Enable building and running on Windows ARM64

I agree to license my contributions to each file under the terms given at the top of each file I changed.

Co-authored-by: Marc-André Moreau <marcandre.moreau@gmail.com>

.github/workflows/ci.yml

@@ -149,6 +149,9 @@ jobs:
           toolchain: stable
           profile: minimal
 
+      - run: echo "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Tools\Llvm\x64\bin" >> $GITHUB_PATH
+        shell: bash
+
       - run: sh mk/package.sh
         shell: bash
 
@@ -167,6 +170,7 @@ jobs:
           - aarch64-apple-ios
           - aarch64-apple-darwin
           - aarch64-linux-android
+          - aarch64-pc-windows-msvc
           - aarch64-unknown-linux-gnu
           - aarch64-unknown-linux-musl
           - arm-unknown-linux-gnueabihf
@@ -234,6 +238,11 @@ jobs:
             # TODO: https://github.com/briansmith/ring/issues/486
             cargo_options: --no-run
 
+          - target: aarch64-pc-windows-msvc
+            host_os: windows-latest
+            # GitHub Actions doesn't have a way to run this target yet.
+            cargo_options: --no-run
+
           - target: aarch64-unknown-linux-gnu
             host_os: ubuntu-18.04
 
@@ -303,6 +312,12 @@ jobs:
           sudo xcode-select -s /Applications/Xcode_12.4.app &&
           sudo rm -Rf /Library/Developer/CommandLineTools/SDKs/*
 
+      - if: ${{ matrix.target == 'aarch64-pc-windows-msvc' }}
+        run: |
+          echo "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\VC\Tools\Llvm\x64\bin" >> $GITHUB_PATH
+          echo "RUSTFLAGS=-C target-feature=+crt-static" >> $GITHUB_ENV
+        shell: bash
+
       - if: ${{ !contains(matrix.host_os, 'windows') }}
         run: |
           mk/cargo.sh test -vv --all-targets --target=${{ matrix.target }} ${{ matrix.cargo_options }} ${{ matrix.features }} ${{ matrix.mode }}

BUILDING.md

@@ -33,9 +33,12 @@ apply when building from crates.io:
   primitives (32- and 64- bit Intel, and 32- and 64-bit ARM), Perl must be
   installed and in `$PATH`.
 
-* For Windows targets, `target/tools/windows/nasm/nasm[.exe]` is used as the
-  assembler. The version to use and how to download it is documented in
-  [.github/workflows/ci.yml](.github/workflows/ci.yml).
+* For Windows targets except ARM64, `target/tools/windows/nasm/nasm[.exe]`
+  is used as the assembler. The version to use and how to download it is
+  documented in [.github/workflows/ci.yml](.github/workflows/ci.yml).
+
+* For Windows ARM64 target, Clang is used as the C compiler and the assembler.
+  See below "Building for Windows ARM64" section.
 
 Cross Compiling
 ---------------
@@ -57,10 +60,12 @@ that the current beta and nightly releases work.
 
 On Windows, *ring* supports the x86_64-pc-windows-msvc and i686-pc-windows-msvc
 targets best. These targets require the “Visual C++ Build Tools
-2015” package or Visual Studio 2015 Update 3 or later to be installed. Patches
-to get it working on other variants, including in particular Visual Studio 2017
-([#338]), Windows ARM platforms, Windows Universal Platform, Windows XP (the
-v140_xp toolchain; [#339]), and the -gnu targets ([#330]) are welcome.
+2015” package or Visual Studio 2015 Update 3 or later to be installed.
+*ring* now also supports the aarch64-pc-windows-msvc target. For the detailed
+instructions please see the next section.
+Patches to get it working on other variants, including in particular Visual Studio 2017
+([#338]), Windows Universal Platform, Windows XP (the v140_xp toolchain; [#339]),
+and the -gnu targets ([#330]) are welcome.
 
 For other platforms, GCC 4.6 or later and Clang 3.5 or later are currently
 supported best. The build script passes options to the C/C++ compiler that are
@@ -78,6 +83,34 @@ defined with a value of at least 21 on 64-bit targets or 18 on 32-bit targets;
 e.g. export `CFLAGS=-D__ANDROID_API__=21`.
 
