Rollup merge of #86652 - nagisa:nagisa/non-leaf-fp, r=petrochenkov

Add support for leaf function frame pointer elimination

This PR adds ability for the target specifications to specify frame
pointer emission type that's not just “always” or “whatever cg decides”.

In particular there's a new mode that allows omission of the frame
pointer for leaf functions (those that don't call any other functions).

We then set this new mode for Aarch64-based Apple targets.

Fixes #86196

Author: GuillaumeGomez

compiler/rustc_codegen_llvm/src/attributes.rs

@@ -12,7 +12,7 @@ use rustc_middle::ty::{self, TyCtxt};
 use rustc_session::config::OptLevel;
 use rustc_session::Session;
 use rustc_target::spec::abi::Abi;
-use rustc_target::spec::{SanitizerSet, StackProbeType};
+use rustc_target::spec::{FramePointer, SanitizerSet, StackProbeType};
 
 use crate::attributes;
 use crate::llvm::AttributePlace::Function;
@@ -69,15 +69,25 @@ fn naked(val: &'ll Value, is_naked: bool) {
     Attribute::Naked.toggle_llfn(Function, val, is_naked);
 }
 
-pub fn set_frame_pointer_elimination(cx: &CodegenCx<'ll, '_>, llfn: &'ll Value) {
-    if cx.sess().must_not_eliminate_frame_pointers() {
-        llvm::AddFunctionAttrStringValue(
-            llfn,
-            llvm::AttributePlace::Function,
-            cstr!("frame-pointer"),
-            cstr!("all"),
-        );
+pub fn set_frame_pointer_type(cx: &CodegenCx<'ll, '_>, llfn: &'ll Value) {
+    let mut fp = cx.sess().target.frame_pointer;
+    // "mcount" function relies on stack pointer.
+    // See <https://sourceware.org/binutils/docs/gprof/Implementation.html>.
+    if cx.sess().instrument_mcount() || matches!(cx.sess().opts.cg.force_frame_pointers, Some(true))
+    {
+        fp = FramePointer::Always;
     }
+    let attr_value = match fp {
+        FramePointer::Always => cstr!("all"),
+        FramePointer::NonLeaf => cstr!("non-leaf"),
+        FramePointer::MayOmit => return,
+    };
+    llvm::AddFunctionAttrStringValue(
+        llfn,
+        llvm::AttributePlace::Function,
+        cstr!("frame-pointer"),
+        attr_value,
+    );
 }
 
 /// Tell LLVM what instrument function to insert.
@@ -254,7 +264,7 @@ pub fn from_fn_attrs(cx: &CodegenCx<'ll, 'tcx>, llfn: &'ll Value, instance: ty::
     }
 
     // FIXME: none of these three functions interact with source level attributes.
-    set_frame_pointer_elimination(cx, llfn);
+    set_frame_pointer_type(cx, llfn);
     set_instrument_function(cx, llfn);
     set_probestack(cx, llfn);
 

compiler/rustc_codegen_llvm/src/context.rs

@@ -410,8 +410,8 @@ impl MiscMethods<'tcx> for CodegenCx<'ll, 'tcx> {
         &self.used_statics
     }
 
-    fn set_frame_pointer_elimination(&self, llfn: &'ll Value) {
-        attributes::set_frame_pointer_elimination(self, llfn)
+    fn set_frame_pointer_type(&self, llfn: &'ll Value) {
+        attributes::set_frame_pointer_type(self, llfn)
     }
 
     fn apply_target_cpu_attr(&self, llfn: &'ll Value) {

compiler/rustc_codegen_llvm/src/intrinsic.rs

@@ -674,7 +674,7 @@ fn gen_fn<'ll, 'tcx>(
 ) -> &'ll Value {
     let fn_abi = FnAbi::of_fn_ptr(cx, rust_fn_sig, &[]);
     let llfn = cx.declare_fn(name, &fn_abi);
-    cx.set_frame_pointer_elimination(llfn);
+    cx.set_frame_pointer_type(llfn);
     cx.apply_target_cpu_attr(llfn);
     // FIXME(eddyb) find a nicer way to do this.
     unsafe { llvm::LLVMRustSetLinkage(llfn, llvm::Linkage::InternalLinkage) };

compiler/rustc_codegen_ssa/src/base.rs

@@ -406,7 +406,7 @@ pub fn maybe_create_entry_wrapper<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>>(
         };
 
