While running my branch profiler that checks for incorrect "likely" and
"unlikely"s around the kernel, there's a large number of them that are
incorrect due to being "static_branches".
As static_branches are rather special, as they are likely or unlikely for
other reasons than normal annotations are used for, there's no reason to
have them be profiled.
Expose the "unlikely_notrace" and "likely_notrace" so that the
static_branch can use them, and have them be ignored by the branch
profilers.
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20201211163754.585174b9@gandalf.local.home
Commit 815f0ddb34 ("include/linux/compiler*.h: make compiler-*.h
mutually exclusive") neglected to copy barrier_data() from
compiler-gcc.h into compiler-clang.h.
The definition in compiler-gcc.h was really to work around clang's more
aggressive optimization, so this broke barrier_data() on clang, and
consequently memzero_explicit() as well.
For example, this results in at least the memzero_explicit() call in
lib/crypto/sha256.c:sha256_transform() being optimized away by clang.
Fix this by moving the definition of barrier_data() into compiler.h.
Also move the gcc/clang definition of barrier() into compiler.h,
__memory_barrier() is icc-specific (and barrier() is already defined
using it in compiler-intel.h) and doesn't belong in compiler.h.
[rdunlap@infradead.org: fix ALPHA builds when SMP is not enabled]
Link: https://lkml.kernel.org/r/20201101231835.4589-1-rdunlap@infradead.org
Fixes: 815f0ddb34 ("include/linux/compiler*.h: make compiler-*.h mutually exclusive")
Signed-off-by: Arvind Sankar <nivedita@alum.mit.edu>
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Tested-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Cc: <stable@vger.kernel.org>
Link: https://lkml.kernel.org/r/20201014212631.207844-1-nivedita@alum.mit.edu
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The __ADDRESSABLE() macro uses the __LINE__ macro to create a temporary
symbol which has a unique name. However, if the macro is used multiple
times from within another macro, the line number will always be the
same, resulting in duplicate symbols.
Make the temporary symbols truly unique by using __UNIQUE_ID instead of
__LINE__.
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Link: https://lore.kernel.org/r/20200818135804.564436253@infradead.org
Pull header cleanup from Ingo Molnar:
"Separate out the instrumentation_begin()/end() bits from compiler.h"
* tag 'core-headers-2020-08-03' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
compiler.h: Move instrumentation_begin()/end() to new <linux/instrumentation.h> header
Linus pointed out that compiler.h - which is a key header that gets included in every
single one of the 28,000+ kernel files during a kernel build - was bloated in:
6553896666: ("vmlinux.lds.h: Create section for protection against instrumentation")
Linus noted:
> I have pulled this, but do we really want to add this to a header file
> that is _so_ core that it gets included for basically every single
> file built?
>
> I don't even see those instrumentation_begin/end() things used
> anywhere right now.
>
> It seems excessive. That 53 lines is maybe not a lot, but it pushed
> that header file to over 12kB, and while it's mostly comments, it's
> extra IO and parsing basically for _every_ single file compiled in the
> kernel.
>
> For what appears to be absolutely zero upside right now, and I really
> don't see why this should be in such a core header file!
Move these primitives into a new header: <linux/instrumentation.h>, and include that
header in the headers that make use of it.
Unfortunately one of these headers is asm-generic/bug.h, which does get included
in a lot of places, similarly to compiler.h. So the de-bloating effect isn't as
good as we'd like it to be - but at least the interfaces are defined separately.
No change to functionality intended.
Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20200604071921.GA1361070@gmail.com
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
The kernel test robot reports that moving READ_ONCE() out into its own
header breaks a W=1 build for parisc, which is relying on the definition
of compiletime_assert() being available:
| In file included from ./arch/parisc/include/generated/asm/rwonce.h:1,
| from ./include/asm-generic/barrier.h:16,
| from ./arch/parisc/include/asm/barrier.h:29,
| from ./arch/parisc/include/asm/atomic.h:11,
| from ./include/linux/atomic.h:7,
| from kernel/locking/percpu-rwsem.c:2:
| ./arch/parisc/include/asm/atomic.h: In function 'atomic_read':
| ./include/asm-generic/rwonce.h:36:2: error: implicit declaration of function 'compiletime_assert' [-Werror=implicit-function-declaration]
| 36 | compiletime_assert(__native_word(t) || sizeof(t) == sizeof(long long), \
| | ^~~~~~~~~~~~~~~~~~
| ./include/asm-generic/rwonce.h:49:2: note: in expansion of macro 'compiletime_assert_rwonce_type'
| 49 | compiletime_assert_rwonce_type(x); \
| | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| ./arch/parisc/include/asm/atomic.h:73:9: note: in expansion of macro 'READ_ONCE'
| 73 | return READ_ONCE((v)->counter);
| | ^~~~~~~~~
Move these macros into compiler_types.h, so that they are available to
READ_ONCE() and friends.