 
+Building for Windows ARM64
+--------------------------
+
+Windows ARM64 target requires the “Visual C++ Build Tools 2019” package or
+Visual Studio 2019 or later to be installed. “Desktop development with C++”
+workflow should be installed, as well as
+“MSVC v142 - VS 2019 C++ ARM64 build tools” component.
+
+To build *ring* for Windows ARM64, you will need to install Clang as it is used
+as the C compiler and the assembler for that platform. You can either use
+the version of Clang installed by Visual Studio, a standalone version from
+llvm.org, or a mingw64 version of Clang, for example, from [llvm-mingw
+project](https://github.com/mstorsjo/llvm-mingw).
+
+If you're buiding *ring* on an ARM64 device like Surface Pro X, please note
+that llvm.org and llvm-mingw have native ARM64 versions of Clang available.
+Also, if you're building *ring* on an ARM64 device, you might want to use
+`aarch64-pc-windows-msvc` Rustup toolchain, which can be installed using
+`rustup toolchain add aarch64-pc-windows-msvc`.
+
+When building on an ARM64 device, due to a bug in the Visual Studio installer,
+if you're using `rustc` version < 1.55 you would need to run `cargo build` /
+`cargo test` commands from x86_arm64 Developer Command Prompt. You can use
+`C:\Program Files (x86)\Microsoft Visual Studio\2019\<edition>\VC\Auxiliary\Build\vcvarsx86_arm64.bat`
+batch script to configure the environment. If you use `rustc` 1.55 beta or newer,
+you can run `cargo` commands without configuring the dev environment beforehand.
+
+
 Additional Features that are Useful for Development
 ---------------------------------------------------
 The `slow_tests` feature runs additional tests that are too slow to run during

Cargo.toml

@@ -162,7 +162,7 @@ name = "ring"
 [dependencies]
 untrusted = { version = "0.9" }
 
-[target.'cfg(any(target_arch = "x86",target_arch = "x86_64", all(any(target_arch = "aarch64", target_arch = "arm"), any(target_os = "android", target_os = "fuchsia", target_os = "linux"))))'.dependencies]
+[target.'cfg(any(target_arch = "x86",target_arch = "x86_64", all(any(target_arch = "aarch64", target_arch = "arm"), any(target_os = "android", target_os = "fuchsia", target_os = "linux", target_os = "windows"))))'.dependencies]
 spin = { version = "0.9.2", default-features = false, features = ["once"] }
 
 [target.'cfg(any(target_os = "android", target_os = "linux"))'.dependencies]
@@ -176,7 +176,7 @@ once_cell = { version = "1.5.2", default-features = false, features=["std"] }
 web-sys = { version = "0.3.37", default-features = false, features = ["Crypto", "Window"] }
 
 [target.'cfg(target_os = "windows")'.dependencies]
-winapi = { version = "0.3.8", default-features = false, features = ["ntsecapi", "wtypesbase"] }
+winapi = { version = "0.3.9", default-features = false, features = ["ntsecapi", "wtypesbase", "processthreadsapi"] }
 
 [target.'cfg(target_arch = "wasm32")'.dev-dependencies]
 wasm-bindgen-test = { version = "0.3.18", default-features = false }

build.rs

@@ -228,6 +228,13 @@ const ASM_TARGETS: &[AsmTarget] = &[
         asm_extension: "asm",
         preassemble: true,
     },
+    AsmTarget {
+        oss: &[WINDOWS],
+        arch: "aarch64",
+        perlasm_format: "win64",
+        asm_extension: "S",
+        preassemble: true,
+    },
 ];
 
 struct AsmTarget {
@@ -378,7 +385,7 @@ fn pregenerate_asm_main() {
             let os = WINDOWS;
 
             if !std::mem::replace(&mut generated_prefix_headers, true) {
-                generate_prefix_symbols_nasm(&pregenerated, &ring_core_prefix()).unwrap();
+                generate_prefix_symbols(&pregenerated, &ring_core_prefix()).unwrap();
             }
             let srcs = asm_srcs(perlasm_src_dsts);
 
@@ -390,10 +397,8 @@ fn pregenerate_asm_main() {
                 force_warnings_into_errors: true,
             };
 
