Support creation of module BTF along with distilled base BTF;
the latter is stored in a .BTF.base ELF section and supplements
split BTF references to base BTF with information about base types,
allowing for later relocation of split BTF with a (possibly
changed) base.  resolve_btfids detects the presence of a .BTF.base
section and will use it instead of the base BTF it is passed in
BTF id resolution.

Modules will be built with a distilled .BTF.base section for external
module build, i.e.

make -C. -M=path2/module

...while in-tree module build as part of a normal kernel build will
not generate distilled base BTF; this is because in-tree modules
change with the kernel and do not require BTF relocation for the
running vmlinux.

Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Reviewed-by: Eduard Zingerman <eddyz87@gmail.com>
Link: https://lore.kernel.org/bpf/20240620091733.1967885-6-alan.maguire@oracle.com

Instead of filtering out the GCOV and KCSAN flags, let's set GCOV_PROFILE
and KCSAN_SANITIZE to 'n', as in other Makefiles.

Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Kees Cook <keescook@chromium.org>
Tested-by: Roberto Sassu <roberto.sassu@huawei.com>

When KCSAN and CONSTRUCTORS are enabled, one can trigger the

  "Unpatched return thunk in use. This should not happen!"

catch-all warning.

Usually, when objtool runs on the .o objects, it does generate a section
.return_sites which contains all offsets in the objects to the return
thunks of the functions present there. Those return thunks then get
patched at runtime by the alternatives.

KCSAN and CONSTRUCTORS add this to the object file's .text.startup
section:

  -------------------
  Disassembly of section .text.startup:

  ...

  0000000000000010 <_sub_I_00099_0>:
    10:   f3 0f 1e fa             endbr64
    14:   e8 00 00 00 00          call   19 <_sub_I_00099_0+0x9>
                          15: R_X86_64_PLT32      __tsan_init-0x4
    19:   e9 00 00 00 00          jmp    1e <__UNIQUE_ID___addressable_cryptd_alloc_aead349+0x6>
                          1a: R_X86_64_PLT32      __x86_return_thunk-0x4
  -------------------

which, if it is built as a module goes through the intermediary stage of
creating a <module>.mod.c file which, when translated, receives a second
constructor:

  -------------------
  Disassembly of section .text.startup:

  0000000000000010 <_sub_I_00099_0>:
    10:   f3 0f 1e fa             endbr64
    14:   e8 00 00 00 00          call   19 <_sub_I_00099_0+0x9>
                          15: R_X86_64_PLT32      __tsan_init-0x4
    19:   e9 00 00 00 00          jmp    1e <_sub_I_00099_0+0xe>
                          1a: R_X86_64_PLT32      __x86_return_thunk-0x4

  ...

  0000000000000030 <_sub_I_00099_0>:
    30:   f3 0f 1e fa             endbr64
    34:   e8 00 00 00 00          call   39 <_sub_I_00099_0+0x9>
                          35: R_X86_64_PLT32      __tsan_init-0x4
    39:   e9 00 00 00 00          jmp    3e <__ksymtab_cryptd_alloc_ahash+0x2>
                          3a: R_X86_64_PLT32      __x86_return_thunk-0x4
  -------------------

in the .ko file.

Objtool has run already so that second constructor's return thunk cannot
be added to the .return_sites section and thus the return thunk remains
unpatched and the warning rightfully fires.

Drop KCSAN flags from the mod.c generation stage as those constructors
do not contain data races one would be interested about.

Debugged together with David Kaplan <David.Kaplan@amd.com> and Nikolay
Borisov <nik.borisov@suse.com>.

Reported-by: Paul Menzel <pmenzel@molgen.mpg.de>
Closes: https://lore.kernel.org/r/0851a207-7143-417e-be31-8bf2b3afb57d@molgen.mpg.de


Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Tested-by: Paul Menzel <pmenzel@molgen.mpg.de> # Dell XPS 13
Reviewed-by: Nikolay Borisov <nik.borisov@suse.com>
Reviewed-by: Marco Elver <elver@google.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>

The '%.ko' rule in arch/*/Makefile.postlink does nothing but call the
'true' command.

