Commit c3a6cf19e6 ("export: avoid code duplication in
include/linux/export.h") refactors export.h quite nicely, but introduces
a slight increase in memory usage due to using the empty string ""
instead of NULL to indicate that an exported symbol has no namespace. As
mentioned in that commit, this meant an increase of 1 byte per exported
symbol without a namespace. For example, if a kernel configuration has
about 10k exported symbols, this would mean that the size of
__ksymtab_strings would increase by roughly 10kB.
We can alleviate this situation by utilizing the SHF_MERGE and
SHF_STRING section flags. SHF_MERGE|SHF_STRING indicate to the linker
that the data in the section are null-terminated strings that can be
merged to eliminate duplication. More specifically, from the binutils
documentation - "for sections with both M and S, a string which is a
suffix of a larger string is considered a duplicate. Thus "def" will be
merged with "abcdef"; A reference to the first "def" will be changed to
a reference to "abcdef"+3". Thus, all the empty strings would be merged
as well as any strings that can be merged according to the cited method
above. For example, "memset" and "__memset" would be merged to just
"__memset" in __ksymtab_strings.
As of v5.4-rc5, the following statistics were gathered with x86
defconfig with approximately 10.7k exported symbols.
Size of __ksymtab_strings in vmlinux:
-------------------------------------
v5.4-rc5: 213834 bytes
v5.4-rc5 with commit c3a6cf19e695: 224455 bytes
v5.4-rc5 with this patch: 205759 bytes
So, we already see memory savings of ~8kB compared to vanilla -rc5 and
savings of nearly 18.7kB compared to -rc5 with commit c3a6cf19e6 on top.
Unfortunately, as of this writing, strings will not get deduplicated for
kernel modules, as ld does not do the deduplication for
SHF_MERGE|SHF_STRINGS sections for relocatable files (ld -r), which
kernel modules are. A patch for ld is currently being worked on to
hopefully allow for string deduplication in relocatable files in the
future.
Suggested-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Reviewed-by: Masahiro Yamada <masahiroy@kernel.org>
Reviewed-by: Matthias Maennich <maennich@google.com>
Signed-off-by: Jessica Yu <jeyu@kernel.org>
Here is a single SPDX fixup for 5.5-rc1
It resolves an issue where we had missed a few .h files with the
auto-tagging scripts because they had "GPL" text in strings within the
file themselves. This single patch fixes up the issue and provides the
proper SPDX tags at the top of them as needed.
This patch has been in linux-next for many many weeks now with no
reported issues.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
-----BEGIN PGP SIGNATURE-----
iG0EABECAC0WIQT0tgzFv3jCIUoxPcsxR9QN2y37KQUCXebGqw8cZ3JlZ0Brcm9h
aC5jb20ACgkQMUfUDdst+yn31wCePtjLIQCW2PCrub9zeZCqUJ62V0sAoK53hYv+
f3LHw9fd2dzhNCcly7nW
=MuKu
-----END PGP SIGNATURE-----
Merge tag 'spdx-5.5-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/spdx
Pull SPDX fix from Greg KH:
"Here is a single SPDX fixup for 5.5-rc1
It resolves an issue where we had missed a few .h files with the
auto-tagging scripts because they had "GPL" text in strings within the
file themselves. This single patch fixes up the issue and provides the
proper SPDX tags at the top of them as needed.
This patch has been in linux-next for many many weeks now with no
reported issues"
* tag 'spdx-5.5-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/spdx:
export,module: add SPDX GPL-2.0 license identifier to headers with no license
Commit b24413180f ("License cleanup: add SPDX GPL-2.0 license
identifier to files with no license") took care of a lot of files
without any license information.
These headers were not processed by the tool perhaps because they
contain "GPL" in the code.
I do not see any license boilerplate in them, so they fall back to
GPL version 2 only, which is the project default.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Link: https://lore.kernel.org/r/20191018045053.8424-1-yamada.masahiro@socionext.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
EXPORT_SYMBOL from assembly code produces an unused symbol __kcrctab_*.
kcrctab is used as a section name (prefixed with three underscores),
but never used as a symbol.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
For EXPORT_SYMBOL from C files, <linux/export.h> defines __ksymtab_*
as local symbols.
