detect: Support PowerPC64 OpenBSD, prepare support for RISC-V

.github/.cspell/project-dictionary.txt

@@ -47,6 +47,7 @@ exynos
 getauxval
 getpid
 HWCAP
+hwprobe
 ifunc
 includepath
 Iotel

CHANGELOG.md

@@ -168,13 +168,13 @@ Note: In this file, do not use the hard wrap in the middle of a sentence for com
 
 - Optimize x86_64 outline-atomics for 128-bit atomics.
   - Support outline-atomics for cmpxchg16b on Rust 1.69+ (i.e., on Rust 1.69+, x86_64 128-bit atomics is lock-free on all Intel chips and almost all AMD chips, even if cmpxchg16b is not available at compile-time.). Previously it was only nightly. ([#80](https://github.com/taiki-e/portable-atomic/pull/80))
-  - portable-atomic no longer enables outline-atomics on target where run-time feature detection is not available. ([#80](https://github.com/taiki-e/portable-atomic/pull/80))
+  - portable-atomic no longer enables outline-atomics on target where run-time CPU feature detection is not available. ([#80](https://github.com/taiki-e/portable-atomic/pull/80))
 
 - Optimize aarch64 outline-atomics for 128-bit atomics.
   - Support more targets and improve performance. ([#63](https://github.com/taiki-e/portable-atomic/pull/63), [#64](https://github.com/taiki-e/portable-atomic/pull/64), [#67](https://github.com/taiki-e/portable-atomic/pull/67), [#69](https://github.com/taiki-e/portable-atomic/pull/69), [#75](https://github.com/taiki-e/portable-atomic/pull/75), [#76](https://github.com/taiki-e/portable-atomic/pull/76), [#77](https://github.com/taiki-e/portable-atomic/pull/77))
     See the [`atomic128` module's readme](https://github.com/taiki-e/portable-atomic/blob/HEAD/src/imp/atomic128/README.md#run-time-feature-detection) for a list of platforms that support outline-atomics.
     Most of these improvements have already been [submitted and accepted in rust-lang/stdarch](https://github.com/rust-lang/stdarch/pulls?q=is%3Apr+author%3Ataiki-e+std_detect) and will soon be available in `std::arch::is_aarch64_feature_detected`.
-  - portable-atomic no longer enables outline-atomics on target where run-time feature detection is not available.
+  - portable-atomic no longer enables outline-atomics on target where run-time CPU feature detection is not available.
 
 - Performance improvements. ([#70](https://github.com/taiki-e/portable-atomic/pull/70), [#81](https://github.com/taiki-e/portable-atomic/pull/81), [6c189ae](https://github.com/taiki-e/portable-atomic/commit/6c189ae1792ce0c08b4f56b6e6c256c223475ce2), [13c92b0](https://github.com/taiki-e/portable-atomic/commit/13c92b015a8e8646a4b885229157547354d03b9e), etc.)
 

DEVELOPMENT.md

@@ -16,7 +16,7 @@ portable-atomic/
 ├── src/
 │   ├── imp/
 │   │   ├── atomic128/            -- 128-bit atomic implementations (mainly by asm)
-│   │   │   └── detect/           -- Run-time feature detection implementations used for outline-atomics
+│   │   │   └── detect/           -- Run-time CPU feature detection implementations used for outline-atomics
 │   │   ├── arm_linux.rs          -- 64-bit atomic implementation for pre-v6 ARM Linux/Android
 │   │   ├── core_atomic.rs        -- wrapper for core::sync::atomic types
 │   │   ├── fallback/             -- fallback implementation based on global locks

src/cfgs.rs

@@ -265,6 +265,7 @@ mod atomic_64_macros {
                         ),
                         target_os = "android",
                         target_os = "freebsd",
+                        all(target_os = "openbsd", portable_atomic_outline_atomics),
                     ),
                     not(any(miri, portable_atomic_sanitize_thread)),
                 ),
@@ -354,6 +355,7 @@ mod atomic_128_macros {
                         ),
                         target_os = "android",
                         target_os = "freebsd",
+                        all(target_os = "openbsd", portable_atomic_outline_atomics),
                     ),
                     not(any(miri, portable_atomic_sanitize_thread)),
                 ),

src/imp/arm_linux.rs

@@ -3,7 +3,7 @@
 // 64-bit atomic implementation using kuser_cmpxchg64 on pre-v6 ARM Linux/Android.
 //
 // Refs:
-// - https://www.kernel.org/doc/Documentation/arm/kernel_user_helpers.txt
+// - https://github.com/torvalds/linux/blob/v6.10/Documentation/arch/arm/kernel_user_helpers.rst
 // - https://github.com/rust-lang/compiler-builtins/blob/0.1.88/src/arm_linux.rs
 //
 // Note: On Miri and ThreadSanitizer which do not support inline assembly, we don't use
@@ -20,7 +20,7 @@ use core::{arch::asm, cell::UnsafeCell, mem, sync::atomic::Ordering};
 
 use crate::utils::{Pair, U64};
 
