Support for disabling the PLT on ELF targets

Disable the PLT where possible to improve performance
for indirect calls into shared libraries.

This optimization is enabled by default where possible.

- Add the `NonLazyBind` attribute to `rustllvm`:
  This attribute informs LLVM to skip PLT calls in codegen.

- Disable PLT unconditionally:
  Apply the `NonLazyBind` attribute on every function.

- Only enable no-plt when full relro is enabled:
  Ensures we only enable it when we have linker support.

- Add `-Z plt` as a compiler option

src/librustc/session/config.rs

@@ -1387,6 +1387,10 @@ options! {DebuggingOptions, DebuggingSetter, basic_debugging_options,
           "output a json file with profiler results"),
     emit_stack_sizes: bool = (false, parse_bool, [UNTRACKED],
           "emits a section containing stack size metadata"),
+    plt: Option<bool> = (None, parse_opt_bool, [TRACKED],
+          "whether to use the PLT when calling into shared libraries;
+          only has effect for PIC code on systems with ELF binaries
+          (default: PLT is disabled if full relro is enabled)"),
 }
 
 pub fn default_lib_output() -> CrateType {

src/librustc/session/mod.rs

@@ -40,8 +40,7 @@ use syntax::parse::{self, ParseSess};
 use syntax_pos::{MultiSpan, Span};
 use util::profiling::SelfProfiler;
 
-use rustc_target::spec::PanicStrategy;
-use rustc_target::spec::{Target, TargetTriple};
+use rustc_target::spec::{PanicStrategy, RelroLevel, Target, TargetTriple};
 use rustc_data_structures::flock;
 use jobserver::Client;
 
@@ -984,6 +983,27 @@ impl Session {
     pub fn edition(&self) -> Edition {
         self.opts.edition
     }
+
+    /// True if we cannot skip the PLT for shared library calls.
+    pub fn needs_plt(&self) -> bool {
+        // Check if the current target usually needs PLT to be enabled.
+        // The user can use the command line flag to override it.
+        let needs_plt = self.target.target.options.needs_plt;
+
+        let dbg_opts = &self.opts.debugging_opts;
+
+        let relro_level = dbg_opts.relro_level
+            .unwrap_or(self.target.target.options.relro_level);
+
+        // Only enable this optimization by default if full relro is also enabled.
+        // In this case, lazy binding was already unavailable, so nothing is lost.
+        // This also ensures `-Wl,-z,now` is supported by the linker.
+        let full_relro = RelroLevel::Full == relro_level;
+
+        // If user didn't explicitly forced us to use / skip the PLT,
+        // then try to skip it where possible.
+        dbg_opts.plt.unwrap_or(needs_plt || !full_relro)
+    }
 }
 
 pub fn build_session(

src/librustc_codegen_llvm/attributes.rs

@@ -138,6 +138,15 @@ pub fn apply_target_cpu_attr(cx: &CodegenCx<'ll, '_>, llfn: &'ll Value) {
             target_cpu.as_c_str());
 }
 
+/// Sets the `NonLazyBind` LLVM attribute on a given function,
+/// assuming the codegen options allow skipping the PLT.
+pub fn non_lazy_bind(sess: &Session, llfn: &'ll Value) {
+    // Don't generate calls through PLT if it's not necessary
+    if !sess.needs_plt() {
+        Attribute::NonLazyBind.apply_llfn(Function, llfn);
+    }
+}
+
 /// Composite function which sets LLVM attributes for function depending on its AST (#[attribute])
 /// attributes.
 pub fn from_fn_attrs(

src/librustc_codegen_llvm/context.rs

@@ -208,6 +208,13 @@ pub unsafe fn create_module(
         llvm::LLVMRustSetModulePIELevel(llmod);
     }
 
+    // If skipping the PLT is enabled, we need to add some module metadata
+    // to ensure intrinsic calls don't use it.
+    if !sess.needs_plt() {
+        let avoid_plt = "RtLibUseGOT\0".as_ptr() as *const _;
+        llvm::LLVMRustAddModuleFlag(llmod, avoid_plt, 1);
+    }
+
     llmod
 }
 

src/librustc_codegen_llvm/declare.rs

@@ -104,6 +104,8 @@ fn declare_raw_fn(
         attributes::unwind(llfn, false);
     }
 
+    attributes::non_lazy_bind(cx.sess(), llfn);
+
     llfn
 }
 

src/librustc_codegen_llvm/llvm/ffi.rs

@@ -122,6 +122,7 @@ pub enum Attribute {
     SanitizeThread  = 20,
     SanitizeAddress = 21,
     SanitizeMemory  = 22,
+    NonLazyBind     = 23,
 }
 
