Avoid using the x86-specific implementations on non-sse2 x86

Patch from Debian Fixes briansmith#1999
redox-os · Oct 18, 2024 · fce20f1 · fce20f1
1 parent fa98b49
commit fce20f1
Show file tree

Hide file tree

Showing 9 changed files with 55 additions and 40 deletions.
diff --git a/build.rs b/build.rs
@@ -427,7 +427,7 @@ fn build_c_code(
 
     generate_prefix_symbols_asm_headers(out_dir, ring_core_prefix).unwrap();
 
-    let (asm_srcs, obj_srcs) = if let Some(asm_target) = asm_target {
+    let (mut asm_srcs, mut obj_srcs) = if let Some(asm_target) = asm_target {
         let perlasm_src_dsts = perlasm_src_dsts(asm_dir, asm_target);
 
         if !use_pregenerated {
@@ -451,6 +451,21 @@ fn build_c_code(
         (vec![], vec![])
     };
 
+    use std::env;
+
+    if target.arch == "x86" {
+        let mut havesse2 = false;
+        for target_feature in env::var("CARGO_CFG_TARGET_FEATURE").unwrap_or("".to_string()).split(",") {
+            if target_feature == "sse2" {
+                havesse2 = true;
+            }
+        }
+        if !havesse2 {
+            asm_srcs = vec![];
+            obj_srcs = vec![];
+        }
+    }
+
     let core_srcs = sources_for_arch(&target.arch)
         .into_iter()
         .filter(|p| !is_perlasm(p))

diff --git a/src/aead/aes.rs b/src/aead/aes.rs
@@ -149,7 +149,7 @@ impl Key {
                 target_arch = "aarch64",
                 target_arch = "arm",
                 target_arch = "x86_64",
-                target_arch = "x86"
+                all(target_arch = "x86", target_feature = "sse2")
             ))]
             Implementation::HWAES => {
                 set_encrypt_key!(aes_hw_set_encrypt_key, bytes, key_bits, &mut key)?
@@ -159,7 +159,7 @@ impl Key {
                 target_arch = "aarch64",
                 target_arch = "arm",
                 target_arch = "x86_64",
-                target_arch = "x86"
+                all(target_arch = "x86", target_feature = "sse2")
             ))]
             Implementation::VPAES_BSAES => {
                 set_encrypt_key!(vpaes_set_encrypt_key, bytes, key_bits, &mut key)?
@@ -180,15 +180,15 @@ impl Key {
                 target_arch = "aarch64",
                 target_arch = "arm",
                 target_arch = "x86_64",
-                target_arch = "x86"
+                all(target_arch = "x86", target_feature = "sse2")
             ))]
             Implementation::HWAES => encrypt_block!(aes_hw_encrypt, a, self),
 
             #[cfg(any(
                 target_arch = "aarch64",
                 target_arch = "arm",
                 target_arch = "x86_64",
-                target_arch = "x86"
+                all(target_arch = "x86", target_feature = "sse2")
             ))]
             Implementation::VPAES_BSAES => encrypt_block!(vpaes_encrypt, a, self),
 
@@ -219,7 +219,7 @@ impl Key {
                 target_arch = "aarch64",
                 target_arch = "arm",
                 target_arch = "x86_64",
-                target_arch = "x86"
+                all(target_arch = "x86", target_feature = "sse2")
             ))]
             Implementation::HWAES => {
                 ctr32_encrypt_blocks!(aes_hw_ctr32_encrypt_blocks, in_out, src, &self.inner, ctr)
@@ -263,7 +263,7 @@ impl Key {
                 ctr32_encrypt_blocks!(vpaes_ctr32_encrypt_blocks, in_out, src, &self.inner, ctr)
             }
 
-            #[cfg(target_arch = "x86")]
+            #[cfg(all(target_arch = "x86", target_feature = "sse2"))]
             Implementation::VPAES_BSAES => {
                 super::shift::shift_full_blocks(in_out, src, |input| {
                     self.encrypt_iv_xor_block(ctr.increment(), Block::from(input), cpu_features)
@@ -365,7 +365,7 @@ pub enum Implementation {
         target_arch = "aarch64",
         target_arch = "arm",
         target_arch = "x86_64",
-        target_arch = "x86"
+        all(target_arch = "x86", target_feature = "sse2")
     ))]
     HWAES = 1,
 