Remove the unneeded code.

Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nicolas Schier <n.schier@avm.de>

With commit c1177979

 ("btf, scripts: Exclude Rust CUs with pahole")
we are now able to use pahole directly to identify Rust compilation
units (CUs) and exclude them from generating BTF debugging information
(when DEBUG_INFO_BTF is enabled).

And if pahole doesn't support the --lang-exclude flag, we can't enable
both RUST and DEBUG_INFO_BTF at the same time.

So, in any case, the script is_rust_module.sh is just redundant and we
can drop it.

NOTE: we may also be able to drop the "Rust loadable module" mark
inside Rust modules, but it seems safer to keep it for now to make sure
we are not breaking any external tool that may potentially rely on it.

Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
Reviewed-by: Nathan Chancellor <nathan@kernel.org>
Tested-by: Eric Curtin <ecurtin@redhat.com>
Reviewed-by: Eric Curtin <ecurtin@redhat.com>
Reviewed-by: Neal Gompa <neal@gompa.dev>
Reviewed-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Martin Rodriguez Reboredo <yakoyoku@gmail.com>
Acked-by: Daniel Xu <dxu@dxuuu.xyz>
Link: https://lore.kernel.org/r/20230704052136.155445-1-andrea.righi@canonical.com


[ Picked the `Reviewed-by`s from the old patch too. ]
Signed-off-by: Miguel Ojeda <ojeda@kernel.org>

With GCOV_PROFILE_ALL, Clang injects __llvm_gcov_* functions to each
object file, including the *.mod.o. As we filter out CC_FLAGS_CFI
for *.mod.o, the compiler won't generate type hashes for the
injected functions, and therefore indirectly calling them during
module loading trips indirect call checking.

Enabling CFI for *.mod.o isn't sufficient to fix this issue after
commit 0c3e806e ("x86/cfi: Add boot time hash randomization"),
as *.mod.o aren't processed by objtool, which means any hashes
emitted there won't be randomized. Therefore, in addition to
disabling CFI for *.mod.o, also disable GCOV, as the object files
don't otherwise contain any executable code.

Fixes: cf68fffb

 ("add support for Clang CFI")
Reported-by: Joe Fradley <joefradley@google.com>
Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>

The cmd-check macro compares $(cmd_$@) and $(cmd_$1), but a pitfall is
that you cannot use cmd_<target> as the variable name for the command.

For example, the following code will not work in the top Makefile
or ./Kbuild.

    quiet_cmd_foo = GEN     $@
          cmd_foo = touch $@

    targets += foo
    foo: FORCE
            $(call if_changed,foo)

In this case, both $@ and $1 are expanded to 'foo', so $(cmd_check)
is always empty.

We do not need to use the same prefix for cmd_$@ and cmd_$1.
Rename the former to savedcmd_$@.

Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nicolas Schier <nicolas@fjasle.eu>

scripts/Makefile.build replaces the suffix .o with .ko, then
scripts/Makefile.modpost calls the sed command to change .ko back
to the original .o suffix.

Instead of converting the suffixes back-and-forth, store the .o paths
in modules.order, and replace it with .ko in 'make modules_install'.

This avoids the unneeded sed command.

Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>

Since GNU Make 4.2, $(file ...) supports the read operater '<', which
is useful to read a file without forking a new process. No warning is
shown even if the input file is missing.

For older Make versions, it falls back to the cat command.

Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nicolas Schier <nicolas@fjasle.eu>
Reviewed-by: Alexander Lobakin <alexandr.lobakin@intel.com>
Tested-by: Alexander Lobakin <alexandr.lobakin@intel.com>

modules.order lists modules in the deterministic order (that is why
"modules order"), and there is no duplication in the list.

$(sort ) is pointless.

Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nicolas Schier <nicolas@fjasle.eu>

Currently, modpost is executed twice; first for vmlinux, second
for modules.

This commit merges them.

Current build flow
==================

  1) build obj-y and obj-m objects
    2) link vmlinux.o
      3) modpost for vmlinux
        4) link vmlinux
          5) modpost for modules
            6) link modules (*.ko)

The build steps 1) through 6) are serialized, that is, modules are
built after vmlinux. You do not get benefits of parallel builds when
scripts/link-vmlinux.sh is being run.

