kernel/include/asm-generic/atomic-instrumented.h, branch linux-5.2.y

locking/atomics: Check atomic headers with sha1sum

2019-02-13T07:07:31Z

We currently check the atomic headers at build-time to ensure they haven't been modified directly, and these checks require regenerating the headers in full. As this takes a few seconds, even when parallelized, this is too slow to run for every kernel build. Instead, we can generate a hash of each header as we generate them, which we can cheaply check at build time (~0.16s for all headers). This patch does so, updating headers with their hashes using the new gen-atomics.sh script. As some users apparently build the kernel wihout coreutils, lacking sha1sum, the checks are skipped in this case. Presumably, most developers have a working coreutils installation. Signed-off-by: Mark Rutland Acked-by: Will Deacon Cc: Andrew Morton Cc: Boqun Feng Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Cc: anders.roxell@linaro.org Cc: linux-kernel@vger.kernel.rg Cc: naresh.kamboju@linaro.org Signed-off-by: Ingo Molnar

locking/atomics: Switch to generated instrumentation

2018-11-01T10:01:03Z

As a step towards ensuring the atomic* APIs are consistent, let's switch to wrappers generated by gen-atomic-instrumented.h, using the same table used to generate the fallbacks and atomic-long wrappers. These are checked in rather than generated with Kbuild, since: * This allows inspection of the atomics with git grep and ctags on a pristine tree, which Linus strongly prefers being able to do. * The fallbacks are not affected by machine details or configuration options, so it is not necessary to regenerate them to take these into account. * These are included by files required *very* early in the build process (e.g. for generating bounds.h), and we'd rather not complicate the top-level Kbuild file with dependencies. Generating the atomic headers means that the instrumented wrappers will remain in sync with the rest of the atomic APIs, and we gain all the ordering variants of each atomic without having to manually expanded them all. The KASAN checks are automatically generated based on the function parameters defined in atomics.tbl. Note that try_cmpxchg() now correctly treats 'old' as a parameter that may be written to, and not only read as the hand-written instrumentation assumed. Other than the change to try_cmpxchg(), existing code should not be affected by this patch. The patch introduces instrumentation for all optional atomics (and ordering variants), along with the ifdeffery this requires, enabling other architectures to make use of the instrumented atomics. Signed-off-by: Mark Rutland Signed-off-by: Peter Zijlstra (Intel) Acked-by: Will Deacon Cc: linux-arm-kernel@lists.infradead.org Cc: catalin.marinas@arm.com Cc: linuxdrivers@attotech.com Cc: Dmitry Vyukov Cc: Boqun Feng Cc: Arnd Bergmann Cc: Andrey Ryabinin Cc: Alexander Potapenko Link: http://lkml.kernel.org/r/20180904104830.2975-5-mark.rutland@arm.com Signed-off-by: Ingo Molnar

locking/atomics: Instrument cmpxchg_double*()

2018-07-25T09:53:59Z

We currently don't instrument cmpxchg_double() and cmpxchg_double_local() due to compilation issues reported in the past, which are supposedly related to GCC bug 72873 [1], reported when GCC 7 was not yet released. This bug only applies to x86-64, and does not apply to other architectures. While the test case for GCC bug 72873 triggers issues with released versions of GCC, the instrumented kernel code compiles fine for all configurations I have tried, and it is unclear how the two cases are/were related. As we can't reproduce the kernel build failures, let's instrument cmpxchg_double*() again. We can revisit the issue if build failures reappear. Signed-off-by: Mark Rutland Acked-by: Peter Zijlstra (Intel) Acked-by: Will Deacon Cc: Arnd Bergmann Cc: Boqun Feng Cc: Dmitry Vyukov Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Cc: andy.shevchenko@gmail.com Cc: aryabinin@virtuozzo.com Cc: catalin.marinas@arm.com Cc: glider@google.com Cc: linux-arm-kernel@lists.infradead.org Cc: parri.andrea@gmail.com Cc: peter@hurleysoftware.com Link: http://lkml.kernel.org/r/20180716113017.3909-6-mark.rutland@arm.com Signed-off-by: Ingo Molnar

locking/atomics: Instrument xchg()