         // `main` should respect same config for frame pointer elimination as rest of code
-        cx.set_frame_pointer_elimination(llfn);
+        cx.set_frame_pointer_type(llfn);
         cx.apply_target_cpu_attr(llfn);
 
         let llbb = Bx::append_block(&cx, llfn, "top");

compiler/rustc_codegen_ssa/src/traits/misc.rs

@@ -16,7 +16,7 @@ pub trait MiscMethods<'tcx>: BackendTypes {
     fn sess(&self) -> &Session;
     fn codegen_unit(&self) -> &'tcx CodegenUnit<'tcx>;
     fn used_statics(&self) -> &RefCell<Vec<Self::Value>>;
-    fn set_frame_pointer_elimination(&self, llfn: Self::Function);
+    fn set_frame_pointer_type(&self, llfn: Self::Function);
     fn apply_target_cpu_attr(&self, llfn: Self::Function);
     fn create_used_variable(&self);
     /// Declares the extern "C" main function for the entry point. Returns None if the symbol already exists.

compiler/rustc_session/src/session.rs

@@ -792,18 +792,6 @@ impl Session {
         !self.target.is_like_windows && !self.target.is_like_osx
     }
 
-    pub fn must_not_eliminate_frame_pointers(&self) -> bool {
-        // "mcount" function relies on stack pointer.
-        // See <https://sourceware.org/binutils/docs/gprof/Implementation.html>.
-        if self.instrument_mcount() {
-            true
-        } else if let Some(x) = self.opts.cg.force_frame_pointers {
-            x
-        } else {
-            !self.target.eliminate_frame_pointer
-        }
-    }
-
     pub fn must_emit_unwind_tables(&self) -> bool {
         // This is used to control the emission of the `uwtable` attribute on
         // LLVM functions.

compiler/rustc_target/src/spec/aarch64_apple_darwin.rs

@@ -1,4 +1,4 @@
-use crate::spec::{LinkerFlavor, SanitizerSet, Target, TargetOptions};
+use crate::spec::{FramePointer, LinkerFlavor, SanitizerSet, Target, TargetOptions};
 
 pub fn target() -> Target {
     let mut base = super::apple_base::opts("macos");
@@ -20,6 +20,10 @@ pub fn target() -> Target {
         pointer_width: 64,
         data_layout: "e-m:o-i64:64-i128:128-n32:64-S128".to_string(),
         arch: arch.to_string(),
-        options: TargetOptions { mcount: "\u{1}mcount".to_string(), ..base },
+        options: TargetOptions {
+            mcount: "\u{1}mcount".to_string(),
+            frame_pointer: FramePointer::NonLeaf,
+            ..base
+        },
     }
 }

compiler/rustc_target/src/spec/aarch64_apple_ios.rs

@@ -1,5 +1,5 @@
 use super::apple_sdk_base::{opts, Arch};
-use crate::spec::{Target, TargetOptions};
+use crate::spec::{FramePointer, Target, TargetOptions};
 
 pub fn target() -> Target {
     let base = opts("ios", Arch::Arm64);
@@ -13,6 +13,7 @@ pub fn target() -> Target {
             max_atomic_width: Some(128),
             unsupported_abis: super::arm_base::unsupported_abis(),
             forces_embed_bitcode: true,
+            frame_pointer: FramePointer::NonLeaf,
             // Taken from a clang build on Xcode 11.4.1.
             // These arguments are not actually invoked - they just have
             // to look right to pass App Store validation.

compiler/rustc_target/src/spec/aarch64_apple_ios_macabi.rs

@@ -1,5 +1,5 @@
 use super::apple_sdk_base::{opts, Arch};
-use crate::spec::{Target, TargetOptions};
+use crate::spec::{FramePointer, Target, TargetOptions};
 
 pub fn target() -> Target {
     let base = opts("ios", Arch::Arm64_macabi);
@@ -13,6 +13,7 @@ pub fn target() -> Target {
             max_atomic_width: Some(128),
             unsupported_abis: super::arm_base::unsupported_abis(),
             forces_embed_bitcode: true,
+            frame_pointer: FramePointer::NonLeaf,
             // Taken from a clang build on Xcode 11.4.1.
             // These arguments are not actually invoked - they just have
             // to look right to pass App Store validation.