For EXPORT_SYMBOL from assembly, in contrast, <asm-generic/export.h>
produces globally-visible __ksymtab_* symbols due to this .globl
directive.
I do not know why this must be global. It still works without this.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Commit 7290d58095 ("module: use relative references for __ksymtab
entries") converted the '__put' #define into an assembly macro in
asm-generic/export.h but forgot to remove the corresponding '#undef'.
Remove the leftover '#undef'.
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Jessica Yu <jeyu@kernel.org>
Signed-off-by: Will Deacon <will@kernel.org>
Signed-off-by: Jessica Yu <jeyu@kernel.org>
Commit 8651ec01da ("module: add support for symbol namespaces.")
broke linking for arm64 defconfig:
| lib/crypto/arc4.o: In function `__ksymtab_arc4_setkey':
| arc4.c:(___ksymtab+arc4_setkey+0x8): undefined reference to `no symbol'
| lib/crypto/arc4.o: In function `__ksymtab_arc4_crypt':
| arc4.c:(___ksymtab+arc4_crypt+0x8): undefined reference to `no symbol'
This is because the dummy initialisation of the 'namespace_offset' field
in 'struct kernel_symbol' when using EXPORT_SYMBOL on architectures with
support for PREL32 locations uses an offset from an absolute address (0)
in an effort to trick 'offset_to_pointer' into behaving as a NOP,
allowing non-namespaced symbols to be treated in the same way as those
belonging to a namespace.
Unfortunately, place-relative relocations require a symbol reference
rather than an absolute value and, although x86 appears to get away with
this due to placing the kernel text at the top of the address space, it
almost certainly results in a runtime failure if the kernel is relocated
dynamically as a result of KASLR.
Rework 'namespace_offset' so that a value of 0, which cannot occur for a
valid namespaced symbol, indicates that the corresponding symbol does
not belong to a namespace.
Cc: Matthias Maennich <maennich@google.com>
Cc: Jessica Yu <jeyu@kernel.org>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Fixes: 8651ec01da ("module: add support for symbol namespaces.")
Reported-by: kbuild test robot <lkp@intel.com>
Tested-by: Matthias Maennich <maennich@google.com>
Tested-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Matthias Maennich <maennich@google.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will@kernel.org>
Signed-off-by: Jessica Yu <jeyu@kernel.org>
The EXPORT_SYMBOL_NS() and EXPORT_SYMBOL_NS_GPL() macros can be used to
export a symbol to a specific namespace. There are no _GPL_FUTURE and
_UNUSED variants because these are currently unused, and I'm not sure
they are necessary.
I didn't add EXPORT_SYMBOL_NS() for ASM exports; this patch sets the
namespace of ASM exports to NULL by default. In case of relative
references, it will be relocatable to NULL. If there's a need, this
should be pretty easy to add.
A module that wants to use a symbol exported to a namespace must add a
MODULE_IMPORT_NS() statement to their module code; otherwise, modpost
will complain when building the module, and the kernel module loader
will emit an error and fail when loading the module.
MODULE_IMPORT_NS() adds a modinfo tag 'import_ns' to the module. That
tag can be observed by the modinfo command, modpost and kernel/module.c
at the time of loading the module.
The ELF symbols are renamed to include the namespace with an asm label;
for example, symbol 'usb_stor_suspend' in namespace USB_STORAGE becomes
'usb_stor_suspend.USB_STORAGE'. This allows modpost to do namespace
checking, without having to go through all the effort of parsing ELF and
relocation records just to get to the struct kernel_symbols.
On x86_64 I saw no difference in binary size (compression), but at
runtime this will require a word of memory per export to hold the
namespace. An alternative could be to store namespaced symbols in their
own section and use a separate 'struct namespaced_kernel_symbol' for
that section, at the cost of making the module loader more complex.
Co-developed-by: Martijn Coenen <maco@android.com>
Signed-off-by: Martijn Coenen <maco@android.com>
Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Matthias Maennich <maennich@google.com>
Signed-off-by: Jessica Yu <jeyu@kernel.org>
This change allows growing struct kernel_symbol without wasting bytes to
alignment. It also concretized the alignment of ksymtab entries if
relative references are used for ksymtab entries.
struct kernel_symbol was already implicitly being aligned to the word
size, except on x86_64 and m68k, where it is aligned to 16 and 2 bytes,
respectively.
