DXR is a code search and navigation tool aimed at making sense of large projects. It supports full-text and regex searches as well as structural queries.

Git (4fb54ed484)

VCS Links

Line Code
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504
use crate::base;
use crate::common::CodegenCx;
use crate::debuginfo;
use crate::llvm::{self, True};
use crate::type_::Type;
use crate::type_of::LayoutLlvmExt;
use crate::value::Value;
use libc::c_uint;
use log::debug;
use rustc_codegen_ssa::traits::*;
use rustc_hir as hir;
use rustc_hir::def_id::DefId;
use rustc_hir::Node;
use rustc_middle::middle::codegen_fn_attrs::{CodegenFnAttrFlags, CodegenFnAttrs};
use rustc_middle::mir::interpret::{
    read_target_uint, Allocation, ConstValue, ErrorHandled, Pointer,
};
use rustc_middle::mir::mono::MonoItem;
use rustc_middle::ty::{self, Instance, Ty};
use rustc_middle::{bug, span_bug};
use rustc_span::symbol::{sym, Symbol};
use rustc_span::Span;
use rustc_target::abi::{Align, HasDataLayout, LayoutOf, Primitive, Scalar, Size};

use std::ffi::CStr;

pub fn const_alloc_to_llvm(cx: &CodegenCx<'ll, '_>, alloc: &Allocation) -> &'ll Value {
    let mut llvals = Vec::with_capacity(alloc.relocations().len() + 1);
    let dl = cx.data_layout();
    let pointer_size = dl.pointer_size.bytes() as usize;

    let mut next_offset = 0;
    for &(offset, ((), alloc_id)) in alloc.relocations().iter() {
        let offset = offset.bytes();
        assert_eq!(offset as usize as u64, offset);
        let offset = offset as usize;
        if offset > next_offset {
            // This `inspect` is okay since we have checked that it is not within a relocation, it
            // is within the bounds of the allocation, and it doesn't affect interpreter execution
            // (we inspect the result after interpreter execution). Any undef byte is replaced with
            // some arbitrary byte value.
            //
            // FIXME: relay undef bytes to codegen as undef const bytes
            let bytes = alloc.inspect_with_undef_and_ptr_outside_interpreter(next_offset..offset);
            llvals.push(cx.const_bytes(bytes));
        }
        let ptr_offset = read_target_uint(
            dl.endian,
            // This `inspect` is okay since it is within the bounds of the allocation, it doesn't
            // affect interpreter execution (we inspect the result after interpreter execution),
            // and we properly interpret the relocation as a relocation pointer offset.
            alloc.inspect_with_undef_and_ptr_outside_interpreter(offset..(offset + pointer_size)),
        )
        .expect("const_alloc_to_llvm: could not read relocation pointer")
            as u64;
        llvals.push(cx.scalar_to_backend(
            Pointer::new(alloc_id, Size::from_bytes(ptr_offset)).into(),
            &Scalar { value: Primitive::Pointer, valid_range: 0..=!0 },
            cx.type_i8p(),
        ));
        next_offset = offset + pointer_size;
    }
    if alloc.len() >= next_offset {
        let range = next_offset..alloc.len();
        // This `inspect` is okay since we have check that it is after all relocations, it is
        // within the bounds of the allocation, and it doesn't affect interpreter execution (we
        // inspect the result after interpreter execution). Any undef byte is replaced with some
        // arbitrary byte value.
        //
        // FIXME: relay undef bytes to codegen as undef const bytes
        let bytes = alloc.inspect_with_undef_and_ptr_outside_interpreter(range);
        llvals.push(cx.const_bytes(bytes));
    }

    cx.const_struct(&llvals, true)
}

pub fn codegen_static_initializer(
    cx: &CodegenCx<'ll, 'tcx>,
    def_id: DefId,
) -> Result<(&'ll Value, &'tcx Allocation), ErrorHandled> {
    let alloc = match cx.tcx.const_eval_poly(def_id)? {
        ConstValue::ByRef { alloc, offset } if offset.bytes() == 0 => alloc,
        val => bug!("static const eval returned {:#?}", val),
    };
    Ok((const_alloc_to_llvm(cx, alloc), alloc))
}