compiler/rustc_target/src/spec/aarch64_apple_ios_sim.rs

@@ -1,5 +1,5 @@
 use super::apple_sdk_base::{opts, Arch};
-use crate::spec::{Target, TargetOptions};
+use crate::spec::{FramePointer, Target, TargetOptions};
 
 pub fn target() -> Target {
     let base = opts("ios", Arch::Arm64_sim);
@@ -21,6 +21,7 @@ pub fn target() -> Target {
             max_atomic_width: Some(128),
             unsupported_abis: super::arm_base::unsupported_abis(),
             forces_embed_bitcode: true,
+            frame_pointer: FramePointer::NonLeaf,
             // Taken from a clang build on Xcode 11.4.1.
             // These arguments are not actually invoked - they just have
             // to look right to pass App Store validation.

compiler/rustc_target/src/spec/aarch64_apple_tvos.rs

@@ -1,5 +1,5 @@
 use super::apple_sdk_base::{opts, Arch};
-use crate::spec::{Target, TargetOptions};
+use crate::spec::{FramePointer, Target, TargetOptions};
 
 pub fn target() -> Target {
     let base = opts("tvos", Arch::Arm64);
@@ -13,6 +13,7 @@ pub fn target() -> Target {
             max_atomic_width: Some(128),
             unsupported_abis: super::arm_base::unsupported_abis(),
             forces_embed_bitcode: true,
+            frame_pointer: FramePointer::NonLeaf,
             ..base
         },
     }

compiler/rustc_target/src/spec/apple_base.rs

@@ -1,6 +1,6 @@
 use std::env;
 
-use crate::spec::{SplitDebuginfo, TargetOptions};
+use crate::spec::{FramePointer, SplitDebuginfo, TargetOptions};
 
 pub fn opts(os: &str) -> TargetOptions {
     // ELF TLS is only available in macOS 10.7+. If you try to compile for 10.6
@@ -27,7 +27,7 @@ pub fn opts(os: &str) -> TargetOptions {
         families: vec!["unix".to_string()],
         is_like_osx: true,
         dwarf_version: Some(2),
-        eliminate_frame_pointer: false,
+        frame_pointer: FramePointer::Always,
         has_rpath: true,
         dll_suffix: ".dylib".to_string(),
         archive_format: "darwin".to_string(),

compiler/rustc_target/src/spec/freebsd_base.rs

@@ -1,4 +1,4 @@
-use crate::spec::{RelroLevel, TargetOptions};
+use crate::spec::{FramePointer, RelroLevel, TargetOptions};
 
 pub fn opts() -> TargetOptions {
     TargetOptions {
@@ -8,7 +8,7 @@ pub fn opts() -> TargetOptions {
         families: vec!["unix".to_string()],
         has_rpath: true,
         position_independent_executables: true,
-        eliminate_frame_pointer: false, // FIXME 43575
+        frame_pointer: FramePointer::Always, // FIXME 43575: should be MayOmit...
         relro_level: RelroLevel::Full,
         abi_return_struct_as_int: true,
         dwarf_version: Some(2),

compiler/rustc_target/src/spec/i686_apple_darwin.rs

@@ -1,4 +1,4 @@
-use crate::spec::{LinkerFlavor, StackProbeType, Target, TargetOptions};
+use crate::spec::{FramePointer, LinkerFlavor, StackProbeType, Target, TargetOptions};
 
 pub fn target() -> Target {
     let mut base = super::apple_base::opts("macos");
@@ -8,7 +8,7 @@ pub fn target() -> Target {
     base.link_env_remove.extend(super::apple_base::macos_link_env_remove());
     // don't use probe-stack=inline-asm until rust#83139 and rust#84667 are resolved
     base.stack_probes = StackProbeType::Call;
-    base.eliminate_frame_pointer = false;
+    base.frame_pointer = FramePointer::Always;
 