Link: http://lists.infradead.org/pipermail/linux-arm-kernel/2020-July/587094.html
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Will Deacon <will@kernel.org>
In preparation for allowing architectures to define their own
implementation of the READ_ONCE() macro, move the generic
{READ,WRITE}_ONCE() definitions out of the unwieldy 'linux/compiler.h'
file and into a new 'rwonce.h' header under 'asm-generic'.
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Will Deacon <will@kernel.org>
A KCSAN build revealed we have explicit annoations through atomic_*()
usage, switch to arch_atomic_*() for the respective functions.
vmlinux.o: warning: objtool: rcu_nmi_exit()+0x4d: call to __kcsan_check_access() leaves .noinstr.text section
vmlinux.o: warning: objtool: rcu_dynticks_eqs_enter()+0x25: call to __kcsan_check_access() leaves .noinstr.text section
vmlinux.o: warning: objtool: rcu_nmi_enter()+0x4f: call to __kcsan_check_access() leaves .noinstr.text section
vmlinux.o: warning: objtool: rcu_dynticks_eqs_exit()+0x2a: call to __kcsan_check_access() leaves .noinstr.text section
vmlinux.o: warning: objtool: __rcu_is_watching()+0x25: call to __kcsan_check_access() leaves .noinstr.text section
Additionally, without the NOP in instrumentation_begin(), objtool would
not detect the lack of the 'else instrumentation_begin();' branch in
rcu_nmi_enter().
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
The volatile accesses no longer need to be wrapped in data_race()
because compilers that emit instrumentation distinguishing volatile
accesses are required for KCSAN.
Consequently, the explicit kcsan_check_atomic*() are no longer required
either since the compiler emits instrumentation distinguishing the
volatile accesses.
Finally, simplify __READ_ONCE_SCALAR() and remove __WRITE_ONCE_SCALAR().
[ bp: Convert commit message to passive voice. ]
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Will Deacon <will@kernel.org>
Link: https://lkml.kernel.org/r/20200521142047.169334-9-elver@google.com
Some compilers incorrectly inline small __no_kcsan functions, which then
results in instrumenting the accesses. For this reason, the 'noinline'
attribute was added to __no_kcsan_or_inline. All known versions of GCC
are affected by this. Supported versions of Clang are unaffected, and
never inline a no_sanitize function.
However, the attribute 'noinline' in __no_kcsan_or_inline causes
unexpected code generation in functions that are __no_kcsan and call a
__no_kcsan_or_inline function.
In certain situations it is expected that the __no_kcsan_or_inline
function is actually inlined by the __no_kcsan function, and *no* calls
are emitted. By removing the 'noinline' attribute, give the compiler
the ability to inline and generate the expected code in __no_kcsan
functions.
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Will Deacon <will@kernel.org>
Link: https://lkml.kernel.org/r/CANpmjNNOpJk0tprXKB_deiNAv_UmmORf1-2uajLhnLWQQ1hvoA@mail.gmail.com
Link: https://lkml.kernel.org/r/20200521142047.169334-6-elver@google.com
Merge the state of the locking kcsan branch before the read/write_once()
and the atomics modifications got merged.
Squash the fallout of the rebase on top of the read/write once and atomic
fallback work into the merge. The history of the original branch is
preserved in tag locking-kcsan-2020-06-02.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Pull READ/WRITE_ONCE rework from Will Deacon:
"This the READ_ONCE rework I've been working on for a while, which
bumps the minimum GCC version and improves code-gen on arm64 when
stack protector is enabled"
[ Side note: I'm _really_ tempted to raise the minimum gcc version to
4.9, so that we can just say that we require _Generic() support.
That would allow us to more cleanly handle a lot of the cases where we
depend on very complex macros with 'sizeof' or __builtin_choose_expr()
with __builtin_types_compatible_p() etc.