-// https://www.kernel.org/doc/Documentation/arm/kernel_user_helpers.txt
+// https://github.com/torvalds/linux/blob/v6.10/Documentation/arch/arm/kernel_user_helpers.rst
 const KUSER_HELPER_VERSION: usize = 0xFFFF0FFC;
 // __kuser_helper_version >= 5 (kernel version 3.1+)
 const KUSER_CMPXCHG64: usize = 0xFFFF0F60;

src/imp/atomic128/README.md

@@ -28,11 +28,11 @@ Implementations with inline assembly generate assemblies almost equivalent to th
 
 As 128-bit atomics-related APIs stabilize in the standard library, implementations with inline assembly are planned to be updated to get the benefits of both.
 
-## Run-time feature detection
+## Run-time CPU feature detection
 
-[detect](detect) module has run-time feature detection implementations.
+[detect](detect) module has run-time CPU feature detection implementations.
 
-Here is the table of targets that support run-time feature detection and the instruction or API used:
+Here is the table of targets that support run-time CPU feature detection and the instruction or API used:
 
 | target_arch | target_os/target_env | instruction/API | features | note |
 | ----------- | -------------------- | --------------- | -------- | ---- |
@@ -47,6 +47,7 @@ Here is the table of targets that support run-time feature detection and the ins
 | aarch64     | fuchsia              | zx_system_get_features | lse | Enabled by default |
 | powerpc64   | linux                | getauxval       | all      | Disabled by default |
 | powerpc64   | freebsd              | elf_aux_info    | all      | Disabled by default |
+| powerpc64   | openbsd              | elf_aux_info    | all      | Disabled by default |
 
 Run-time detection is enabled by default on most targets and can be disabled with `--cfg portable_atomic_no_outline_atomics`.
 

src/imp/atomic128/aarch64.rs

@@ -119,6 +119,12 @@ mod detect;
 #[cfg(any(target_os = "linux", target_os = "android", target_os = "freebsd"))]
 #[path = "detect/aarch64_aa64reg.rs"]
 mod detect_aa64reg;
+// TODO: OpenBSD 7.6+
+// #[cfg(test)]
+// #[cfg(not(portable_atomic_no_outline_atomics))]
+// #[cfg(target_os = "openbsd")]
+// #[path = "detect/auxv.rs"]
+// mod detect_auxv;
 #[cfg(test)]
 #[cfg(not(portable_atomic_no_outline_atomics))]
 #[cfg(target_os = "macos")]

src/imp/atomic128/detect/aarch64_aa64reg.rs

@@ -1,14 +1,14 @@
 // SPDX-License-Identifier: Apache-2.0 OR MIT
 
-// Run-time feature detection on aarch64 Linux/FreeBSD/NetBSD/OpenBSD by parsing system registers.
+// Run-time CPU feature detection on AArch64 Linux/Android/FreeBSD/NetBSD/OpenBSD by parsing system registers.
 //
 // As of nightly-2023-01-23, is_aarch64_feature_detected doesn't support run-time detection on NetBSD/OpenBSD.
 // https://github.com/rust-lang/stdarch/blob/a0c30f3e3c75adcd6ee7efc94014ebcead61c507/crates/std_detect/src/detect/mod.rs
 // https://github.com/rust-lang/stdarch/pull/1374
 //
 // Refs:
 // - https://developer.arm.com/documentation/ddi0601/latest/AArch64-Registers
-// - https://www.kernel.org/doc/Documentation/arm64/cpu-feature-registers.txt
+// - https://github.com/torvalds/linux/blob/v6.10/Documentation/arch/arm64/cpu-feature-registers.rst
 // - https://github.com/rust-lang/stdarch/blob/a0c30f3e3c75adcd6ee7efc94014ebcead61c507/crates/std_detect/src/detect/os/aarch64.rs
 //
 // Supported platforms:
@@ -22,9 +22,10 @@
 //   https://github.com/openbsd/src/commit/d335af936b9d7dd9cf655cae1ce19560c45de6c8
 //
 // For now, this module is only used on NetBSD/OpenBSD.
-// On Linux/FreeBSD, this module is test-only:
-// - On Linux, this approach requires a higher kernel version than Rust supports,
-//   and also does not work with qemu-user (as of QEMU 7.2) and Valgrind.
+//
+// On Linux/Android/FreeBSD, we use auxv.rs and this module is test-only because:
+// - On Linux/Android, this approach requires a higher kernel version than Rust supports,
+//   and also does not work with qemu-user (as of 7.2) and Valgrind (as of 3.19).
 //   (Looking into HWCAP_CPUID in auxvec, it appears that Valgrind is setting it
 //   to false correctly, but qemu-user is setting it to true.)
 // - On FreeBSD, this approach does not work on FreeBSD 12 on QEMU (confirmed on
@@ -49,31 +50,31 @@ fn _detect(info: &mut CpuInfo) {
         aa64mmfr2,
     } = imp::aa64reg();
 