2018-07-25T09:53:59Z

While we instrument all of the (non-relaxed) atomic_*() functions and cmpxchg(), we missed xchg(). Let's add instrumentation for xchg(), fixing up x86 to implement arch_xchg(). Signed-off-by: Mark Rutland Acked-by: Peter Zijlstra (Intel) Acked-by: Will Deacon Cc: Boqun Feng Cc: Dmitry Vyukov Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Cc: andy.shevchenko@gmail.com Cc: arnd@arndb.de Cc: aryabinin@virtuozzo.com Cc: catalin.marinas@arm.com Cc: glider@google.com Cc: linux-arm-kernel@lists.infradead.org Cc: parri.andrea@gmail.com Cc: peter@hurleysoftware.com Link: http://lkml.kernel.org/r/20180716113017.3909-5-mark.rutland@arm.com Signed-off-by: Ingo Molnar

locking/atomics: Simplify cmpxchg() instrumentation

2018-07-25T09:53:58Z

Currently we define some fairly verbose wrappers for the cmpxchg() family so that we can pass a pointer and size into kasan_check_write(). The wrappers duplicate the size-switching logic necessary in arch code, and only work for scalar types. On some architectures, (cmp)xchg are used on non-scalar types, and thus the instrumented wrappers need to be able to handle this. We could take the type-punning logic from {READ,WRITE}_ONCE(), but this makes the wrappers even more verbose, and requires several local variables in the macros. Instead, let's simplify the wrappers into simple macros which: * snapshot the pointer into a single local variable, called __ai_ptr to avoid conflicts with variables in the scope of the caller. * call kasan_check_write() on __ai_ptr. * invoke the relevant arch_*() function, passing the original arguments, bar __ai_ptr being substituted for ptr. There should be no functional change as a result of this patch. Signed-off-by: Mark Rutland Acked-by: Peter Zijlstra (Intel) Acked-by: Will Deacon Cc: Boqun Feng Cc: Dmitry Vyukov Cc: Linus Torvalds Cc: Peter Zijlstra Cc: Thomas Gleixner Cc: andy.shevchenko@gmail.com Cc: arnd@arndb.de Cc: aryabinin@virtuozzo.com Cc: catalin.marinas@arm.com Cc: glider@google.com Cc: linux-arm-kernel@lists.infradead.org Cc: parri.andrea@gmail.com Cc: peter@hurleysoftware.com Link: http://lkml.kernel.org/r/20180716113017.3909-4-mark.rutland@arm.com Signed-off-by: Ingo Molnar

atomics/treewide: Make conditional inc/dec ops optional

2018-06-21T12:25:24Z

The conditional inc/dec ops differ for atomic_t and atomic64_t: - atomic_inc_unless_positive() is optional for atomic_t, and doesn't exist for atomic64_t. - atomic_dec_unless_negative() is optional for atomic_t, and doesn't exist for atomic64_t. - atomic_dec_if_positive is optional for atomic_t, and is mandatory for atomic64_t. Let's make these consistently optional for both. At the same time, let's clean up the existing fallbacks to use atomic_try_cmpxchg(). The instrumented atomics are updated accordingly. There should be no functional change as a result of this patch. Signed-off-by: Mark Rutland Reviewed-by: Will Deacon Acked-by: Peter Zijlstra (Intel) Cc: Boqun Feng Cc: Linus Torvalds Cc: Thomas Gleixner Link: https://lore.kernel.org/lkml/20180621121321.4761-18-mark.rutland@arm.com Signed-off-by: Ingo Molnar