@@ -374,7 +374,7 @@ pub enum Implementation {
         target_arch = "aarch64",
         target_arch = "arm",
         target_arch = "x86_64",
-        target_arch = "x86"
+        all(target_arch = "x86", target_feature = "sse2")
     ))]
     VPAES_BSAES = 2,
 
@@ -387,7 +387,7 @@ fn detect_implementation(cpu_features: cpu::Features) -> Implementation {
         target_arch = "aarch64",
         target_arch = "arm",
         target_arch = "x86_64",
-        target_arch = "x86"
+        all(target_arch = "x86", target_feature = "sse2")
     )))]
     let _cpu_features = cpu_features;
 
@@ -398,14 +398,14 @@ fn detect_implementation(cpu_features: cpu::Features) -> Implementation {
         }
     }
 
-    #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
+    #[cfg(any(target_arch = "x86_64", all(target_arch = "x86", target_feature = "sse2")))]
     {
         if cpu::intel::AES.available(cpu_features) {
             return Implementation::HWAES;
         }
     }
 
-    #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
+    #[cfg(any(target_arch = "x86_64", all(target_arch = "x86", target_feature = "sse2")))]
     {
         if cpu::intel::SSSE3.available(cpu_features) {
             return Implementation::VPAES_BSAES;

diff --git a/src/aead/chacha.rs b/src/aead/chacha.rs
@@ -20,7 +20,7 @@ use super::{quic::Sample, Nonce};
     not(any(
         target_arch = "aarch64",
         target_arch = "arm",
-        target_arch = "x86",
+        all(target_arch = "x86", target_feature = "sse2"),
         target_arch = "x86_64"
     ))
 ))]
@@ -75,7 +75,7 @@ impl Key {
         // has this limitation and come up with a better solution.
         //
         // https://rt.openssl.org/Ticket/Display.html?id=4362
-        if cfg!(any(target_arch = "arm", target_arch = "x86")) && src.start != 0 {
+        if cfg!(any(target_arch = "arm", all(target_arch = "x86", target_feature = "sse2"))) && src.start != 0 {
             let len = in_out.len() - src.start;
             in_out.copy_within(src, 0);
             self.encrypt_in_place(counter, &mut in_out[..len]);
@@ -91,7 +91,7 @@ impl Key {
         #[cfg(any(
             target_arch = "aarch64",
             target_arch = "arm",
-            target_arch = "x86",
+            all(target_arch = "x86", target_feature = "sse2"),
             target_arch = "x86_64"
         ))]
         #[inline(always)]
@@ -128,7 +128,7 @@ impl Key {
         #[cfg(not(any(
             target_arch = "aarch64",
             target_arch = "arm",
-            target_arch = "x86",
+            all(target_arch = "x86", target_feature = "sse2"),
             target_arch = "x86_64"
         )))]
         use fallback::ChaCha20_ctr32;
@@ -169,7 +169,7 @@ impl Counter {
         not(any(
             target_arch = "aarch64",
             target_arch = "arm",
-            target_arch = "x86",
+            all(target_arch = "x86", target_feature = "sse2"),
             target_arch = "x86_64"
         ))
     ))]
@@ -219,7 +219,7 @@ mod tests {
         let max_offset = if cfg!(any(
             target_arch = "aarch64",
             target_arch = "arm",
-            target_arch = "x86",
+            all(target_arch = "x86", target_feature = "sse2"),
             target_arch = "x86_64"
         )) {
             MAX_ALIGNMENT_AND_OFFSET