New build flow
==============

  1) build obj-y and obj-m objects
    2) link vmlinux.o
      3) modpost for vmlinux and modules
        4a) link vmlinux
        4b) link modules (*.ko)

In the new build flow, modpost is invoked just once.

vmlinux and modules are built in parallel. One exception is
CONFIG_DEBUG_INFO_BTF_MODULES=y, where modules depend on vmlinux.

Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Tested-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Nicolas Schier <nicolas@fjasle.eu>

Having most of the new files in place, we now enable Rust support
in the build system, including `Kconfig` entries related to Rust,
the Rust configuration printer and a few other bits.

Reviewed-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Tested-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Co-developed-by: Alex Gaynor <alex.gaynor@gmail.com>
Signed-off-by: Alex Gaynor <alex.gaynor@gmail.com>
Co-developed-by: Finn Behrens <me@kloenk.de>
Signed-off-by: Finn Behrens <me@kloenk.de>
Co-developed-by: Adam Bratschi-Kaye <ark.email@gmail.com>
Signed-off-by: Adam Bratschi-Kaye <ark.email@gmail.com>
Co-developed-by: Wedson Almeida Filho <wedsonaf@google.com>
Signed-off-by: Wedson Almeida Filho <wedsonaf@google.com>
Co-developed-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Co-developed-by: Sven Van Asbroeck <thesven73@gmail.com>
Si...

When CONFIG_LTO_CLANG=y, additional intermediate *.prelink.o is created
for each module. Also, objtool is postponed until LLVM IR is converted
to ELF.

CONFIG_X86_KERNEL_IBT works in a similar way to postpone objtool until
objects are merged together.

This commit stops generating *.prelink.o, so the build flow will look
similar with/without LTO.

The following figures show how the LTO build currently works, and
how this commit is changing it.

Current build flow
==================

 [1] single-object module

                                      $(LD)
           $(CC)                     +objtool              $(LD)
    foo.c --------------------> foo.o -----> foo.prelink.o -----> foo.ko
                              (LLVM IR)          (ELF)       |    (ELF)
                                                             |
                                                 foo.mod.o --/
                                                 (LLVM IR)

 [2] multi-object module
                                      $(LD)
           $(CC)         $(AR)       +objtool               $(LD)
    foo1.c -----> foo1.o -----> foo.o -----> foo.prelink.o -----> foo.ko
                           |  (archive)          (ELF)       |    (ELF)
    foo2.c -----> foo2.o --/                                 |
                 (LLVM IR)                       foo.mod.o --/
                                                 (LLVM IR)

  One confusion is that foo.o in multi-object module is an archive
  despite of its suffix.

New build flow
==============

 [1] single-object module

  Since there is only one object, there is no need to keep the LLVM IR.
  Use $(CC)+$(LD) to generate an ELF object in one build rule. When LTO
  is disabled, $(LD) is unneeded because $(CC) produces an ELF object.

               $(CC)+$(LD)+objtool              $(LD)
    foo.c ----------------------------> foo.o ---------> foo.ko
                                        (ELF)     |      (ELF)
                                                  |
                                      foo.mod.o --/
                                      (LLVM IR)

 [2] multi-object module

  Previously, $(AR) was used to combine LLVM IR files into an archive,
  but there was no technical reason to do so. Use $(LD) to merge them
  into a single ELF object.

                               $(LD)
             $(CC)            +objtool          $(LD)
    foo1.c ---------> foo1.o ---------> foo.o ---------> foo.ko
                                 |      (ELF)     |      (ELF)
    foo2.c ---------> foo2.o ----/                |
                     (LLVM IR)        foo.mod.o --/
                                      (LLVM IR)

Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nicolas Schier <nicolas@fjasle.eu>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Reviewed-by: Sami Tolvanen <samitolvanen@google.com>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com> # LLVM-14 (x86-64)
Acked-by: Josh Poimboeuf <jpoimboe@kernel.org>

Using new PAHOLE_FLAGS variable to pass extra arguments to
pahole for both vmlinux and modules BTF data generation.