-    // ID_AA64ISAR0_EL1, Instruction Set Attribute Register 0
-    // https://developer.arm.com/documentation/ddi0601/2023-06/AArch64-Registers/ID-AA64ISAR0-EL1--AArch64-Instruction-Set-Attribute-Register-0?lang=en
+    // ID_AA64ISAR0_EL1, AArch64 Instruction Set Attribute Register 0
+    // https://developer.arm.com/documentation/ddi0601/2024-06/AArch64-Registers/ID-AA64ISAR0-EL1--AArch64-Instruction-Set-Attribute-Register-0
     let atomic = extract(aa64isar0, 23, 20);
-    if atomic >= 2 {
+    if atomic >= 0b0010 {
         info.set(CpuInfo::HAS_LSE);
         // we currently only use FEAT_LSE and FEAT_LSE2 in outline-atomics.
         #[cfg(test)]
         {
-            if atomic >= 3 {
+            if atomic >= 0b0011 {
                 info.set(CpuInfo::HAS_LSE128);
             }
         }
     }
     // we currently only use FEAT_LSE and FEAT_LSE2 in outline-atomics.
     #[cfg(test)]
     {
-        // ID_AA64ISAR1_EL1, Instruction Set Attribute Register 1
-        // https://developer.arm.com/documentation/ddi0601/2023-06/AArch64-Registers/ID-AA64ISAR1-EL1--AArch64-Instruction-Set-Attribute-Register-1?lang=en
-        if extract(aa64isar1, 23, 20) >= 3 {
+        // ID_AA64ISAR1_EL1, AArch64 Instruction Set Attribute Register 1
+        // https://developer.arm.com/documentation/ddi0601/2024-06/AArch64-Registers/ID-AA64ISAR1-EL1--AArch64-Instruction-Set-Attribute-Register-1
+        if extract(aa64isar1, 23, 20) >= 0b0011 {
             info.set(CpuInfo::HAS_RCPC3);
         }
     }
     // ID_AA64MMFR2_EL1, AArch64 Memory Model Feature Register 2
-    // https://developer.arm.com/documentation/ddi0601/2023-06/AArch64-Registers/ID-AA64MMFR2-EL1--AArch64-Memory-Model-Feature-Register-2?lang=en
-    if extract(aa64mmfr2, 35, 32) >= 1 {
+    // https://developer.arm.com/documentation/ddi0601/2024-06/AArch64-Registers/ID-AA64MMFR2-EL1--AArch64-Memory-Model-Feature-Register-2
+    if extract(aa64mmfr2, 35, 32) >= 0b0001 {
         info.set(CpuInfo::HAS_LSE2);
     }
 }

src/imp/atomic128/detect/aarch64_fuchsia.rs

@@ -1,6 +1,6 @@
 // SPDX-License-Identifier: Apache-2.0 OR MIT
 
-// Run-time feature detection on aarch64 Fuchsia by using zx_system_get_features.
+// Run-time CPU feature detection on aarch64 Fuchsia by using zx_system_get_features.
 //
 // As of nightly-2023-01-23, is_aarch64_feature_detected doesn't support run-time detection on Fuchsia.
 // https://github.com/rust-lang/stdarch/blob/a0c30f3e3c75adcd6ee7efc94014ebcead61c507/crates/std_detect/src/detect/mod.rs

src/imp/atomic128/detect/aarch64_macos.rs

@@ -1,6 +1,6 @@
 // SPDX-License-Identifier: Apache-2.0 OR MIT
 
-// Run-time feature detection on aarch64 macOS by using sysctl.
+// Run-time CPU feature detection on aarch64 macOS by using sysctl.
 //
 // This module is currently only enabled on tests because aarch64 macOS always supports FEAT_LSE and FEAT_LSE2.
 // https://github.com/llvm/llvm-project/blob/llvmorg-18.1.2/llvm/include/llvm/TargetParser/AArch64TargetParser.h#L728

src/imp/atomic128/detect/aarch64_windows.rs

@@ -1,6 +1,6 @@
 // SPDX-License-Identifier: Apache-2.0 OR MIT
 
-// Run-time feature detection on aarch64 Windows by using IsProcessorFeaturePresent.
+// Run-time CPU feature detection on aarch64 Windows by using IsProcessorFeaturePresent.
 //
 // As of nightly-2023-01-23, is_aarch64_feature_detected doesn't support run-time detection of FEAT_LSE on Windows.
 // https://github.com/rust-lang/stdarch/blob/a0c30f3e3c75adcd6ee7efc94014ebcead61c507/crates/std_detect/src/detect/os/windows/aarch64.rs