fn set_global_alignment(cx: &CodegenCx<'ll, '_>, gv: &'ll Value, mut align: Align) {
    // The target may require greater alignment for globals than the type does.
    // Note: GCC and Clang also allow `__attribute__((aligned))` on variables,
    // which can force it to be smaller.  Rust doesn't support this yet.
    if let Some(min) = cx.sess().target.target.options.min_global_align {
        match Align::from_bits(min) {
            Ok(min) => align = align.max(min),
            Err(err) => {
                cx.sess().err(&format!("invalid minimum global alignment: {}", err));
            }
        }
    }
    unsafe {
        llvm::LLVMSetAlignment(gv, align.bytes() as u32);
    }
}

fn check_and_apply_linkage(
    cx: &CodegenCx<'ll, 'tcx>,
    attrs: &CodegenFnAttrs,
    ty: Ty<'tcx>,
    sym: Symbol,
    span: Span,
) -> &'ll Value {
    let llty = cx.layout_of(ty).llvm_type(cx);
    let sym = sym.as_str();
    if let Some(linkage) = attrs.linkage {
        debug!("get_static: sym={} linkage={:?}", sym, linkage);

        // If this is a static with a linkage specified, then we need to handle
        // it a little specially. The typesystem prevents things like &T and
        // extern "C" fn() from being non-null, so we can't just declare a
        // static and call it a day. Some linkages (like weak) will make it such
        // that the static actually has a null value.
        let llty2 = if let ty::RawPtr(ref mt) = ty.kind {
            cx.layout_of(mt.ty).llvm_type(cx)
        } else {
            cx.sess().span_fatal(
                span,
                "must have type `*const T` or `*mut T` due to `#[linkage]` attribute",
            )
        };
        unsafe {
            // Declare a symbol `foo` with the desired linkage.
            let g1 = cx.declare_global(&sym, llty2);
            llvm::LLVMRustSetLinkage(g1, base::linkage_to_llvm(linkage));

            // Declare an internal global `extern_with_linkage_foo` which
            // is initialized with the address of `foo`.  If `foo` is
            // discarded during linking (for example, if `foo` has weak
            // linkage and there are no definitions), then
            // `extern_with_linkage_foo` will instead be initialized to
            // zero.
            let mut real_name = "_rust_extern_with_linkage_".to_string();
            real_name.push_str(&sym);
            let g2 = cx.define_global(&real_name, llty).unwrap_or_else(|| {
                cx.sess().span_fatal(span, &format!("symbol `{}` is already defined", &sym))
            });
            llvm::LLVMRustSetLinkage(g2, llvm::Linkage::InternalLinkage);
            llvm::LLVMSetInitializer(g2, g1);
            g2
        }
    } else {
        // Generate an external declaration.
        // FIXME(nagisa): investigate whether it can be changed into define_global
        cx.declare_global(&sym, llty)
    }
}

pub fn ptrcast(val: &'ll Value, ty: &'ll Type) -> &'ll Value {
    unsafe { llvm::LLVMConstPointerCast(val, ty) }
}

impl CodegenCx<'ll, 'tcx> {
    crate fn const_bitcast(&self, val: &'ll Value, ty: &'ll Type) -> &'ll Value {
        unsafe { llvm::LLVMConstBitCast(val, ty) }
    }

    crate fn static_addr_of_mut(
        &self,
        cv: &'ll Value,
        align: Align,
        kind: Option<&str>,
    ) -> &'ll Value {
        unsafe {
            let gv = match kind {
                Some(kind) if !self.tcx.sess.fewer_names() => {
                    let name = self.generate_local_symbol_name(kind);
                    let gv = self.define_global(&name[..], self.val_ty(cv)).unwrap_or_else(|| {
                        bug!("symbol `{}` is already defined", name);
                    });
                    llvm::LLVMRustSetLinkage(gv, llvm::Linkage::PrivateLinkage);
                    gv
                }
                _ => self.define_private_global(self.val_ty(cv)),
            };
            llvm::LLVMSetInitializer(gv, cv);
            set_global_alignment(&self, gv, align);
            llvm::SetUnnamedAddress(gv, llvm::UnnamedAddr::Global);
            gv
        }
    }