atomics/treewide: Make unconditional inc/dec ops optional

2018-06-21T12:25:24Z

Many of the inc/dec ops are mandatory, but for most architectures inc/dec are simply trivial wrappers around their corresponding add/sub ops. Let's make all the inc/dec ops optional, so that we can get rid of these boilerplate wrappers. The instrumented atomics are updated accordingly. There should be no functional change as a result of this patch. Signed-off-by: Mark Rutland Reviewed-by: Will Deacon Acked-by: Peter Zijlstra (Intel) Acked-by: Palmer Dabbelt Cc: Boqun Feng Cc: Linus Torvalds Cc: Thomas Gleixner Link: https://lore.kernel.org/lkml/20180621121321.4761-17-mark.rutland@arm.com Signed-off-by: Ingo Molnar

atomics/treewide: Make test ops optional

2018-06-21T12:25:24Z

Some of the atomics return the result of a test applied after the atomic operation, and almost all architectures implement these as trivial wrappers around the underlying atomic. Specifically: * _inc_and_test(v) is (_inc_return(v) == 0) * _dec_and_test(v) is (_dec_return(v) == 0) * _sub_and_test(i, v) is (_sub_return(i, v) == 0) * _add_negative(i, v) is (_add_return(i, v) < 0) Rather than have these definitions duplicated in all architectures, with minor inconsistencies in formatting and documentation, let's make these operations optional, with default fallbacks as above. Implementations must now provide a preprocessor symbol. The instrumented atomics are updated accordingly. Both x86 and m68k have custom implementations, which are left as-is, given preprocessor symbols to avoid being overridden. There should be no functional change as a result of this patch. Signed-off-by: Mark Rutland Reviewed-by: Will Deacon Acked-by: Geert Uytterhoeven Acked-by: Peter Zijlstra (Intel) Acked-by: Palmer Dabbelt Cc: Boqun Feng Cc: Linus Torvalds Cc: Thomas Gleixner Link: https://lore.kernel.org/lkml/20180621121321.4761-16-mark.rutland@arm.com Signed-off-by: Ingo Molnar

atomics/treewide: Make atomic64_fetch_add_unless() optional

2018-06-21T12:25:24Z

Architectures with atomic64_fetch_add_unless() provide a preprocessor symbol if they do so, and all other architectures have trivial C implementations of atomic64_add_unless() which are near-identical. Let's unify the trivial definitions of atomic64_fetch_add_unless() in , so that we always have both atomic64_fetch_add_unless() and atomic64_add_unless() with less boilerplate code. This means that atomic64_add_unless() is always implemented in core code, and the instrumented atomics are updated accordingly. There should be no functional change as a result of this patch. Signed-off-by: Mark Rutland Reviewed-by: Will Deacon Acked-by: Peter Zijlstra (Intel) Cc: Boqun Feng Cc: Linus Torvalds Cc: Thomas Gleixner Link: https://lore.kernel.org/lkml/20180621121321.4761-15-mark.rutland@arm.com Signed-off-by: Ingo Molnar

atomics: Prepare for atomic64_fetch_add_unless()

2018-06-21T12:22:34Z

Currently all architectures must implement atomic_fetch_add_unless(), with common code providing atomic_add_unless(). Architectures must also implement atomic64_add_unless() directly, with no corresponding atomic64_fetch_add_unless(). This divergence is unfortunate, and means that the APIs for atomic_t, atomic64_t, and atomic_long_t differ. In preparation for unifying things, with architectures providing atomic64_fetch_add_unless, this patch adds a generic atomic64_add_unless() which will use atomic64_fetch_add_unless(). The instrumented atomics are updated to take this case into account. There should be no functional change as a result of this patch. Signed-off-by: Mark Rutland Reviewed-by: Will Deacon Acked-by: Peter Zijlstra (Intel) Cc: Albert Ou Cc: Arnd Bergmann Cc: Benjamin Herrenschmidt Cc: Boqun Feng Cc: Ivan Kokshaysky Cc: Linus Torvalds Cc: Matt Turner Cc: Michael Ellerman Cc: Palmer Dabbelt Cc: Paul Mackerras Cc: Richard Henderson Cc: Russell King Cc: Thomas Gleixner Cc: Vineet Gupta Link: https://lore.kernel.org/lkml/20180621121321.4761-8-mark.rutland@arm.com Signed-off-by: Ingo Molnar