Adding new scripts/pahole-flags.sh script that detect and
prints pahole options.

[ fixed issues found by kernel test robot ]

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20211029125729.70002-1-jolsa@kernel.org

This commit moves BTF ID lookup into the newly added registration
helper, in a way that the bbr, cubic, and dctcp implementation set up
their sets in the bpf_tcp_ca kfunc_btf_set list, while the ones not
dependent on modules are looked up from the wrapper function.

This lifts the restriction for them to be compiled as built in objects,
and can be loaded as modules if required. Also modify Makefile.modfinal
to call resolve_btfids for each module.

Note that since kernel kfunc_ids never overlap with module kfunc_ids, we
only match the owner for module btf id sets.

See following commits for background on use of:

 CONFIG_X86 ifdef:
 569c484f (bpf: Limit static tcp-cc functions in the .BTF_ids list to x86)

 CONFIG_DYNAMIC_FTRACE ifdef:
 7aae231a

 (bpf: tcp: Limit calling some tcp cc functions to CONFIG_DYNAMIC_FTRACE)

Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20211002011757.311265-6-memxor@gmail.com

With CONFIG_LTO_CLANG, we currently link modules into native
code just before modpost, which means with TRIM_UNUSED_KSYMS
enabled, we still look at the LLVM bitcode in the .o files when
generating the list of used symbols. As the bitcode doesn't
yet have calls to compiler intrinsics and llvm-nm doesn't see
function references that only exist in function-level inline
assembly, we currently need a whitelist for TRIM_UNUSED_KSYMS to
work with LTO.

This change moves module LTO linking to happen earlier, and
thus avoids the issue with LLVM bitcode and TRIM_UNUSED_KSYMS
entirely, allowing us to also drop the whitelist from
gen_autoksyms.sh.

Link: https://github.com/ClangBuiltLinux/linux/issues/1369


Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
Reviewed-by: Alexander Lobakin <alobakin@pm.me>
Tested-by: Alexander Lobakin <alobakin@pm.me>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>

The ccache tool can be used to speed up cross-compilation, by calling the
compiler and binutils through ccache. For example, following should work:

    $ export ARCH=arm64 CROSS_COMPILE="ccache aarch64-linux-gnu-"

    $ make M=drivers/gpu/drm/rockchip/

but pahole fails to extract the BTF info from DWARF, breaking the build:

      CC [M]  drivers/gpu/drm/rockchip//rockchipdrm.mod.o
      LD [M]  drivers/gpu/drm/rockchip//rockchipdrm.ko
      BTF [M] drivers/gpu/drm/rockchip//rockchipdrm.ko
    aarch64-linux-gnu-objcopy: invalid option -- 'J'
    Usage: aarch64-linux-gnu-objcopy [option(s)] in-file [out-file]
     Copies a binary file, possibly transforming it in the process
    ...
    make[1]: *** [scripts/Makefile.modpost:156: __modpost] Error 2
    make: *** [Makefile:1866: modules] Error 2

this fails because OBJCOPY is set to "ccache aarch64-linux-gnu-copy" and
later pahole is executed with the following command line:

    LLVM_OBJCOPY=$(OBJCOPY) $(PAHOLE) -J --btf_base vmlinux $@

which gets expanded to:

    LLVM_OBJCOPY=ccache aarch64-linux-gnu-objcopy pahole -J ...

instead of:

    LLVM_OBJCOPY="ccache aarch64-linux-gnu-objcopy" pahole -J ...

Fixes: 5f9ae91f

 ("kbuild: Build kernel module BTFs if BTF is enabled and pahole supports it")
Signed-off-by: Javier Martinez Canillas <javierm@redhat.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Link: https://lore.kernel.org/bpf/20210526215228.3729875-1-javierm@redhat.com

I do not see a good reason why only the libelf development package must
be so carefully checked.

Kbuild generally does not check host tools or libraries.