    crate fn get_static(&self, def_id: DefId) -> &'ll Value {
        let instance = Instance::mono(self.tcx, def_id);
        if let Some(&g) = self.instances.borrow().get(&instance) {
            return g;
        }

        let defined_in_current_codegen_unit =
            self.codegen_unit.items().contains_key(&MonoItem::Static(def_id));
        assert!(
            !defined_in_current_codegen_unit,
            "consts::get_static() should always hit the cache for \
                 statics defined in the same CGU, but did not for `{:?}`",
            def_id
        );

        let ty = instance.monomorphic_ty(self.tcx);
        let sym = self.tcx.symbol_name(instance).name;

        debug!("get_static: sym={} instance={:?}", sym, instance);

        let g = if let Some(def_id) = def_id.as_local() {
            let id = self.tcx.hir().as_local_hir_id(def_id);
            let llty = self.layout_of(ty).llvm_type(self);
            // FIXME: refactor this to work without accessing the HIR
            let (g, attrs) = match self.tcx.hir().get(id) {
                Node::Item(&hir::Item { attrs, span, kind: hir::ItemKind::Static(..), .. }) => {
                    let sym_str = sym.as_str();
                    if let Some(g) = self.get_declared_value(&sym_str) {
                        if self.val_ty(g) != self.type_ptr_to(llty) {
                            span_bug!(span, "Conflicting types for static");
                        }
                    }

                    let g = self.declare_global(&sym_str, llty);

                    if !self.tcx.is_reachable_non_generic(def_id) {
                        unsafe {
                            llvm::LLVMRustSetVisibility(g, llvm::Visibility::Hidden);
                        }
                    }

                    (g, attrs)
                }

                Node::ForeignItem(&hir::ForeignItem {
                    ref attrs,
                    span,
                    kind: hir::ForeignItemKind::Static(..),
                    ..
                }) => {
                    let fn_attrs = self.tcx.codegen_fn_attrs(def_id);
                    (check_and_apply_linkage(&self, &fn_attrs, ty, sym, span), &**attrs)
                }

                item => bug!("get_static: expected static, found {:?}", item),
            };

            debug!("get_static: sym={} attrs={:?}", sym, attrs);

            for attr in attrs {
                if attr.check_name(sym::thread_local) {
                    llvm::set_thread_local_mode(g, self.tls_model);
                }
            }

            g
        } else {
            // FIXME(nagisa): perhaps the map of externs could be offloaded to llvm somehow?
            debug!("get_static: sym={} item_attr={:?}", sym, self.tcx.item_attrs(def_id));

            let attrs = self.tcx.codegen_fn_attrs(def_id);
            let span = self.tcx.def_span(def_id);
            let g = check_and_apply_linkage(&self, &attrs, ty, sym, span);

            // Thread-local statics in some other crate need to *always* be linked
            // against in a thread-local fashion, so we need to be sure to apply the
            // thread-local attribute locally if it was present remotely. If we
            // don't do this then linker errors can be generated where the linker
            // complains that one object files has a thread local version of the
            // symbol and another one doesn't.
            if attrs.flags.contains(CodegenFnAttrFlags::THREAD_LOCAL) {
                llvm::set_thread_local_mode(g, self.tls_model);
            }

            let needs_dll_storage_attr = self.use_dll_storage_attrs && !self.tcx.is_foreign_item(def_id) &&
                // ThinLTO can't handle this workaround in all cases, so we don't
                // emit the attrs. Instead we make them unnecessary by disallowing
                // dynamic linking when linker plugin based LTO is enabled.
                !self.tcx.sess.opts.cg.linker_plugin_lto.enabled();

