Currently, CONFIG_MODVERSIONS needs extra link to embed the symbol
versions into ELF objects. Then, modpost extracts the version CRCs
from them.

The following figures show how it currently works, and how I am trying
to change it.

Current implementation
======================
                                                           |----------|
                 embed CRC      -------------------------->| final    |
       $(CC)       $(LD)       /  |---------|              | link for |
       -----> *.o -------> *.o -->| modpost |              | vmlinux  |
      /              /            |         |-- *.mod.c -->| or       |
     / genksyms     /             |---------|              | module   |
  *.c ------> *.symversions                                |----------|

Genksyms outputs the calculated CRCs in the form of linker script
(*.symversions), which is used by $(LD) to update the object.

If CONFIG_LTO_CLANG=y, the build process is much more complex. Embedding
the CRCs is postponed until the LLVM bitcode is converted into ELF,
creating another intermediate *.prelink.o.

However, this complexity is unneeded. There is no reason why we must
embed version CRCs in objects so early.

There is final link stage for vmlinux (scripts/link-vmlinux.sh) and
modules (scripts/Makefile.modfinal). We can link CRCs at the very last
moment.

New implementation
==================
                                                           |----------|
                   --------------------------------------->| final    |
       $(CC)      /    |---------|                         | link for |
       -----> *.o ---->|         |                         | vmlinux  |
      /                | modpost |--- .vmlinux.export.c -->| or       |
     / genksyms        |         |--- *.mod.c ------------>| module   |
  *.c ------> *.cmd -->|---------|                         |----------|

Pass the symbol versions to modpost as separate text data, which are
available in *.cmd files.

This commit changes modpost to extract CRCs from *.cmd files instead of
from ELF objects.
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)

find_symbol() returns the first symbol found in the hash table. This
table is global, so it may return a symbol from an unexpected module.

There is a case where we want to search for a symbol with a given name
in a specified module.

Add sym_find_with_module(), which receives the module pointer as the
second argument. It is equivalent to find_module() if NULL is passed
as the module pointer.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nicolas Schier <nicolas@fjasle.eu>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com> # LLVM-14 (x86-64)

There were more EXPORT_SYMBOL types in the past. The following commits
removed unused ones.

 - f1c3d73e ("module: remove EXPORT_SYMBOL_GPL_FUTURE")
 - 36794822 ("module: remove EXPORT_UNUSED_SYMBOL*")

There are 3 remaining in enum export, but export_unknown does not make
any sense because we never expect such a situation like "we do not know
how it was exported".

If the symbol name starts with "__ksymtab_", but the section name
does not start with "___ksymtab+" or "___ksymtab_gpl+", it is not an
exported symbol.

It occurs when a variable starting with "__ksymtab_" is directly defined:

   int __ksymtab_foo;

Presumably, there is no practical issue for using such a weird variable
name (but there is no good reason for doing so, either).

Anyway, that is not an exported symbol. Setting export_unknown is not
the right thing to do. Do not call sym_add_exported() in this case.

With pointless export_unknown removed, the export type finally becomes
boolean (either EXPORT_SYMBOL or EXPORT_SYMBOL_GPL).

I renamed the field name to is_gpl_only. EXPORT_SYMBOL_GPL sets it true.
Only GPL-compatible modules can use it.

I removed the orphan comment, "How a symbol is exported", which is
unrelated to sec_mismatch_count. It is about enum export.
See commit bd5cbced ("kbuild: export-type enhancement to modpost.c")
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nicolas Schier <nicolas@fjasle.eu>
Tested-by: Nathan Chancellor <nathan@kernel.org>

This is a remnant of commit 6543becf ("mod/file2alias: make
modalias generation safe for cross compiling").
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

When CONFIG_MODVERSIONS=y, the output from genksyms is saved in
separate *.symversions files, and will be used much later when
CONFIG_LTO_CLANG=y because it is impossible to update LLVM bit code
here.

This approach is not robust because:

 - *.symversions may or may not exist. If *.symversions does not
   exist, we never know if it is missing for legitimate reason
   (i.e. no EXPORT_SYMBOL) or something bad has happened (for
   example, the user accidentally deleted it). Once it occurs,
   it is not self-healing because *.symversions is generated
   as a side effect.

