the byte offset to each section (sh_offset) in
relocatable ELF is no longer guaranteed to be aligned.
modpost parses MODULE_DEVICE_TABLE() data structures, which are usually
located in the .rodata section. If it is not properly aligned, unaligned
access errors may occur.
To address the issue, this commit imports the get_unaligned() helper
from include/linux/unaligned.h.
The get_unaligned_native() helper caters to the endianness in addition
to handling the unaligned access.
I slightly refactored do_pcmcia_entry() and do_input() to avoid writing
back to an unaligned address. (We would need the put_unaligned() helper
to do that.)
The addend_*_rel() functions need similar adjustments because the .text
sections are not aligned either.
It seems that the .symtab, .rel.* and .rela.* sections are still aligned.
Keep normal pointer access for these sections to avoid unnecessary
performance costs.
Reported-by: Paulo Pisati <paolo.pisati@canonical.com>
Reported-by: Matthias Klose <doko@debian.org>
Closes: https://sourceware.org/bugzilla/show_bug.cgi?id=32435
Reported-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Closes: https://sourceware.org/bugzilla/show_bug.cgi?id=32493
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Tested-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