            // If this assertion triggers, there's something wrong with commandline
            // argument validation.
            debug_assert!(
                !(self.tcx.sess.opts.cg.linker_plugin_lto.enabled()
                    && self.tcx.sess.target.target.options.is_like_msvc
                    && self.tcx.sess.opts.cg.prefer_dynamic)
            );

            if needs_dll_storage_attr {
                // This item is external but not foreign, i.e., it originates from an external Rust
                // crate. Since we don't know whether this crate will be linked dynamically or
                // statically in the final application, we always mark such symbols as 'dllimport'.
                // If final linkage happens to be static, we rely on compiler-emitted __imp_ stubs
                // to make things work.
                //
                // However, in some scenarios we defer emission of statics to downstream
                // crates, so there are cases where a static with an upstream DefId
                // is actually present in the current crate. We can find out via the
                // is_codegened_item query.
                if !self.tcx.is_codegened_item(def_id) {
                    unsafe {
                        llvm::LLVMSetDLLStorageClass(g, llvm::DLLStorageClass::DllImport);
                    }
                }
            }
            g
        };

        if self.use_dll_storage_attrs && self.tcx.is_dllimport_foreign_item(def_id) {
            // For foreign (native) libs we know the exact storage type to use.
            unsafe {
                llvm::LLVMSetDLLStorageClass(g, llvm::DLLStorageClass::DllImport);
            }
        }

        self.instances.borrow_mut().insert(instance, g);
        g
    }
}

impl StaticMethods for CodegenCx<'ll, 'tcx> {
    fn static_addr_of(&self, cv: &'ll Value, align: Align, kind: Option<&str>) -> &'ll Value {
        if let Some(&gv) = self.const_globals.borrow().get(&cv) {
            unsafe {
                // Upgrade the alignment in cases where the same constant is used with different
                // alignment requirements
                let llalign = align.bytes() as u32;
                if llalign > llvm::LLVMGetAlignment(gv) {
                    llvm::LLVMSetAlignment(gv, llalign);
                }
            }
            return gv;
        }
        let gv = self.static_addr_of_mut(cv, align, kind);
        unsafe {
            llvm::LLVMSetGlobalConstant(gv, True);
        }
        self.const_globals.borrow_mut().insert(cv, gv);
        gv
    }

    fn codegen_static(&self, def_id: DefId, is_mutable: bool) {
        unsafe {
            let attrs = self.tcx.codegen_fn_attrs(def_id);

            let (v, alloc) = match codegen_static_initializer(&self, def_id) {
                Ok(v) => v,
                // Error has already been reported
                Err(_) => return,
            };

            let g = self.get_static(def_id);

            // boolean SSA values are i1, but they have to be stored in i8 slots,
            // otherwise some LLVM optimization passes don't work as expected
            let mut val_llty = self.val_ty(v);
            let v = if val_llty == self.type_i1() {
                val_llty = self.type_i8();
                llvm::LLVMConstZExt(v, val_llty)
            } else {
                v
            };

            let instance = Instance::mono(self.tcx, def_id);
            let ty = instance.monomorphic_ty(self.tcx);
            let llty = self.layout_of(ty).llvm_type(self);
            let g = if val_llty == llty {
                g
            } else {
                // If we created the global with the wrong type,
                // correct the type.
                let name = llvm::get_value_name(g).to_vec();
                llvm::set_value_name(g, b"");

                let linkage = llvm::LLVMRustGetLinkage(g);
                let visibility = llvm::LLVMRustGetVisibility(g);

                let new_g = llvm::LLVMRustGetOrInsertGlobal(
                    self.llmod,
                    name.as_ptr().cast(),
                    name.len(),
                    val_llty,
                );

                llvm::LLVMRustSetLinkage(new_g, linkage);
                llvm::LLVMRustSetVisibility(new_g, visibility);

                // To avoid breaking any invariants, we leave around the old
                // global for the moment; we'll replace all references to it
                // with the new global later. (See base::codegen_backend.)
                self.statics_to_rauw.borrow_mut().push((g, new_g));
                new_g
            };
            set_global_alignment(&self, g, self.align_of(ty));
            llvm::LLVMSetInitializer(g, v);

            // As an optimization, all shared statics which do not have interior
            // mutability are placed into read-only memory.
            if !is_mutable {
                if self.type_is_freeze(ty) {
                    llvm::LLVMSetGlobalConstant(g, llvm::True);
                }
            }

            debuginfo::create_global_var_metadata(&self, def_id, g);

            if attrs.flags.contains(CodegenFnAttrFlags::THREAD_LOCAL) {
                llvm::set_thread_local_mode(g, self.tls_model);