 /// LLVMIntPredicate

src/librustc_target/spec/mod.rs

@@ -576,6 +576,9 @@ pub struct TargetOptions {
     /// the functions in the executable are not randomized and can be used
     /// during an exploit of a vulnerability in any code.
     pub position_independent_executables: bool,
+    /// Determines if the target always requires using the PLT for indirect
+    /// library calls or not. This controls the default value of the `-Z plt` flag.
+    pub needs_plt: bool,
     /// Either partial, full, or off. Full RELRO makes the dynamic linker
     /// resolve all symbols at startup and marks the GOT read-only before
     /// starting the program, preventing overwriting the GOT.
@@ -720,6 +723,7 @@ impl Default for TargetOptions {
             has_rpath: false,
             no_default_libraries: true,
             position_independent_executables: false,
+            needs_plt: false,
             relro_level: RelroLevel::None,
             pre_link_objects_exe: Vec::new(),
             pre_link_objects_exe_crt: Vec::new(),
@@ -1009,6 +1013,7 @@ impl Target {
         key!(has_rpath, bool);
         key!(no_default_libraries, bool);
         key!(position_independent_executables, bool);
+        key!(needs_plt, bool);
         try!(key!(relro_level, RelroLevel));
         key!(archive_format);
         key!(allow_asm, bool);
@@ -1217,6 +1222,7 @@ impl ToJson for Target {
         target_option_val!(has_rpath);
         target_option_val!(no_default_libraries);
         target_option_val!(position_independent_executables);
+        target_option_val!(needs_plt);
         target_option_val!(relro_level);
         target_option_val!(archive_format);
         target_option_val!(allow_asm);

src/librustc_target/spec/x86_64_unknown_linux_gnux32.rs

@@ -17,6 +17,9 @@ pub fn target() -> TargetResult {
     base.pre_link_args.get_mut(&LinkerFlavor::Gcc).unwrap().push("-mx32".to_string());
     base.stack_probes = true;
     base.has_elf_tls = false;
+    // BUG(GabrielMajeri): disabling the PLT on x86_64 Linux with x32 ABI
+    // breaks code gen. See LLVM bug 36743
+    base.needs_plt = true;
 
     Ok(Target {
         llvm_target: "x86_64-unknown-linux-gnux32".to_string(),

src/rustllvm/RustWrapper.cpp

@@ -178,6 +178,8 @@ static Attribute::AttrKind fromRust(LLVMRustAttribute Kind) {
     return Attribute::SanitizeAddress;
   case SanitizeMemory:
     return Attribute::SanitizeMemory;
+  case NonLazyBind:
+    return Attribute::NonLazyBind;
   }
   report_fatal_error("bad AttributeKind");
 }

src/rustllvm/rustllvm.h

@@ -97,6 +97,7 @@ enum LLVMRustAttribute {
   SanitizeThread = 20,
   SanitizeAddress = 21,
   SanitizeMemory = 22,
+  NonLazyBind = 23,
 };
 
 typedef struct OpaqueRustString *RustStringRef;

src/test/codegen/naked-functions.rs

@@ -15,7 +15,7 @@
 #![crate_type = "lib"]
 #![feature(naked_functions)]
 
-// CHECK: Function Attrs: naked uwtable
+// CHECK: Function Attrs: naked
 // CHECK-NEXT: define void @naked_empty()
 #[no_mangle]
 #[naked]
@@ -24,7 +24,7 @@ pub fn naked_empty() {
     // CHECK-NEXT: ret void
 }
 
-// CHECK: Function Attrs: naked uwtable
+// CHECK: Function Attrs: naked
 #[no_mangle]
 #[naked]
 // CHECK-NEXT: define void @naked_with_args(i{{[0-9]+}})
@@ -35,7 +35,7 @@ pub fn naked_with_args(a: isize) {
     // CHECK: ret void
 }
 
-// CHECK: Function Attrs: naked uwtable
+// CHECK: Function Attrs: naked
 // CHECK-NEXT: define i{{[0-9]+}} @naked_with_return()
 #[no_mangle]
 #[naked]
@@ -45,7 +45,7 @@ pub fn naked_with_return() -> isize {
     0
 }
 
-// CHECK: Function Attrs: naked uwtable
+// CHECK: Function Attrs: naked
 // CHECK-NEXT: define i{{[0-9]+}} @naked_with_args_and_return(i{{[0-9]+}})
 #[no_mangle]
 #[naked]
@@ -57,7 +57,7 @@ pub fn naked_with_args_and_return(a: isize) -> isize {
     a
 }
 
-// CHECK: Function Attrs: naked uwtable
+// CHECK: Function Attrs: naked
 // CHECK-NEXT: define void @naked_recursive()
 #[no_mangle]
 #[naked]

src/test/codegen/no-plt.rs

@@ -0,0 +1,27 @@
+// Copyright 2018 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// compile-flags: -C relocation-model=pic -Z plt=no
+
+#![crate_type = "lib"]
+
+// We need a function which is normally called through the PLT.
+extern "C" {
+    // CHECK: Function Attrs: nounwind nonlazybind
+    fn getenv(name: *const u8) -> *mut u8;
+}
+
+// Ensure the function gets referenced.
+pub unsafe fn call_through_plt() -> *mut u8 {
+    getenv(b"\0".as_ptr())
+}
+
+// Ensure intrinsics also skip the PLT
+// CHECK: !"RtLibUseGOT"