cpufeatures: check OS register support

cpufeatures/CHANGELOG.md

@@ -5,6 +5,12 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## 0.2.8 (2023-06-15)
+### Fixed
+- Check OS register support on x86 targets ([#919])
+
+[#919]: https://github.com/RustCrypto/utils/issues/919
+
 ## 0.2.7 (2023-04-20)
 ### Added
 - Support freestanding/UEFI `x86` targets ([#821])

cpufeatures/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "cpufeatures"
-version = "0.2.7"
+version = "0.2.8"
 description = """
 Lightweight runtime CPU feature detection for x86/x86_64 and aarch64 with
 no_std support and support for mobile targets including Android and iOS

cpufeatures/src/x86.rs

@@ -3,23 +3,23 @@
 //! Portable, `no_std`-friendly implementation that relies on the x86 `CPUID`
 //! instruction for feature detection.
 
-// Evaluate the given `$body` expression any of the supplied target features
-// are not enabled. Otherwise returns true.
-//
-// The `$body` expression is not evaluated on SGX targets, and returns false
-// on these targets unless *all* supplied target features are enabled.
+/// Evaluate the given `$body` expression any of the supplied target features
+/// are not enabled. Otherwise returns true.
+///
+/// The `$body` expression is not evaluated on SGX targets, and returns false
+/// on these targets unless *all* supplied target features are enabled.
 #[macro_export]
 #[doc(hidden)]
 macro_rules! __unless_target_features {
     ($($tf:tt),+ => $body:expr ) => {{
         #[cfg(not(all($(target_feature=$tf,)*)))]
         {
-            #[cfg(not(any(target_env = "sgx", target_os = "none", target_os = "uefi")))]
+            #[cfg(not(any(target_env = "sgx", target_os = "", target_os = "uefi")))]
             $body
 
             // CPUID is not available on SGX. Freestanding and UEFI targets
             // do not support SIMD features with default compilation flags.
-            #[cfg(any(target_env = "sgx", target_os = "none", target_os = "uefi"))]
+            #[cfg(any(target_env = "sgx", target_os = "", target_os = "uefi"))]
             false
         }
 
@@ -28,7 +28,7 @@ macro_rules! __unless_target_features {
     }};
 }
 
-// Use CPUID to detect the presence of all supplied target features.
+/// Use CPUID to detect the presence of all supplied target features.
 #[macro_export]
 #[doc(hidden)]
 macro_rules! __detect_target_features {
@@ -61,54 +61,85 @@ macro_rules! __detect_target_features {
     }};
 }
 
+/// Check that OS supports required SIMD registers
+#[macro_export]
+#[doc(hidden)]
+macro_rules! __xgetbv {
+    ($cr:expr, $mask:expr) => {{
+        #[cfg(target_arch = "x86")]
+        use core::arch::x86 as arch;
+        #[cfg(target_arch = "x86_64")]
+        use core::arch::x86_64 as arch;
+
+        // Check bits 26 and 27
+        let xmask = 0b11 << 26;
+        let xsave = $cr[0].ecx & xmask == xmask;
+        if xsave {
+            let xcr0 = unsafe { arch::_xgetbv(arch::_XCR_XFEATURE_ENABLED_MASK) };
+            (xcr0 & $mask) == $mask
+        } else {
+            false
+        }
+    }};
+}
+
 macro_rules! __expand_check_macro {
-    ($(($name:tt $(, $i:expr, $reg:ident, $offset:expr)*)),* $(,)?) => {
+    ($(($name:tt, $reg_cap:tt $(, $i:expr, $reg:ident, $offset:expr)*)),* $(,)?) => {
         #[macro_export]
         #[doc(hidden)]
         macro_rules! check {
             $(
-                ($cr:expr, $name) => {
-                    true
+                ($cr:expr, $name) => {{
+                    // Register bits are listed here:
+                    // https://wiki.osdev.org/CPU_Registers_x86#Extended_Control_Registers
+                    let reg_cap = match $reg_cap {
+                        // Bit 1
+                        "xmm" => $crate::__xgetbv!($cr, 0b10),
+                        // Bits 1 and 2
+                        "ymm" => $crate::__xgetbv!($cr, 0b110),
+                        // Bits 1, 2, 5, 6, and 7
+                        "zmm" => $crate::__xgetbv!($cr, 0b1110_0110),
+                        _ => true,
+                    };
+                    reg_cap
                     $(
                         & ($cr[$i].$reg & (1 << $offset) != 0)
                     )*
-                };
+                }};
             )*
         }
     };
 }
 