 - stale (i.e. invalid) *.symversions might be picked up if an
   object is generated in a non-ordinary way, and corresponding
   *.symversions (, which was generated by old builds) just happen
   to exist.

A more robust approach is to save symbol versions in *.cmd files
because:

 - *.cmd always exists (if the object is generated by if_changed
   rule or friends). Even if the user accidentally deletes it,
   it will be regenerated in the next build.

 - *.cmd is always re-generated when the object is updated. This
   avoid stale version information being picked up.

I will remove *.symversions later.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Tested-by: Nicolas Schier <nicolas@fjasle.eu>
Reviewed-by: Nicolas Schier <nicolas@fjasle.eu>
Tested-by: Nathan Chancellor <nathan@kernel.org>

A *.mod file lists the member objects of a module, but vmlinux does
not have such a file.

Generate this list to allow modpost to know all the member objects.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nicolas Schier <nicolas@fjasle.eu>
Tested-by: Nathan Chancellor <nathan@kernel.org>

A later commit will add more code to this list_for_each_entry loop.

Before that, move the loop body into a separate helper function.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nicolas Schier <nicolas@fjasle.eu>
Tested-by: Nathan Chancellor <nathan@kernel.org>

add_intree_flag(), add_retpoline(), and add_staging_flag() are small
enough to be merged into add_header().
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nicolas Schier <nicolas@fjasle.eu>
Tested-by: Nathan Chancellor <nathan@kernel.org>

If the new-kernel-pkg utility isn't present, try using kernel-install.
This is what the %preun scriptlet in scripts/package/mkspec does too.
Signed-off-by: Reza Arbab <arbab@linux.ibm.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>

Many architectures have similar install.sh scripts.

The first half is really generic; it verifies that the kernel image
and System.map exist, then executes ~/bin/${INSTALLKERNEL} or
/sbin/${INSTALLKERNEL} if available.

The second half is kind of arch-specific; it copies the kernel image
and System.map to the destination, but the code is slightly different.

Factor out the generic part into scripts/install.sh.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nicolas Schier <n.schier@avm.de>

new_symbol() does two things; allocate a new symbol and register it
to the hash table.

Using a separate function for each is easier to understand.

Replace new_symbol() with hash_add_symbol(). Remove the second parameter
of alloc_symbol().
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

Currently, sym_add_exported() does not allocate a symbol if the same
name symbol already exists in the hash table.

This does not reflect the real use cases. You can let an external
module override the in-tree one. In this case, the external module
will export the same name symbols as the in-tree one. However,
modpost simply ignores those symbols, then Module.symvers for the
external module loses its symbols.

sym_add_exported() should allocate a new symbol.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

This is currently a warning, but I think modpost should stop building
in this case.

If the same symbol is exported multiple times and we let it keep going,
the sanity check becomes difficult.

Only the legitimate case is that an external module overrides the
corresponding in-tree module to provide a different implementation
with the same interface.

Also, there exists an upstream example that exploits this feature.

  $ make M=tools/testing/nvdimm

... builds tools/testing/nvdimm/libnvdimm.ko. This is a mocked module
that overrides the symbols from drivers/nvdimm/libnvdimm.ko.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

modpost dumps the exported symbols into Module.symvers, but currently
in random order because it iterates in the hash table.

Add a linked list of exported symbols in struct module, so we can
iterate on symbols per module.

This commit makes Module.symvers much more readable; the outer loop in
write_dump() iterates over the modules in the order of modules.order,
and the inner loop dumps symbols in each module.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

Use the doubly linked list to traverse the list in the added order.
This makes the code more consistent.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

This looks easier to understand (just because this is a pattern in
the kernel code). No functional change is intended.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

Currently, modpost manages unresolved in a singly linked list; it adds
a new node to the head, and traverses the list from new to old.

Use a doubly linked list to keep the order in the symbol table in the
ELF file.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

Add a small helper, sym_add_unresolved() to ease the further
refactoring.