As far as I can tell there is no requirement for aligning struct
kernel_symbol to 16 bytes on x86_64, but gcc aligns structs to their
size, and the linker aligns the custom __ksymtab sections to the largest
data type contained within, so setting KSYM_ALIGN to 16 was necessary to
stay consistent with the code generated for non-ASM EXPORT_SYMBOL(). Now
that non-ASM EXPORT_SYMBOL() explicitly aligns to word size (8),
KSYM_ALIGN is no longer necessary.
In case of relative references, the alignment has been changed
accordingly to not waste space when adding new struct members.
As for m68k, struct kernel_symbol is aligned to 2 bytes even though the
structure itself is 8 bytes; using a 4-byte alignment shouldn't hurt.
I manually verified the output of the __ksymtab sections didn't change
on x86, x86_64, arm, arm64 and m68k. As expected, the section contents
didn't change, and the ELF section alignment only changed on x86_64 and
m68k. Feedback from other archs more than welcome.
Co-developed-by: Martijn Coenen <maco@android.com>
Signed-off-by: Martijn Coenen <maco@android.com>
Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Matthias Maennich <maennich@google.com>
Signed-off-by: Jessica Yu <jeyu@kernel.org>
My main motivation of this commit is to clean up scripts/Kbuild.include
and scripts/Makefile.build.
Currently, CONFIG_TRIM_UNUSED_KSYMS works with a tricky gimmick;
possibly exported symbols are detected by letting $(CPP) replace
EXPORT_SYMBOL* with a special string '=== __KSYM_*===', which is
post-processed by sed, and passed to fixdep. The extra preprocessing
is costly, and hacking cmd_and_fixdep is ugly.
I came up with a new way to find exported symbols; insert a dummy
symbol __ksym_marker_* to each potentially exported symbol. Those
dummy symbols are picked up by $(NM), post-processed by sed, then
appended to .*.cmd files. I collected the post-process part to a
new shell script scripts/gen_ksymdeps.sh for readability. The dummy
symbols are put into the .discard.* section so that the linker
script rips them off the final vmlinux or modules.
A nice side-effect is building with CONFIG_TRIM_UNUSED_KSYMS will
be much faster.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Nicolas Pitre <nico@linaro.org>
An ordinary arm64 defconfig build has ~64 KB worth of __ksymtab entries,
each consisting of two 64-bit fields containing absolute references, to
the symbol itself and to a char array containing its name, respectively.
When we build the same configuration with KASLR enabled, we end up with an
additional ~192 KB of relocations in the .init section, i.e., one 24 byte
entry for each absolute reference, which all need to be processed at boot
time.
Given how the struct kernel_symbol that describes each entry is completely
local to module.c (except for the references emitted by EXPORT_SYMBOL()
itself), we can easily modify it to contain two 32-bit relative references
instead. This reduces the size of the __ksymtab section by 50% for all
64-bit architectures, and gets rid of the runtime relocations entirely for
architectures implementing KASLR, either via standard PIE linking (arm64)
or using custom host tools (x86).
Note that the binary search involving __ksymtab contents relies on each
section being sorted by symbol name. This is implemented based on the
input section names, not the names in the ksymtab entries, so this patch
does not interfere with that.
Given that the use of place-relative relocations requires support both in
the toolchain and in the module loader, we cannot enable this feature for
all architectures. So make it dependent on whether
CONFIG_HAVE_ARCH_PREL32_RELOCATIONS is defined.
Link: http://lkml.kernel.org/r/20180704083651.24360-4-ard.biesheuvel@linaro.org
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Acked-by: Jessica Yu <jeyu@kernel.org>
Acked-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Will Deacon <will.deacon@arm.com>
Acked-by: Ingo Molnar <mingo@kernel.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Bjorn Helgaas <bhelgaas@google.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: James Morris <james.morris@microsoft.com>
Cc: James Morris <jmorris@namei.org>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Nicolas Pitre <nico@linaro.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Petr Mladek <pmladek@suse.com>
Cc: Russell King <linux@armlinux.org.uk>
Cc: "Serge E. Hallyn" <serge@hallyn.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Garnier <thgarnie@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
CONFIG_HAVE_UNDERSCORE_SYMBOL_PREFIX was selected by BLACKFIN, METAG.