For example, x86_64 defconfig fails to build with no libssl development
package installed.

scripts/extract-cert.c:21:10: fatal error: openssl/bio.h: No such file or directory
   21 | #include <openssl/bio.h>
      |          ^~~~~~~~~~~~~~~

To solve the build error, you need to install libssl-dev or openssl-devel
package, depending on your distribution.

'apt-file search', 'dnf provides', etc. is your frined to find a proper
package to install.

This commit removes all the libelf checks from the top Makefile.

If libelf is missing, objtool will fail to build in a similar pattern:

.../linux/tools/objtool/include/objtool/elf.h:10:10: fatal error: gelf.h: No such file or directory
   10 | #include <gelf.h>

You need to install libelf-dev, libelf-devel, or elfutils-libelf-devel
to proceed.

Another remarkable change is, CONFIG_STACK_VALIDATION (without
CONFIG_UNWINDER_ORC) previously continued to build with a warning,
but now it will treat missing libelf as an error.

This is just a one-time installation, so it should not hurt to break
a build and make a user install the package.

BTW, the traditional way to handle such checks is autotool, but according
to [1], I do not expect the kernel build would have similar scripting
like './configure' does.

[1]: https://lore.kernel.org/lkml/CA+55aFzr2HTZVOuzpHYDwmtRJLsVzE-yqg2DHpHi_9ePsYp5ug@mail.gmail.com/



Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Acked-by: Andrii Nakryiko <andrii@kernel.org>

This change adds support for Clang’s forward-edge Control Flow
Integrity (CFI) checking. With CONFIG_CFI_CLANG, the compiler
injects a runtime check before each indirect function call to ensure
the target is a valid function with the correct static type. This
restricts possible call targets and makes it more difficult for
an attacker to exploit bugs that allow the modification of stored
function pointers. For more details, see:

  https://clang.llvm.org/docs/ControlFlowIntegrity.html



Clang requires CONFIG_LTO_CLANG to be enabled with CFI to gain
visibility to possible call targets. Kernel modules are supported
with Clang’s cross-DSO CFI mode, which allows checking between
independently compiled components.

With CFI enabled, the compiler injects a __cfi_check() function into
the kernel and each module for validating local call targets. For
cross-module calls that cannot be validated locally, the compiler
calls the global __cfi_slowpath_diag() function, which determines
the target module and calls the correct __cfi_check() function. This
patch includes a slowpath implementation that uses __module_address()
to resolve call targets, and with CONFIG_CFI_CLANG_SHADOW enabled, a
shadow map that speeds up module look-ups by ~3x.

Clang implements indirect call checking using jump tables and
offers two methods of generating them. With canonical jump tables,
the compiler renames each address-taken function to <function>.cfi
and points the original symbol to a jump table entry, which passes
__cfi_check() validation. This isn’t compatible with stand-alone
assembly code, which the compiler doesn’t instrument, and would
result in indirect calls to assembly code to fail. Therefore, we
default to using non-canonical jump tables instead, where the compiler
generates a local jump table entry <function>.cfi_jt for each
address-taken function, and replaces all references to the function
with the address of the jump table entry.

Note that because non-canonical jump table addresses are local
to each component, they break cross-module function address
equality. Specifically, the address of a global function will be
different in each module, as it's replaced with the address of a local
jump table entry. If this address is passed to a different module,
it won’t match the address of the same function taken there. This
may break code that relies on comparing addresses passed from other
components.

CFI checking can be disabled in a function with the __nocfi attribute.
Additionally, CFI can be disabled for an entire compilation unit by
filtering out CC_FLAGS_CFI.

By default, CFI failures result in a kernel panic to stop a potential
exploit. CONFIG_CFI_PERMISSIVE enables a permissive mode, where the
kernel prints out a rate-limited warning instead, and allows execution
to continue. This option is helpful for locating type mismatches, but
should only be enabled during development.

Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/20210408182843.1754385-2-samitolvanen@google.com

With LTO, LLVM bitcode won't be compiled into native code until
modpost_link, or modfinal for modules. This change postpones calls
to objtool until after these steps, and moves objtool_args to
Makefile.lib, so the arguments can be reused in Makefile.modfinal.

