Compiler CSE and SSA GVN optimizations can cause the address dependency of addresses returned by rcu_dereference to be lost when comparing those pointers with either constants or previously loaded pointers.
Introduce ptr_eq() to compare two addresses while preserving the address dependencies for later use of the address. It should be used when comparing an address returned by rcu_dereference(). This is needed to prevent the compiler CSE and SSA GVN optimizations from using @a (or @b) in places where the source refers to @b (or @a) based on the fact that after the comparison, the two are known to be equal, which does not preserve address dependencies and allows the following misordering speculations: - If @b is a constant, the compiler can issue the loads which depend on @a before loading @a. - If @b is a register populated by a prior load, weakly-ordered CPUs can speculate loads which depend on @a before loading @a. The same logic applies with @a and @b swapped. The header "linux/ptr_eq.h" is the expected include target. It contains the documentation of the ptr_eq() API. An architecture specific implementation of the pointer comparison can be implemented by each architecture as asm/ptr_eq.h. The x86 implementation is provided initially. If no implementation header is present for the architecture, an arch-agnostic fallback based on OPTIMIZER_HIDE_VAR is included from asm-generic. The arguments to ptr_eq() are const volatile to take advantage of implicit conversion that adds cv-qualifiers, which are thus added implicitly. Suggested-by: Linus Torvalds <[email protected]> Suggested-by: Boqun Feng <[email protected]> Signed-off-by: Mathieu Desnoyers <[email protected]> Reviewed-by: Boqun Feng <[email protected]> Cc: Greg Kroah-Hartman <[email protected]> Cc: Sebastian Andrzej Siewior <[email protected]> Cc: "Paul E. McKenney" <[email protected]> Cc: Will Deacon <[email protected]> Cc: Peter Zijlstra <[email protected]> Cc: Boqun Feng <[email protected]> Cc: Alan Stern <[email protected]> Cc: John Stultz <[email protected]> Cc: Linus Torvalds <[email protected]> Cc: Boqun Feng <[email protected]> Cc: Frederic Weisbecker <[email protected]> Cc: Joel Fernandes <[email protected]> Cc: Josh Triplett <[email protected]> Cc: Uladzislau Rezki <[email protected]> Cc: Steven Rostedt <[email protected]> Cc: Lai Jiangshan <[email protected]> Cc: Zqiang <[email protected]> Cc: Ingo Molnar <[email protected]> Cc: Waiman Long <[email protected]> Cc: Mark Rutland <[email protected]> Cc: Thomas Gleixner <[email protected]> Cc: Vlastimil Babka <[email protected]> Cc: [email protected] Cc: Mateusz Guzik <[email protected]> Cc: Gary Guo <[email protected]> Cc: [email protected] Cc: [email protected] Cc: [email protected] Cc: Nikita Popov <[email protected]> Cc: [email protected] Cc: Lian Wang <[email protected]> Cc: Kunwu Chan <[email protected]> --- Changes since v2: - Rename ptreq.h to ptr_eq.h (x86, asm-generic and linux headers). Adapt the include guards accordingly. Changes since v1: - Move to linux/ptreq.h, implement x86-specific comparison and asm-generic header (fallback). Changes since v0: - Include feedback from Alan Stern. --- arch/x86/include/asm/ptr_eq.h | 16 +++++++++ include/asm-generic/Kbuild | 1 + include/asm-generic/ptr_eq.h | 16 +++++++++ include/linux/ptr_eq.h | 63 +++++++++++++++++++++++++++++++++++ 4 files changed, 96 insertions(+) create mode 100644 arch/x86/include/asm/ptr_eq.h create mode 100644 include/asm-generic/ptr_eq.h create mode 100644 include/linux/ptr_eq.h diff --git a/arch/x86/include/asm/ptr_eq.h b/arch/x86/include/asm/ptr_eq.h new file mode 100644 