They were removed by commit 4ba66a9760 ("arch: remove blackfin port"),
commit bb6fb6dfcc ("metag: Remove arch/metag/"), respectively.
No more architecture enables CONFIG_HAVE_UNDERSCORE_SYMBOL_PREFIX.
Clean up the export.h headers. I am keeping VMLINUX_SYMBOL() and
VMLINUX_SYMBOL_STR() because they are widely used.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Reviewed-by: Sam Ravnborg <sam@ravnborg.org>
The modversion symbol CRCs are emitted as ELF symbols, which allows us
to easily populate the kcrctab sections by relying on the linker to
associate each kcrctab slot with the correct value.
This has a couple of downsides:
- Given that the CRCs are treated as memory addresses, we waste 4 bytes
for each CRC on 64 bit architectures,
- On architectures that support runtime relocation, a R_<arch>_RELATIVE
relocation entry is emitted for each CRC value, which identifies it
as a quantity that requires fixing up based on the actual runtime
load offset of the kernel. This results in corrupted CRCs unless we
explicitly undo the fixup (and this is currently being handled in the
core module code)
- Such runtime relocation entries take up 24 bytes of __init space
each, resulting in a x8 overhead in [uncompressed] kernel size for
CRCs.
Switching to explicit 32 bit values on 64 bit architectures fixes most
of these issues, given that 32 bit values are not treated as quantities
that require fixing up based on the actual runtime load offset. Note
that on some ELF64 architectures [such as PPC64], these 32-bit values
are still emitted as [absolute] runtime relocatable quantities, even if
the value resolves to a build time constant. Since relative relocations
are always resolved at build time, this patch enables MODULE_REL_CRCS on
powerpc when CONFIG_RELOCATABLE=y, which turns the absolute CRC
references into relative references into .rodata where the actual CRC
value is stored.
So redefine all CRC fields and variables as u32, and redefine the
__CRC_SYMBOL() macro for 64 bit builds to emit the CRC reference using
inline assembler (which is necessary since 64-bit C code cannot use
32-bit types to hold memory addresses, even if they are ultimately
resolved using values that do not exceed 0xffffffff). To avoid
potential problems with legacy 32-bit architectures using legacy
toolchains, the equivalent C definition of the kcrctab entry is retained
for 32-bit architectures.
Note that this mostly reverts commit d4703aefdb ("module: handle ppc64
relocating kcrctabs when CONFIG_RELOCATABLE=y")
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This reverts commit 8ab2ae655b.
I loved that commit because of how it explained what the problem with
newer versions of binutils were, but the actual patch itself turns out
to not work very well.
It has two problems:
- a zero CRC value isn't actually right. It happens to work for the
case where both sides of the equation fail at giving the symbol a
crc, but there are cases where the users of the exported symbol get
the right crc (due to seeing the C declarations), but the actual
exporting itself does not (due to the whole weak asm symbol issue).
So then the module load fails after all - we did have a crc for the
symbol, but we couldn't match it with the loaded module.
- it seems that the alpha assembler has special semantics for the
'.set' directive, and on alpha it doesn't actually set the value of
the specified symbol at all, it is instead used to set various
assembly modes (eg ".set noat" and ".set noreorder").
So using ".set" to set the symbol value would just cause build
failures on alpha.
I'm sure we'll find some other workaround for these issues (hopefully
that involves getting rid of modversions entirely some day, but people
are also talking about just using smarter tools). But for now we'll
just fall back on commit faaae2a581 ("Re-enable CONFIG_MODVERSIONS in
a slightly weaker form") that just let's a missing crc through.
Reported-by: Jan Stancek <jstancek@redhat.com>
Reported-by: Philip Müller <philm@manjaro.org>
Reported-by: Guenter Roeck <linux@roeck-us.net>
Cc: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
With binutils-2.26 and before, a weak missing symbol was kept during the
final link, and a missing CRC for an export would lead to that CRC being
treated as zero implicitly. With binutils-2.27, the crc symbol gets
dropped, and any module trying to use it will fail to load.