As we didn't have objects to process earlier, we use --duplicate
when processing vmlinux.o. This change also disables unreachable
instruction warnings with LTO to avoid warnings about the int3
padding between functions.

Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
Reviewed-by: Kees Cook <keescook@chromium.org>

This change adds build system support for Clang's Link Time
Optimization (LTO). With -flto, instead of ELF object files, Clang
produces LLVM bitcode, which is compiled into native code at link
time, allowing the final binary to be optimized globally. For more
details, see:

  https://llvm.org/docs/LinkTimeOptimization.html

The Kconfig option CONFIG_LTO_CLANG is implemented as a choice,
which defaults to LTO being disabled. To use LTO, the architecture
must select ARCH_SUPPORTS_LTO_CLANG and support:

  - compiling with Clang,
  - compiling all assembly code with Clang's integrated assembler,
  - and linking with LLD.

While using CONFIG_LTO_CLANG_FULL results in the best runtime
performance, the compilation is not scalable in time or
memory. CONFIG_LTO_CLANG_THIN enables ThinLTO, which allows
parallel optimization and faster incremental builds. ThinLTO is
used by default if the architecture also selects
ARCH_SUPPORTS_LTO_CLANG_THIN:

  https://clang.llvm.org/docs/ThinLTO...

In some modes of operation, Kbuild allows to build modules without having
vmlinux image around. In such case, generation of module BTF is impossible.
This patch changes the behavior to emit a warning about impossibility of
generating kernel module BTF, instead of breaking the build. This is especially
important for out-of-tree external module builds.

In vmlinux-less mode:

$ make clean
$ make modules_prepare
$ touch drivers/acpi/button.c
$ make M=drivers/acpi
...
  CC [M]  drivers/acpi/button.o
  MODPOST drivers/acpi/Module.symvers
  LD [M]  drivers/acpi/button.ko
  BTF [M] drivers/acpi/button.ko
Skipping BTF generation for drivers/acpi/button.ko due to unavailability of vmlinux
...
$ readelf -S ~/linux-build/default/drivers/acpi/button.ko | grep BTF -A1
... empty ...

Now with normal build:

$ make all
...
LD [M]  drivers/acpi/button.ko
BTF [M] drivers/acpi/button.ko
...
$ readelf -S ~/linux-build/default/drivers/acpi/button.ko | grep BTF -A1
  [60] .BTF              PROGBITS         0000000000000000  00029310
       000000000000ab3f  0000000000000000           0     0     1

Fixes: 5f9ae91f

 ("kbuild: Build kernel module BTFs if BTF is enabled and pahole supports it")
Reported-by: Bruce Allan <bruce.w.allan@intel.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jessica Yu <jeyu@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Masahiro Yamada <yamada.masahiro@socionext.com>
Link: https://lore.kernel.org/bpf/20201121070829.2612884-1-andrii@kernel.org

Detect if pahole supports split BTF generation, and generate BTF for each
selected kernel module, if it does. This is exposed to Makefiles and C code as
CONFIG_DEBUG_INFO_BTF_MODULES flag.

Kernel module BTF has to be re-generated if either vmlinux's BTF changes or
module's .ko changes. To achieve that, I needed a helper similar to
if_changed, but that would allow to filter out vmlinux from the list of
updated dependencies for .ko building. I've put it next to the only place that
uses and needs it, but it might be a better idea to just add it along the
other if_changed variants into scripts/Kbuild.include.

Each kernel module's BTF deduplication is pretty fast, as it does only
incremental BTF deduplication on top of already deduplicated vmlinux BTF. To
show the added build time, I've first ran make only just built kernel (to
establish the baseline) and then forced only BTF re-generation, without
regenerating .ko files. The build was performed with -j60 parallelization on
56-core machine. The final time also includes bzImage building, so it's not
a pure BTF overhead.

$ time make -j60
...
make -j60  27.65s user 10.96s system 782% cpu 4.933 total
$ touch ~/linux-build/default/vmlinux && time make -j60
...
make -j60  123.69s user 27.85s system 1566% cpu 9.675 total

So 4.6 seconds real time, with noticeable part spent in compressed vmlinux and
bzImage building.