This branch has a workaround for sparse not handling _Generic(),
either, but that was already fixed in the sparse development branch,
so it's really just gcc-4.9 that we'd require. - Linus ]
* 'rwonce/rework' of git://git.kernel.org/pub/scm/linux/kernel/git/will/linux:
compiler_types.h: Use unoptimized __unqual_scalar_typeof for sparse
compiler_types.h: Optimize __unqual_scalar_typeof compilation time
compiler.h: Enforce that READ_ONCE_NOCHECK() access size is sizeof(long)
compiler-types.h: Include naked type in __pick_integer_type() match
READ_ONCE: Fix comment describing 2x32-bit atomicity
gcov: Remove old GCC 3.4 support
arm64: barrier: Use '__unqual_scalar_typeof' for acquire/release macros
locking/barriers: Use '__unqual_scalar_typeof' for load-acquire macros
READ_ONCE: Drop pointer qualifiers when reading from scalar types
READ_ONCE: Enforce atomicity for {READ,WRITE}_ONCE() memory accesses
READ_ONCE: Simplify implementations of {READ,WRITE}_ONCE()
arm64: csum: Disable KASAN for do_csum()
fault_inject: Don't rely on "return value" from WRITE_ONCE()
net: tls: Avoid assigning 'const' pointer to non-const pointer
netfilter: Avoid assigning 'const' pointer to non-const pointer
compiler/gcc: Raise minimum GCC version for kernel builds to 4.8
READ_ONCE_NOCHECK() unconditionally performs a sizeof(long)-sized access,
so enforce that the size of the pointed-to object that we are loading
from is the same size as 'long'.
Reported-by: Marco Elver <elver@google.com>
Signed-off-by: Will Deacon <will@kernel.org>
READ_ONCE() permits 64-bit accesses on 32-bit architectures, since this
crops up in a few places and is generally harmless because either the
upper bits are always zero (e.g. for a virtual address or 32-bit time_t)
or the architecture provides 64-bit atomicity anyway.
Update the corresponding comment above compiletime_assert_rwonce_type(),
which incorrectly states that 32-bit x86 provides 64-bit atomicity, and
instead reference 32-bit Armv7 with LPAE.
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Will Deacon <will@kernel.org>
Pull kprobes updates from Ingo Molnar:
"Various kprobes updates, mostly centered around cleaning up the
no-instrumentation logic.
Instead of the current per debug facility blacklist, use the more
generic .noinstr.text approach, combined with a 'noinstr' marker for
functions.
Also add instrumentation_begin()/end() to better manage the exact
place in entry code where instrumentation may be used.
And add a kprobes blacklist for modules"
* tag 'core-kprobes-2020-06-01' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
kprobes: Prevent probes in .noinstr.text section
vmlinux.lds.h: Create section for protection against instrumentation
samples/kprobes: Add __kprobes and NOKPROBE_SYMBOL() for handlers.
kprobes: Support NOKPROBE_SYMBOL() in modules
kprobes: Support __kprobes blacklist in modules
kprobes: Lock kprobe_mutex while showing kprobe_blacklist
Some code pathes, especially the low level entry code, must be protected
against instrumentation for various reasons:
- Low level entry code can be a fragile beast, especially on x86.
- With NO_HZ_FULL RCU state needs to be established before using it.
Having a dedicated section for such code allows to validate with tooling
that no unsafe functions are invoked.
Add the .noinstr.text section and the noinstr attribute to mark
functions. noinstr implies notrace. Kprobes will gain a section check
later.
Provide also a set of markers: instrumentation_begin()/end()
These are used to mark code inside a noinstr function which calls
into regular instrumentable text section as safe.
The instrumentation markers are only active when CONFIG_DEBUG_ENTRY is
enabled as the end marker emits a NOP to prevent the compiler from merging
the annotation points. This means the objtool verification requires a
kernel compiled with this option.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Alexandre Chartre <alexandre.chartre@oracle.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20200505134100.075416272@linutronix.de
... or the odyssey of trying to disable the stack protector for the
function which generates the stack canary value.
The whole story started with Sergei reporting a boot crash with a kernel
built with gcc-10:
Kernel panic — not syncing: stack-protector: Kernel stack is corrupted in: start_secondary
CPU: 1 PID: 0 Comm: swapper/1 Not tainted 5.6.0-rc5—00235—gfffb08b37df9 #139
Hardware name: Gigabyte Technology Co., Ltd. To be filled by O.E.M./H77M—D3H, BIOS F12 11/14/2013
Call Trace:
dump_stack
panic
? start_secondary
__stack_chk_fail
start_secondary
secondary_startup_64
-—-[ end Kernel panic — not syncing: stack—protector: Kernel stack is corrupted in: start_secondary
This happens because gcc-10 tail-call optimizes the last function call
in start_secondary() - cpu_startup_entry() - and thus emits a stack
canary check which fails because the canary value changes after the
boot_init_stack_canary() call.