diff --git a/src/aead/gcm.rs b/src/aead/gcm.rs
@@ -57,7 +57,7 @@ impl Key {
                 target_arch = "aarch64",
                 target_arch = "arm",
                 target_arch = "x86_64",
-                target_arch = "x86"
+                all(target_arch = "x86", target_feature = "sse2")
             ))]
             Implementation::CLMUL => {
                 prefixed_extern! {
@@ -185,7 +185,7 @@ impl Context {
                 target_arch = "aarch64",
                 target_arch = "arm",
                 target_arch = "x86_64",
-                target_arch = "x86"
+                all(target_arch = "x86", target_feature = "sse2")
             ))]
             Implementation::CLMUL => {
                 prefixed_extern! {
@@ -236,7 +236,7 @@ impl Context {
                 target_arch = "aarch64",
                 target_arch = "arm",
                 target_arch = "x86_64",
-                target_arch = "x86"
+                all(target_arch = "x86", target_feature = "sse2")
             ))]
             Implementation::CLMUL => {
                 prefixed_extern! {
@@ -339,7 +339,7 @@ enum Implementation {
         target_arch = "aarch64",
         target_arch = "arm",
         target_arch = "x86_64",
-        target_arch = "x86"
+        all(target_arch = "x86", target_feature = "sse2")
     ))]
     CLMUL,
 
@@ -356,7 +356,7 @@ fn detect_implementation(cpu_features: cpu::Features) -> Implementation {
         target_arch = "aarch64",
         target_arch = "arm",
         target_arch = "x86_64",
-        target_arch = "x86"
+        all(target_arch = "x86", target_feature = "sse2")
     )))]
     let _cpu_features = cpu_features;
 
@@ -367,7 +367,7 @@ fn detect_implementation(cpu_features: cpu::Features) -> Implementation {
         }
     }
 
-    #[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
+    #[cfg(any(target_arch = "x86_64", all(target_arch = "x86", target_feature = "sse2")))]
     {
         if cpu::intel::FXSR.available(cpu_features) && cpu::intel::PCLMULQDQ.available(cpu_features)
         {

diff --git a/src/aead/shift.rs b/src/aead/shift.rs
@@ -14,7 +14,7 @@
 
 use super::block::{Block, BLOCK_LEN};
 
-#[cfg(target_arch = "x86")]
+#[cfg(all(target_arch = "x86", target_feature = "sse2"))]
 pub fn shift_full_blocks<F>(in_out: &mut [u8], src: core::ops::RangeFrom<usize>, mut transform: F)
 where
     F: FnMut(&[u8; BLOCK_LEN]) -> Block,

diff --git a/src/arithmetic/montgomery.rs b/src/arithmetic/montgomery.rs
@@ -128,7 +128,7 @@ unsafe fn mul_mont(
 #[cfg(not(any(
     target_arch = "aarch64",
     target_arch = "arm",
-    target_arch = "x86",
+    all(target_arch = "x86", target_feature = "sse2"),
     target_arch = "x86_64"
 )))]
 // TODO: Stop calling this from C and un-export it.
@@ -168,7 +168,7 @@ prefixed_export! {
     not(any(
         target_arch = "aarch64",
         target_arch = "arm",
-        target_arch = "x86",
+        all(target_arch = "x86", target_feature = "sse2"),
         target_arch = "x86_64"
     ))
 ))]
@@ -201,7 +201,7 @@ pub(super) fn limbs_from_mont_in_place(r: &mut [Limb], tmp: &mut [Limb], m: &[Li
 #[cfg(not(any(
     target_arch = "aarch64",
     target_arch = "arm",
-    target_arch = "x86",
+    all(target_arch = "x86", target_feature = "sse2"),
     target_arch = "x86_64"
 )))]
 fn limbs_mul(r: &mut [Limb], a: &[Limb], b: &[Limb]) {
@@ -223,7 +223,7 @@ fn limbs_mul(r: &mut [Limb], a: &[Limb], b: &[Limb]) {
         target_arch = "aarch64",
         target_arch = "arm",
         target_arch = "x86_64",
-        target_arch = "x86"
+        all(target_arch = "x86", target_feature = "sse2")
     ))
 ))]
 prefixed_extern! {
@@ -236,7 +236,7 @@ prefixed_extern! {
     target_arch = "aarch64",
     target_arch = "arm",
     target_arch = "x86_64",
-    target_arch = "x86"
+    all(target_arch = "x86", target_feature = "sse2")
 ))]
 prefixed_extern! {
     // `r` and/or 'a' and/or 'b' may alias.