index 000000000000..8e521cabe24b --- /dev/null +++ b/arch/x86/include/asm/ptr_eq.h @@ -0,0 +1,16 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _ASM_X86_PTR_EQ_H +#define _ASM_X86_PTR_EQ_H + +#include <linux/types.h> +#include <asm/asm.h> + +static __always_inline +bool ptr_eq(const volatile void *a, const volatile void *b) +{ + bool ret; + asm(__ASM_SIZE(cmp) " %1,%2" : "=@ccz" (ret) : "r" (a), "r" (b)); + return ret; +} + +#endif /* _ASM_X86_PTR_EQ_H */ diff --git a/include/asm-generic/Kbuild b/include/asm-generic/Kbuild index 2bc00c67dc54..86cee0dfd47f 100644 --- a/include/asm-generic/Kbuild +++ b/include/asm-generic/Kbuild @@ -47,6 +47,7 @@ mandatory-y += percpu.h mandatory-y += percpu_types.h mandatory-y += pgalloc.h mandatory-y += preempt.h +mandatory-y += ptr_eq.h mandatory-y += rqspinlock.h mandatory-y += runtime-const.h mandatory-y += rwonce.h diff --git a/include/asm-generic/ptr_eq.h b/include/asm-generic/ptr_eq.h new file mode 100644 index 000000000000..2bd1c3b8a56f --- /dev/null +++ b/include/asm-generic/ptr_eq.h @@ -0,0 +1,16 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef __ASM_GENERIC_PTR_EQ_H +#define __ASM_GENERIC_PTR_EQ_H + +#include <linux/types.h> +#include <linux/compiler.h> + +static __always_inline +bool ptr_eq(const volatile void *a, const volatile void *b) +{ + OPTIMIZER_HIDE_VAR(a); + OPTIMIZER_HIDE_VAR(b); + return a == b; +} + +#endif /* __ASM_GENERIC_PTR_EQ_H */ diff --git a/include/linux/ptr_eq.h b/include/linux/ptr_eq.h new file mode 100644 index 000000000000..7a2bbf6e64b3 --- /dev/null +++ b/include/linux/ptr_eq.h @@ -0,0 +1,63 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _LINUX_PTR_EQ_H +#define _LINUX_PTR_EQ_H + +/* + * Compare two addresses while preserving the address dependencies for + * later use of the address. It should be used when comparing an address + * returned by rcu_dereference(). + * + * This is needed to prevent the compiler CSE and SSA GVN optimizations + * from using @a (or @b) in places where the source refers to @b (or @a) + * based on the fact that after the comparison, the two are known to be + * equal, which does not preserve address dependencies and allows the + * following misordering speculations: + * + * - If @b is a constant, the compiler can issue the loads which depend + * on @a before loading @a. + * - If @b is a register populated by a prior load, weakly-ordered + * CPUs can speculate loads which depend on @a before loading @a. + * + * The same logic applies with @a and @b swapped. + * + * Return value: true if pointers are equal, false otherwise. + * + * The compiler barrier() is ineffective at fixing this issue. It does + * not prevent the compiler CSE from losing the address dependency: + * + * int fct_2_volatile_barriers(void) + * { + * int *a, *b; + * + * do { + * a = READ_ONCE(p); + * asm volatile ("" : : : "memory"); + * b = READ_ONCE(p); + * } while (a != b); + * asm volatile ("" : : : "memory"); <-- barrier() + * return *b; + * } + * + * With gcc 14.2 (arm64): + * + * fct_2_volatile_barriers: + * adrp x0, .LANCHOR0 + * add x0, x0, :lo12:.LANCHOR0 + * .L2: + * ldr x1, [x0] <-- x1 populated by first load. + * ldr x2, [x0] + * cmp x1, x2 + * bne .L2 + * ldr w0, [x1] <-- x1 is used for access which should depend on b. + * ret + * + * On weakly-ordered architectures, this lets CPU speculation use the + * result from the first load to speculate "ldr w0, [x1]" before + * "ldr x2, [x0]". + * Based on the RCU documentation, the control dependency does not + * prevent the CPU from speculating loads. + */ + +#include <asm/ptr_eq.h> + +#endif /* _LINUX_PTR_EQ_H */ -- 2.43.0