-            let compiler = cc::Build::new().get_compiler();
-
             for src in srcs {
-                compile(&src, &target, &compiler, &pregenerated);
+                compile(&src, &target, &pregenerated);
             }
         }
     }
@@ -441,7 +446,7 @@ fn build_c_code(
         out_dir
     };
 
-    generate_prefix_symbols(target, out_dir, ring_core_prefix).unwrap();
+    generate_prefix_symbols(out_dir, ring_core_prefix).unwrap();
 
     let asm_srcs = if let Some(asm_target) = asm_target {
         let perlasm_src_dsts = perlasm_src_dsts(asm_dir, asm_target);
@@ -500,13 +505,11 @@ fn build_library(
     srcs: &[PathBuf],
     additional_srcs: &[PathBuf],
 ) {
-    let compiler = cc::Build::default().get_compiler();
-
     // Compile all the (dirty) source files into object files.
     let objs = additional_srcs
         .iter()
         .chain(srcs.iter())
-        .map(|f| compile(f, target, &compiler, out_dir))
+        .map(|f| compile(f, target, out_dir))
         .collect::<Vec<_>>();
 
     // Rebuild the library if necessary.
@@ -545,14 +548,14 @@ fn build_library(
     println!("cargo:rustc-link-lib=static={}", lib_name);
 }
 
-fn compile(p: &Path, target: &Target, compiler: &cc::Tool, out_dir: &Path) -> String {
+fn compile(p: &Path, target: &Target, out_dir: &Path) -> String {
     let ext = p.extension().unwrap().to_str().unwrap();
     if ext == "o" {
         p.to_str().expect("Invalid path").into()
     } else {
         let out_path = obj_path(out_dir, p);
         let cmd = if target.os != WINDOWS || ext != "asm" {
-            cc(p, ext, target, compiler, &out_path, out_dir)
+            cc(p, ext, target, &out_path, out_dir)
         } else {
             nasm(p, &target.arch, &out_path, out_dir)
         };
@@ -571,27 +574,28 @@ fn obj_path(out_dir: &Path, src: &Path) -> PathBuf {
     out_path
 }
 
-fn cc(
-    file: &Path,
-    ext: &str,
-    target: &Target,
-    compiler: &cc::Tool,
-    out_path: &Path,
-    include_dir: &Path,
-) -> Command {
+fn cc(file: &Path, ext: &str, target: &Target, out_path: &Path, include_dir: &Path) -> Command {
     let mut c = cc::Build::new();
+
+    // FIXME: On Windows AArch64 we currently must use Clang to compile C code
+    if target.os == WINDOWS && target.arch == AARCH64 && !c.get_compiler().is_like_clang() {
+        let _ = c.compiler("clang");
+    }
+
+    let compiler = c.get_compiler();
+
     let _ = c.include("include");
     let _ = c.include(include_dir);
     match ext {
         "c" => {
-            for f in c_flags(compiler) {
+            for f in c_flags(&compiler) {
                 let _ = c.flag(f);
             }
         }
         "S" => (),
         e => panic!("Unsupported file extension: {:?}", e),
     };
-    for f in cpp_flags(compiler) {
+    for f in cpp_flags(&compiler) {
         let _ = c.flag(f);
     }
     if target.os != "none"
@@ -843,24 +847,18 @@ fn ring_core_prefix() -> String {
 
 /// Creates the necessary header file for symbol renaming and returns the path of the
 /// generated include directory.
-fn generate_prefix_symbols(
-    target: &Target,
-    out_dir: &Path,
-    prefix: &str,
-) -> Result<(), std::io::Error> {
+fn generate_prefix_symbols(out_dir: &Path, prefix: &str) -> Result<(), std::io::Error> {
     generate_prefix_symbols_header(out_dir, "prefix_symbols.h", '#', None, prefix)?;
 
-    if target.os == "windows" {
-        let _ = generate_prefix_symbols_nasm(out_dir, prefix)?;
-    } else {
-        generate_prefix_symbols_header(
-            out_dir,
-            "prefix_symbols_asm.h",
-            '#',
-            Some("#if defined(__APPLE__)"),
-            prefix,
-        )?;
-    }
+    generate_prefix_symbols_nasm(out_dir, prefix)?;
+
+    generate_prefix_symbols_header(
+        out_dir,
+        "prefix_symbols_asm.h",
+        '#',
+        Some("#if defined(__APPLE__)"),
+        prefix,
+    )?;
 