diff --git a/src/cpu.rs b/src/cpu.rs
@@ -27,13 +27,13 @@ pub(crate) fn features() -> Features {
     #[cfg(any(target_arch = "aarch64", target_arch = "arm"))]
     use arm::init_global_shared_with_assembly;
 
-    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+    #[cfg(any(all(target_arch = "x86", target_feature = "sse2"), target_arch = "x86_64"))]
     use intel::init_global_shared_with_assembly;
 
     #[cfg(any(
         target_arch = "aarch64",
         target_arch = "arm",
-        target_arch = "x86",
+        all(target_arch = "x86", target_feature = "sse2"),
         target_arch = "x86_64",
     ))]
     {
@@ -47,5 +47,5 @@ pub(crate) fn features() -> Features {
 #[cfg(any(target_arch = "aarch64", target_arch = "arm"))]
 pub mod arm;
 
-#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+#[cfg(any(all(target_arch = "x86", target_feature = "sse2"), target_arch = "x86_64"))]
 pub mod intel;
diff --git a/src/cpu/intel.rs b/src/cpu/intel.rs
@@ -13,11 +13,11 @@
 // CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 
 #![cfg_attr(
-    not(any(target_arch = "x86", target_arch = "x86_64")),
+    not(any(all(target_arch = "x86", target_feature = "sse2"), target_arch = "x86_64")),
     allow(dead_code)
 )]
 
-#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+#[cfg(any(all(target_arch = "x86", target_feature = "sse2"), target_arch = "x86_64"))]
 mod abi_assumptions {
     // TOOD: Support targets that do not have SSE and SSE2 enabled, such as
     // x86_64-unknown-linux-none. See
@@ -29,7 +29,7 @@ mod abi_assumptions {
 
     #[cfg(target_arch = "x86_64")]
     const _ASSUMED_POINTER_SIZE: usize = 8;
-    #[cfg(target_arch = "x86")]
+    #[cfg(all(target_arch = "x86", target_feature = "sse2"))]
     const _ASSUMED_POINTER_SIZE: usize = 4;
     const _ASSUMED_USIZE_SIZE: () = assert!(core::mem::size_of::<usize>() == _ASSUMED_POINTER_SIZE);
     const _ASSUMED_REF_SIZE: () =
@@ -43,7 +43,7 @@ pub(crate) struct Feature {
     mask: u32,
 }
 
-#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+#[cfg(any(all(target_arch = "x86", target_feature = "sse2"), target_arch = "x86_64"))]
 pub(super) unsafe fn init_global_shared_with_assembly() {
     prefixed_extern! {
         fn OPENSSL_cpuid_setup();
@@ -57,15 +57,15 @@ impl Feature {
     #[allow(clippy::needless_return)]
     #[inline(always)]
     pub fn available(&self, _: super::Features) -> bool {
-        #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+        #[cfg(any(all(target_arch = "x86", target_feature = "sse2"), target_arch = "x86_64"))]
         {
             prefixed_extern! {
                 static mut OPENSSL_ia32cap_P: [u32; 4];
             }
             return self.mask == self.mask & unsafe { OPENSSL_ia32cap_P[self.word] };
         }
 
-        #[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
+        #[cfg(not(any(all(target_arch = "x86", target_feature = "sse2"), target_arch = "x86_64")))]
         {
             return false;
         }

diff --git a/src/prefixed.rs b/src/prefixed.rs
@@ -44,7 +44,7 @@ macro_rules! prefixed_extern {
 #[cfg(not(any(
     target_arch = "aarch64",
     target_arch = "arm",
-    target_arch = "x86",
+    all(target_arch = "x86", target_feature = "sse2"),
     target_arch = "x86_64"
 )))]
 macro_rules! prefixed_export {