To fix that, the initial attempt was to mark the one function which
generates the stack canary with:
__attribute__((optimize("-fno-stack-protector"))) ... start_secondary(void *unused)
however, using the optimize attribute doesn't work cumulatively
as the attribute does not add to but rather replaces previously
supplied optimization options - roughly all -fxxx options.
The key one among them being -fno-omit-frame-pointer and thus leading to
not present frame pointer - frame pointer which the kernel needs.
The next attempt to prevent compilers from tail-call optimizing
the last function call cpu_startup_entry(), shy of carving out
start_secondary() into a separate compilation unit and building it with
-fno-stack-protector, was to add an empty asm("").
This current solution was short and sweet, and reportedly, is supported
by both compilers but we didn't get very far this time: future (LTO?)
optimization passes could potentially eliminate this, which leads us
to the third attempt: having an actual memory barrier there which the
compiler cannot ignore or move around etc.
That should hold for a long time, but hey we said that about the other
two solutions too so...
Reported-by: Sergei Trofimovich <slyfox@gentoo.org>
Signed-off-by: Borislav Petkov <bp@suse.de>
Tested-by: Kalle Valo <kvalo@codeaurora.org>
Cc: <stable@vger.kernel.org>
Link: https://lkml.kernel.org/r/20200314164451.346497-1-slyfox@gentoo.org
Passing a volatile-qualified pointer to READ_ONCE() is an absolute
trainwreck for code generation: the use of 'typeof()' to define a
temporary variable inside the macro means that the final evaluation in
macro scope ends up forcing a read back from the stack. When stack
protector is enabled (the default for arm64, at least), this causes
the compiler to vomit up all sorts of junk.
Unfortunately, dropping pointer qualifiers inside the macro poses quite
a challenge, especially since the pointed-to type is permitted to be an
aggregate, and this is relied upon by mm/ code accessing things like
'pmd_t'. Based on numerous hacks and discussions on the mailing list,
this is the best I've managed to come up with.
Introduce '__unqual_scalar_typeof()' which takes an expression and, if
the expression is an optionally qualified 8, 16, 32 or 64-bit scalar
type, evaluates to the unqualified type. Other input types, including
aggregates, remain unchanged. Hopefully READ_ONCE() on volatile aggregate
pointers isn't something we do on a fast-path.
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Reported-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Will Deacon <will@kernel.org>
{READ,WRITE}_ONCE() cannot guarantee atomicity for arbitrary data sizes.
This can be surprising to callers that might incorrectly be expecting
atomicity for accesses to aggregate structures, although there are other
callers where tearing is actually permissable (e.g. if they are using
something akin to sequence locking to protect the access).
Linus sayeth:
| We could also look at being stricter for the normal READ/WRITE_ONCE(),
| and require that they are
|
| (a) regular integer types
|
| (b) fit in an atomic word
|
| We actually did (b) for a while, until we noticed that we do it on
| loff_t's etc and relaxed the rules. But maybe we could have a
| "non-atomic" version of READ/WRITE_ONCE() that is used for the
| questionable cases?
The slight snag is that we also have to support 64-bit accesses on 32-bit
architectures, as these appear to be widespread and tend to work out ok
if either the architecture supports atomic 64-bit accesses (x86, armv7)
or if the variable being accesses represents a virtual address and
therefore only requires 32-bit atomicity in practice.
Take a step in that direction by introducing a variant of
'compiletime_assert_atomic_type()' and use it to check the pointer
argument to {READ,WRITE}_ONCE(). Expose __{READ,WRITE}_ONCE() variants
which are allowed to tear and convert the one broken caller over to the
new macros.
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Will Deacon <will@kernel.org>
The implementations of {READ,WRITE}_ONCE() suffer from a significant
amount of indirection and complexity due to a historic GCC bug:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58145
which was originally worked around by 230fa253df ("kernel: Provide
READ_ONCE and ASSIGN_ONCE").
Since GCC 4.8 is fairly vintage at this point and we emit a warning if
we detect it during the build, return {READ,WRITE}_ONCE() to their former
glory with an implementation that is easier to understand and, crucially,
more amenable to optimisation. A side effect of this simplification is
that WRITE_ONCE() no longer returns a value, but nobody seems to be
relying on that and the new behaviour is aligned with smp_store_release().
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Will Deacon <will@kernel.org>