This sets the weak CRC symbol to zero explicitly, making it defined in
vmlinux, which in turn lets us load the modules referring to that CRC.
The comment above the __CRC_SYMBOL macro suggests that this was always
the intention, although it also seems that all symbols defined in C have
a correct CRC these days, and only the exports that are now done in
assembly need this.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Tested-by: Adam Borowski <kilobyte@angband.pl>
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The use of config_enabled() is ambiguous. For config options,
IS_ENABLED(), IS_REACHABLE(), etc. will make intention clearer.
Sometimes config_enabled() has been used for non-config options because
it is useful to check whether the given symbol is defined or not.
I have been tackling on deprecating config_enabled(), and now is the
time to finish this work.
Some new users have appeared for v4.9-rc1, but it is trivial to replace
them:
- arch/x86/mm/kaslr.c
replace config_enabled() with IS_ENABLED() because
CONFIG_X86_ESPFIX64 and CONFIG_EFI are boolean.
- include/asm-generic/export.h
replace config_enabled() with __is_defined().
Then, config_enabled() can be removed now.
Going forward, please use IS_ENABLED(), IS_REACHABLE(), etc. for config
options, and __is_defined() for non-config symbols.
Link: http://lkml.kernel.org/r/1476616078-32252-1-git-send-email-yamada.masahiro@socionext.com
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
Acked-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Cc: Peter Oberparleiter <oberpar@linux.vnet.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Kees Cook <keescook@chromium.org>
Cc: Michal Marek <mmarek@suse.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Thomas Garnier <thgarnie@google.com>
Cc: Paul Bolle <pebolle@tiscali.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add asm-usable variants of EXPORT_SYMBOL/EXPORT_SYMBOL_GPL. This
commit just adds the default implementation; most of the architectures
can simply add export.h to asm/Kbuild and start using <asm/export.h>
from assembler. The rest needs to have their <asm/export.h> define
everal macros and then explicitly include <asm-generic/export.h>
One area where the things might diverge from default is the alignment;
normally it's 8 bytes on 64bit targets and 4 on 32bit ones, both for
unsigned long and for struct kernel_symbol. Unfortunately, amd64 and
m68k are unusual - m68k aligns to 2 bytes (for both) and amd64 aligns
struct kernel_symbol to 16 bytes. For those we'll need asm/export.h to
override the constants used by generic version - KSYM_ALIGN and KCRC_ALIGN
for kernel_symbol and unsigned long resp. And no, __alignof__ would not
do the trick - on amd64 __alignof__ of struct kernel_symbol is 8, not 16.
More serious source of unpleasantness is treatment of function
descriptors on architectures that have those. Things like ppc64,
parisc, ia64, etc. need more than the address of the first insn to
call an arbitrary function. As the result, their representation of
pointers to functions is not the typical "address of the entry point" -
it's an address of a small static structure containing all the required
information (including the entry point, of course). Sadly, the asm-side
conventions differ in what the function name refers to - entry point or
the function descriptor. On ppc64 we do the latter;
bar: .quad foo
is what void (*bar)(void) = foo; turns into and the rare places where
we need to explicitly work with the label of entry point are dealt with
as DOTSYM(foo). For our purposes it's ideal - generic macros are usable.
However, parisc would have foo and P%foo used for label of entry point
and address of the function descriptor and
bar: .long P%foo
woudl be used instead. ia64 goes similar to parisc in that respect,
except that there it's @fptr(foo) rather than P%foo. Such architectures
need to define KSYM_FUNC that would turn a function name into whatever
is needed to refer to function descriptor.
What's more, on such architectures we need to know whether we are exporting
a function or an object - in assembler we have to tell that explicitly, to
decide whether we want EXPORT_SYMBOL(foo) produce e.g.
__ksymtab_foo: .quad foo
or
__ksymtab_foo: .quad @fptr(foo)
For that reason we introduce EXPORT_DATA_SYMBOL{,_GPL}(), to be used for
exports of data objects. On normal architectures it's the same thing
as EXPORT_SYMBOL{,_GPL}(), but on parisc-like ones they differ and the
right one needs to be used. Most of the exports are functions, so we
keep EXPORT_SYMBOL for those...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Jessica Yu