                // Do not allow LLVM to change the alignment of a TLS on macOS.
                //
                // By default a global's alignment can be freely increased.
                // This allows LLVM to generate more performant instructions
                // e.g., using load-aligned into a SIMD register.
                //
                // However, on macOS 10.10 or below, the dynamic linker does not
                // respect any alignment given on the TLS (radar 24221680).
                // This will violate the alignment assumption, and causing segfault at runtime.
                //
                // This bug is very easy to trigger. In `println!` and `panic!`,
                // the `LOCAL_STDOUT`/`LOCAL_STDERR` handles are stored in a TLS,
                // which the values would be `mem::replace`d on initialization.
                // The implementation of `mem::replace` will use SIMD
                // whenever the size is 32 bytes or higher. LLVM notices SIMD is used
                // and tries to align `LOCAL_STDOUT`/`LOCAL_STDERR` to a 32-byte boundary,
                // which macOS's dyld disregarded and causing crashes
                // (see issues #51794, #51758, #50867, #48866 and #44056).
                //
                // To workaround the bug, we trick LLVM into not increasing
                // the global's alignment by explicitly assigning a section to it
                // (equivalent to automatically generating a `#[link_section]` attribute).
                // See the comment in the `GlobalValue::canIncreaseAlignment()` function
                // of `lib/IR/Globals.cpp` for why this works.
                //
                // When the alignment is not increased, the optimized `mem::replace`
                // will use load-unaligned instructions instead, and thus avoiding the crash.
                //
                // We could remove this hack whenever we decide to drop macOS 10.10 support.
                if self.tcx.sess.target.target.options.is_like_osx {
                    // The `inspect` method is okay here because we checked relocations, and
                    // because we are doing this access to inspect the final interpreter state
                    // (not as part of the interpreter execution).
                    //
                    // FIXME: This check requires that the (arbitrary) value of undefined bytes
                    // happens to be zero. Instead, we should only check the value of defined bytes
                    // and set all undefined bytes to zero if this allocation is headed for the
                    // BSS.
                    let all_bytes_are_zero = alloc.relocations().is_empty()
                        && alloc
                            .inspect_with_undef_and_ptr_outside_interpreter(0..alloc.len())
                            .iter()
                            .all(|&byte| byte == 0);

                    let sect_name = if all_bytes_are_zero {
                        CStr::from_bytes_with_nul_unchecked(b"__DATA,__thread_bss\0")
                    } else {
                        CStr::from_bytes_with_nul_unchecked(b"__DATA,__thread_data\0")
                    };
                    llvm::LLVMSetSection(g, sect_name.as_ptr());
                }
            }

            // Wasm statics with custom link sections get special treatment as they
            // go into custom sections of the wasm executable.
            if self.tcx.sess.opts.target_triple.triple().starts_with("wasm32") {
                if let Some(section) = attrs.link_section {
                    let section = llvm::LLVMMDStringInContext(
                        self.llcx,
                        section.as_str().as_ptr().cast(),
                        section.as_str().len() as c_uint,
                    );
                    assert!(alloc.relocations().is_empty());

                    // The `inspect` method is okay here because we checked relocations, and
                    // because we are doing this access to inspect the final interpreter state (not
                    // as part of the interpreter execution).
                    let bytes =
                        alloc.inspect_with_undef_and_ptr_outside_interpreter(0..alloc.len());
                    let alloc = llvm::LLVMMDStringInContext(
                        self.llcx,
                        bytes.as_ptr().cast(),
                        bytes.len() as c_uint,
                    );
                    let data = [section, alloc];
                    let meta = llvm::LLVMMDNodeInContext(self.llcx, data.as_ptr(), 2);
                    llvm::LLVMAddNamedMetadataOperand(
                        self.llmod,
                        "wasm.custom_sections\0".as_ptr().cast(),
                        meta,
                    );
                }
            } else {
                base::set_link_section(g, &attrs);
            }

            if attrs.flags.contains(CodegenFnAttrFlags::USED) {
                // This static will be stored in the llvm.used variable which is an array of i8*
                let cast = llvm::LLVMConstPointerCast(g, self.type_i8p());
                self.used_statics.borrow_mut().push(cast);
            }
        }
    }
}