     // Clang automatically chooses a more specific target based on
     // MACOSX_DEPLOYMENT_TARGET.  To enable cross-language LTO to work

compiler/rustc_target/src/spec/i686_pc_windows_gnu.rs

@@ -1,12 +1,12 @@
-use crate::spec::{LinkerFlavor, LldFlavor, Target};
+use crate::spec::{FramePointer, LinkerFlavor, LldFlavor, Target};
 
 pub fn target() -> Target {
     let mut base = super::windows_gnu_base::opts();
     base.cpu = "pentium4".to_string();
     base.pre_link_args
         .insert(LinkerFlavor::Lld(LldFlavor::Ld), vec!["-m".to_string(), "i386pe".to_string()]);
     base.max_atomic_width = Some(64);
-    base.eliminate_frame_pointer = false; // Required for backtraces
+    base.frame_pointer = FramePointer::Always; // Required for backtraces
     base.linker = Some("i686-w64-mingw32-gcc".to_string());
 
     // Mark all dynamic libraries and executables as compatible with the larger 4GiB address

compiler/rustc_target/src/spec/i686_unknown_linux_musl.rs

@@ -1,4 +1,4 @@
-use crate::spec::{LinkerFlavor, StackProbeType, Target};
+use crate::spec::{FramePointer, LinkerFlavor, StackProbeType, Target};
 
 pub fn target() -> Target {
     let mut base = super::linux_musl_base::opts();
@@ -21,7 +21,7 @@ pub fn target() -> Target {
     //
     // This may or may not be related to this bug:
     // https://llvm.org/bugs/show_bug.cgi?id=30879
-    base.eliminate_frame_pointer = false;
+    base.frame_pointer = FramePointer::Always;
 
     Target {
         llvm_target: "i686-unknown-linux-musl".to_string(),

compiler/rustc_target/src/spec/i686_uwp_windows_gnu.rs

@@ -1,12 +1,12 @@
-use crate::spec::{LinkerFlavor, LldFlavor, Target};
+use crate::spec::{FramePointer, LinkerFlavor, LldFlavor, Target};
 
 pub fn target() -> Target {
     let mut base = super::windows_uwp_gnu_base::opts();
     base.cpu = "pentium4".to_string();
     base.pre_link_args
         .insert(LinkerFlavor::Lld(LldFlavor::Ld), vec!["-m".to_string(), "i386pe".to_string()]);
     base.max_atomic_width = Some(64);
-    base.eliminate_frame_pointer = false; // Required for backtraces
+    base.frame_pointer = FramePointer::Always; // Required for backtraces
 
     // Mark all dynamic libraries and executables as compatible with the larger 4GiB address
     // space available to x86 Windows binaries on x86_64.

compiler/rustc_target/src/spec/illumos_base.rs

@@ -1,4 +1,4 @@
-use crate::spec::{LinkArgs, LinkerFlavor, TargetOptions};
+use crate::spec::{FramePointer, LinkArgs, LinkerFlavor, TargetOptions};
 use std::default::Default;
 
 pub fn opts() -> TargetOptions {
@@ -35,7 +35,7 @@ pub fn opts() -> TargetOptions {
         is_like_solaris: true,
         linker_is_gnu: false,
         limit_rdylib_exports: false, // Linker doesn't support this
-        eliminate_frame_pointer: false,
+        frame_pointer: FramePointer::Always,
         eh_frame_header: false,
         late_link_args,
 

compiler/rustc_target/src/spec/linux_kernel_base.rs

@@ -1,4 +1,5 @@
-use crate::spec::{PanicStrategy, RelocModel, RelroLevel, StackProbeType, TargetOptions};
+use crate::spec::TargetOptions;
+use crate::spec::{FramePointer, PanicStrategy, RelocModel, RelroLevel, StackProbeType};
 
 pub fn opts() -> TargetOptions {
     TargetOptions {
@@ -7,7 +8,7 @@ pub fn opts() -> TargetOptions {
         panic_strategy: PanicStrategy::Abort,
         // don't use probe-stack=inline-asm until rust#83139 and rust#84667 are resolved
         stack_probes: StackProbeType::Call,
-        eliminate_frame_pointer: false,
+        frame_pointer: FramePointer::Always,
         position_independent_executables: true,
         needs_plt: true,
         relro_level: RelroLevel::Full,