     Ok(())
 }

crypto/crypto.c

@@ -15,6 +15,7 @@
 #include <ring-core/cpu.h>
 #include "internal.h"
 
+#if defined(OPENSSL_X86) || defined(OPENSSL_X86_64)
 // Our assembly does not use the GOT to reference symbols, which means
 // references to visible symbols will often require a TEXTREL. This is
 // undesirable, so all assembly-referenced symbols should be hidden. CPU
@@ -26,7 +27,6 @@
 #define HIDDEN __attribute__((visibility("hidden")))
 #endif
 
-#if defined(OPENSSL_X86) || defined(OPENSSL_X86_64)
 // This value must be explicitly initialised to zero in order to work around a
 // bug in libtool or the linker on OS X.
 //

crypto/fipsmodule/ec/ecp_nistz384.inl

@@ -252,6 +252,6 @@ void nistz384_point_mul(P384_POINT *r, const BN_ULONG p_scalar[P384_LIMBS],
   add_precomputed_w5(r, wvalue, table);
 }
 
-#if defined(__GNUC__)
+#if defined(__GNUC__) || defined(__clang__)
 #pragma GCC diagnostic pop
 #endif

mk/package.sh

@@ -10,5 +10,6 @@ if [[ $(git status --porcelain | wc -c) -ne 0 ]]; then
 fi
 
 cargo clean --target-dir=target/pregenerate_asm
-RING_PREGENERATE_ASM=1 cargo build --target-dir=target/pregenerate_asm
+RING_PREGENERATE_ASM=1 CC_AARCH64_PC_WINDOWS_MSVC=clang \
+  cargo build --target-dir=target/pregenerate_asm
 cargo package --allow-dirty

src/cpu.rs

@@ -31,7 +31,12 @@ pub(crate) fn features() -> Features {
         target_arch = "x86_64",
         all(
             any(target_arch = "aarch64", target_arch = "arm"),
-            any(target_os = "android", target_os = "fuchsia", target_os = "linux")
+            any(
+                target_os = "android",
+                target_os = "fuchsia",
+                target_os = "linux",
+                target_os = "windows"
+            )
         )
     ))]
     {
@@ -49,7 +54,12 @@ pub(crate) fn features() -> Features {
 
             #[cfg(all(
                 any(target_arch = "aarch64", target_arch = "arm"),
-                any(target_os = "android", target_os = "fuchsia", target_os = "linux")
+                any(
+                    target_os = "android",
+                    target_os = "fuchsia",
+                    target_os = "linux",
+                    target_os = "windows"
+                )
             ))]
             {
                 arm::setup();
@@ -162,6 +172,27 @@ pub(crate) mod arm {
         }
     }
 
+    #[cfg(all(target_os = "windows", target_arch = "aarch64"))]
+    pub fn setup() {
+        // We do not need to check for the presence of NEON, as Armv8-A always has it
+        let mut features = NEON.mask;
+
+        let result = unsafe {
+            winapi::um::processthreadsapi::IsProcessorFeaturePresent(
+                winapi::um::winnt::PF_ARM_V8_CRYPTO_INSTRUCTIONS_AVAILABLE,
+            )
+        };
+
+        if result != 0 {
+            // These are all covered by one call in Windows
+            features |= AES.mask;
+            features |= PMULL.mask;
+            features |= SHA256.mask;
+        }
+
+        unsafe { OPENSSL_armcap_P = features };
+    }
+
     macro_rules! features {
         {
             $(
@@ -237,7 +268,12 @@ pub(crate) mod arm {
             }
 
             #[cfg(all(
-                any(target_os = "android", target_os = "fuchsia", target_os = "linux"),
+                any(
+                    target_os = "android",
+                    target_os = "fuchsia",
+                    target_os = "linux",
+                    target_os = "windows"
+                ),
                 any(target_arch = "arm", target_arch = "aarch64")
             ))]
             {