Remove the 'weak' argument from alloc_symbol() because it is sensible
only for unresolved symbols.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

Currently, modpost manages modules in a singly linked list; it adds
a new node to the head, and traverses the list from new to old.

It works, but the error messages are shown in the reverse order.

If you have a Makefile like this:

  obj-m += foo.o bar.o

then, modpost shows error messages in bar.o, foo.o, in this order.

Use a doubly linked list to keep the order in modules.order; use
list_add_tail() for the node addition and list_for_each_entry() for
the list traverse.

Now that the kernel's list macros have been imported to modpost, I will
use them actively going forward.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

Import include/linux/list.h to use convenient list macros in modpost.

I dropped kernel-space code such as {WRITE,READ}_ONCE etc. and unneeded
macros.

I also imported container_of() from include/linux/container_of.h and
type definitions from include/linux/types.h.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

Currently, mod->gpl_compatible is tristate; it is set to -1 by default,
then to 1 or 0 when MODULE_LICENSE() is found.

Maybe, -1 was chosen to represent the 'unknown' license, but it is not
useful.

The current code:

    if (!mod->gpl_compatible)
            check_for_gpl_usage(exp->export, basename, exp->name);

... only cares whether gpl_compatible is zero or not.

Change it to a bool type with the initial value 'true', which has no
functional change.

The default value should be 'true' instead of 'false'.

Since commit 1d6cd392 ("modpost: turn missing MODULE_LICENSE() into
error"), unknown module license is an error.

The error message, "missing MODULE_LICENSE()" is enough to explain the
issue. It is not sensible to show another message, "GPL-incompatible
module ... uses GPL-only symbol".

Add comments to explain this.

While I was here, I renamed gpl_compatible to is_gpl_compatible for
clarification, and also slightly refactored the code.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

Use 'bool' to clarify that the valid value is true or false.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

The install target should not depend on any build artifact.

The reason is explained in commit 19514fc6 ("arm, kbuild: make
"make install" not depend on vmlinux").
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>

I think this hack is a bad idea. arch/powerpc/boot/Makefile is the
only and last user. Let's stop doing this.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc)

Fix typos in comments so that they make sense.
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>

There is no good reason to define struct namespace_list in modpost.h

struct module has pointers to struct namespace_list, but that does
not require the definition of struct namespace_list.

Move it to modpost.c.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

Do not repeat the similar code.

It is simpler to do this in check_exports() instead of add_versions().
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

It took me a while to understand the intent of "exp->module == mod".

This code goes back to 2003. [1]

The commit is not in this git repository, and might be worth a little
explanation.

You can add EXPORT_SYMBOL() without having its definition in the same
file (but you need to put a declaration).

This is typical when EXPORT_SYMBOL() is added in a C file, but the
actual implementation is in a separate assembly file.

One example is arch/arm/kernel/armksyms.c

In the old days, EXPORT_SYMBOL() was only available in C files (but
this limitation does not exist any more). If you forget to add the
definition, this error occurs.

Add a separate, clearer message for this case. It should be an error
even if KBUILD_MODPOST_WARN is given.

[1]: https://git.kernel.org/pub/scm/linux/kernel/git/history/history.git/commit/?id=2763b6bcb96e6a38a2fe31108fe5759ec5bcc80aSigned-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

The description,

  it may have already been added without a
  CRC, in this case just update the CRC

... is no longer valid.

In the old days, this function was used to update the CRC as well.

Commit 040fcc81 ("kbuild: improved modversioning support for
external modules") started to use a separate function (sym_update_crc)
for updating the CRC.

The first part, "Add an exported symbol" is correct, but it is too
obvious from the function name. Drop this comment entirely.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

If an error occurs, modpost will fail anyway. Do not write out
any content (, which might be invalid).
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

Use snprintf() to avoid the potential buffer overflow, and also
check the return value to detect the too long path.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

The macros defined in this file are for testing only and are purposely
not used. When compiled with W=2, both gcc and clang yield some
-Wunused-macros warnings. Ignore them.
Signed-off-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>

CONFIG_PAHOLE_VERSION is a part of a config since the commit below. And
when multiple people update the config, this value constantly changes.
Even if they use dummy scripts.