To show size savings, I've built my kernel configuration with about 700 kernel
modules with full BTF per each kernel module (without deduplicating against
vmlinux) and with split BTF against deduplicated vmlinux (approach in this
patch). Below are top 10 modules with biggest BTF sizes. And total size of BTF
data across all kernel modules.

It shows that split BTF "compresses" 115MB down to 5MB total. And the biggest
kernel modules get a downsize from 500-570KB down to 200-300KB.

FULL BTF
========

$ for f in $(find . -name '*.ko'); do size -A -d $f | grep BTF | awk '{print $2}'; done | awk '{ s += $1 } END { print s }'
115710691

$ for f in $(find . -name '*.ko'); do printf "%s %d\n" $f $(size -A -d $f | grep BTF | awk '{print $2}'); done | sort -nr -k2 | head -n10
./drivers/gpu/drm/i915/i915.ko 570570
./drivers/net/ethernet/mellanox/mlx5/core/mlx5_core.ko 520240
./drivers/gpu/drm/radeon/radeon.ko 503849
./drivers/infiniband/hw/mlx5/mlx5_ib.ko 491777
./fs/xfs/xfs.ko 411544
./drivers/net/ethernet/intel/i40e/i40e.ko 403904
./drivers/net/ethernet/broadcom/bnx2x/bnx2x.ko 398754
./drivers/infiniband/core/ib_core.ko 397224
./fs/cifs/cifs.ko 386249
./fs/nfsd/nfsd.ko 379738

SPLIT BTF
=========

$ for f in $(find . -name '*.ko'); do size -A -d $f | grep BTF | awk '{print $2}'; done | awk '{ s += $1 } END { print s }'
5194047

$ for f in $(find . -name '*.ko'); do printf "%s %d\n" $f $(size -A -d $f | grep BTF | awk '{print $2}'); done | sort -nr -k2 | head -n10
./drivers/gpu/drm/i915/i915.ko 293206
./drivers/gpu/drm/radeon/radeon.ko 282103
./fs/xfs/xfs.ko 222150
./drivers/net/ethernet/mellanox/mlx5/core/mlx5_core.ko 198503
./drivers/infiniband/hw/mlx5/mlx5_ib.ko 198356
./drivers/net/ethernet/broadcom/bnx2x/bnx2x.ko 113444
./fs/cifs/cifs.ko 109379
./arch/x86/kvm/kvm.ko 100225
./drivers/gpu/drm/drm.ko 94827
./drivers/infiniband/core/ib_core.ko 91188

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20201110011932.3201430-4-andrii@kernel.org

There was a request to preprocess the module linker script like we
do for the vmlinux one. (https://lkml.org/lkml/2020/8/21/512

)

The difference between vmlinux.lds and module.lds is that the latter
is needed for external module builds, thus must be cleaned up by
'make mrproper' instead of 'make clean'. Also, it must be created
by 'make modules_prepare'.

You cannot put it in arch/$(SRCARCH)/kernel/, which is cleaned up by
'make clean'. I moved arch/$(SRCARCH)/kernel/module.lds to
arch/$(SRCARCH)/include/asm/module.lds.h, which is included from
scripts/module.lds.S.

scripts/module.lds is fine because 'make clean' keeps all the
build artifacts under scripts/.

You can add arch-specific sections in <asm/module.lds.h>.

Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Tested-by: Jessica Yu <jeyu@kernel.org>
Acked-by: Will Deacon <will@kernel.org>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Palmer Dabbelt <palmerdabbelt@google.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Acked-by: Jessica Yu <jeyu@kernel.org>

Makefile.lib is included by Makefile.modfinal as well as Makefile.build.

Move modkern_cflags to Makefile.lib in order to simplify cmd_cc_o_c
in Makefile.modfinal. Move modkern_cflags as well for consistency.

Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>

I think splitting the modpost and linking modules into separate
Makefiles will be useful especially when more complex build steps
come in. The main motivation of this commit is to integrate the
proposed klp-convert feature cleanly.

I moved the logging 'Building modules, stage 2.' to Makefile.modpost
to avoid the code duplication although I do not know whether or not
this message is needed in the first place.

Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>