-// Note that according to the [Intel manual][0] AVX2 and FMA require
-// that we check availability of AVX before using them.
-//
-// [0]: https://www.intel.com/content/dam/develop/external/us/en/documents/36945
 __expand_check_macro! {
-    ("mmx", 0, edx, 23),
-    ("sse", 0, edx, 25),
-    ("sse2", 0, edx, 26),
-    ("sse3", 0, ecx, 0),
-    ("pclmulqdq", 0, ecx, 1),
-    ("ssse3", 0, ecx, 9),
-    ("fma", 0, ecx, 28, 0, ecx, 12),
-    ("sse4.1", 0, ecx, 19),
-    ("sse4.2", 0, ecx, 20),
-    ("popcnt", 0, ecx, 23),
-    ("aes", 0, ecx, 25),
-    ("avx", 0, ecx, 28),
-    ("rdrand", 0, ecx, 30),
-    ("sgx", 1, ebx, 2),
-    ("bmi1", 1, ebx, 3),
-    ("avx2", 0, ecx, 28, 1, ebx, 5),
-    ("bmi2", 1, ebx, 8),
-    ("avx512f", 1, ebx, 16),
-    ("avx512dq", 1, ebx, 17),
-    ("rdseed", 1, ebx, 18),
-    ("adx", 1, ebx, 19),
-    ("avx512ifma", 1, ebx, 21),
-    ("avx512pf", 1, ebx, 26),
-    ("avx512er", 1, ebx, 27),
-    ("avx512cd", 1, ebx, 28),
-    ("sha", 1, ebx, 29),
-    ("avx512bw", 1, ebx, 30),
-    ("avx512vl", 1, ebx, 31),
+    ("sse3", "xmm", 0, ecx, 0),
+    ("pclmulqdq", "xmm", 0, ecx, 1),
+    ("ssse3", "xmm", 0, ecx, 9),
+    ("fma", "xmm", 0, ecx, 12, 0, ecx, 28),
+    ("sse4.1", "xmm", 0, ecx, 19),
+    ("sse4.2", "xmm", 0, ecx, 20),
+    ("popcnt", "", 0, ecx, 23),
+    ("aes", "xmm", 0, ecx, 25),
+    ("avx", "xmm", 0, ecx, 28),
+    ("rdrand", "", 0, ecx, 30),
+
+    ("mmx", "", 0, edx, 23),
+    ("sse", "xmm", 0, edx, 25),
+    ("sse2", "xmm", 0, edx, 26),
+
+    ("sgx", "", 1, ebx, 2),
+    ("bmi1", "", 1, ebx, 3),
+    ("bmi2", "", 1, ebx, 8),
+    ("avx2", "ymm", 1, ebx, 5, 0, ecx, 28),
+    ("avx512f", "zmm", 1, ebx, 16),
+    ("avx512dq", "zmm", 1, ebx, 17),
+    ("rdseed", "", 1, ebx, 18),
+    ("adx", "", 1, ebx, 19),
+    ("avx512ifma", "zmm", 1, ebx, 21),
+    ("avx512pf", "zmm", 1, ebx, 26),
+    ("avx512er", "zmm", 1, ebx, 27),
+    ("avx512cd", "zmm", 1, ebx, 28),
+    ("sha", "xmm", 1, ebx, 29),
+    ("avx512bw", "zmm", 1, ebx, 30),
+    ("avx512vl", "zmm", 1, ebx, 31),
 }