To fix this, add a pahole dummy script returning v99.99. (This is
translated into 9999 later in the process.)

Thereafter, this script can be invoked easily for example as:
make PAHOLE=scripts/dummy-tools/pahole oldconfig

Fixes: 613fe169 (kbuild: Add CONFIG_PAHOLE_VERSION)
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>

For out-of-tree builds, this script invokes cpio twice to copy header
files from the srctree and subsequently from the objtree. According to a
comment in the script, there might be situations in which certain files
already exist in the destination directory when header files are copied
from the objtree:

"The second CPIO can complain if files already exist which can happen
with out of tree builds having stale headers in srctree. Just silence
CPIO for now."

GNU cpio might simply print a warning like "newer or same age version
exists", but toybox cpio exits with a non-zero exit code unless the
command line option "-u" is specified.

To improve compatibility with toybox cpio, add the command line option
"-u" to unconditionally replace existing files in the destination
directory.
Signed-off-by: Daniel Mentz <danielmentz@google.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>

When developing new code/feature, CONFIG_WERROR is most
often turned off, especially for people using make W=12 to
get more warnings.

In such case, turning on -Werror temporarily would require
switching on CONFIG_WERROR in the configuration, building,
then switching off CONFIG_WERROR.

For this use case, this patch introduces a new 'e' modifier
to W= as a short hand for KCFLAGS+=-Werror" so that -Werror
got added to the kernel (built-in) and modules' CFLAGS.
Signed-off-by: Yann Droneaud <ydroneaud@opteya.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>

ld and ar support @file, which command-line options are read from.

Now that *.mod lists the member objects in the correct order, without
duplication, it is ready to be passed to ld and ar.

By using the @file syntax, people will not be worried about the pitfall
described in the NOTE.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>

The dependency

    $(obj)/%.mod: $(obj)/%$(mod-prelink-ext).o

... exists because *.mod files previously contained undefined symbols,
which are computed from *.o files when CONFIG_TRIM_UNUSED_KSYMS=y.

Now that the undefined symbols are put into separate *.usyms files,
there is no reason to make *.mod depend on *.o files.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>

It is allowed to add the same objects multiple times to obj-y / obj-m:

  obj-y += foo.o foo.o foo.o
  obj-m += bar.o bar.o bar.o

It is also allowed to add the same objects multiple times to a composite
module:

  obj-m += foo.o
  foo-y := foo1.o foo2.o foo2.o foo1.o

This flexibility is useful because the same object might be selected by
different CONFIG options, like this:

  obj-m               += foo.o
  foo-y               := foo1.o
  foo-$(CONFIG_FOO_X) += foo2.o
  foo-$(CONFIG_FOO_Y) += foo2.o

The duplicated objects are omitted at link time. It works naturally in
Makefiles because GNU Make removes duplication in $^ without changing
the order.

It is working well, almost...

A small flaw I notice is, *.mod contains duplication in such a case.

This is probably not a big deal. As far as I know, the only small
problem is scripts/mod/sumversion.c parses the same file multiple
times.

I am fixing this because I plan to reuse *.mod for other purposes,
where the duplication can be problematic.

The code change is quite simple. We already use awk to drop duplicated
lines in modules.order (see cmd_modules_order in the same file).
I copied the code, but changed RS to use spaces as record separators.

I also changed the file format to list one object per line.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>

The *.mod files have two lines; the first line lists the member objects
of the module, and the second line, if CONFIG_TRIM_UNUSED_KSYMS=y, lists
the undefined symbols.

Currently, we generate *.mod after constructing composite modules,
otherwise, we cannot compute the second line. No prerequisite is
required to print the first line.

They are orthogonal. Splitting them into separate commands will ease
further cleanups.

This commit splits the list of undefined symbols out to *.usyms files.

Previously, the list of undefined symbols ended up with a very long
line, but now it has one symbol per line.

Use sed like we did before commit 7d32358b ("kbuild: avoid split
lines in .mod files").
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nicolas Schier <nicolas@fjasle.eu>

The first command in cmd_mod is similar to the real-search macro.
Reuse it.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>