Rollup merge of #62809 - alexcrichton:wasm-llvm-9, r=nikic

rustc: Update wasm32 support for LLVM 9

This commit brings in a number of minor updates for rustc's support for
the wasm target which has changed in the LLVM 9 update. Notable updates
include:

* The compiler now no longer manually inserts the `producers` section,
  instead relying on LLVM to do so. LLVM uses the `llvm.ident` metadata
  for the `processed-by` directive (which is now emitted on the wasm
  target in this PR) and it uses debuginfo to figure out what `language`
  to put in the `producers` section.

* Threaded WebAssembly code now requires different flags to be passed
  with LLD. In LLD we now pass:

  * `--shared-memory` - required since objects are compiled with
    atomics. This also means that the generated memory will be marked as
    `shared`.
  * `--max-memory=1GB` - required with the `--shared-memory` argument
    since shared memories in WebAssembly must have a maximum size. The
    1GB number is intended to be a conservative estimate for rustc, but
    it should be overridable with `-C link-arg` if necessary.
  * `--passive-segments` - this has become the default for multithreaded
    memory, but when compiling a threaded module all data segments need
    to be marked as passive to ensure they don't re-initialize memory
    for each thread. This will also cause LLD to emit a synthetic
    function to initialize memory which users will have to arrange to
    call.
  * The `__heap_base` and `__data_end` globals are explicitly exported
    since they're now hidden by default due to the `--export` flags we
    pass to LLD.

4 files changed, 46 insertions, 201 deletions

diff --git a/src/librustc_codegen_ssa/back/link.rs b/src/librustc_codegen_ssa/back/link.rs
index 707b7cae16c..3f6a1a72ea6 100644
--- a/src/librustc_codegen_ssa/back/link.rs
+++ b/src/librustc_codegen_ssa/back/link.rs
@@ -678,14 +678,6 @@ fn link_natively<'a, B: ArchiveBuilder<'a>>(sess: &'a Session,
             sess.fatal(&format!("failed to run dsymutil: {}", e))
         }
     }
-
-    if sess.opts.target_triple.triple() == "wasm32-unknown-unknown" {
-        super::wasm::add_producer_section(
-            &out_filename,
-            &sess.edition().to_string(),
-            option_env!("CFG_VERSION").unwrap_or("unknown"),
-        );
-    }
 }
 
 /// Returns a boolean indicating whether the specified crate should be ignored
diff --git a/src/librustc_codegen_ssa/back/linker.rs b/src/librustc_codegen_ssa/back/linker.rs
index cb8870d0be9..26091005f25 100644
--- a/src/librustc_codegen_ssa/back/linker.rs
+++ b/src/librustc_codegen_ssa/back/linker.rs
@@ -901,7 +901,45 @@ pub struct WasmLd<'a> {
 }
 
 impl<'a> WasmLd<'a> {
-    fn new(cmd: Command, sess: &'a Session, info: &'a LinkerInfo) -> WasmLd<'a> {
+    fn new(mut cmd: Command, sess: &'a Session, info: &'a LinkerInfo) -> WasmLd<'a> {
+        // If the atomics feature is enabled for wasm then we need a whole bunch
+        // of flags:
+        //
+        // * `--shared-memory` - the link won't even succeed without this, flags
+        //   the one linear memory as `shared`
+        //
+        // * `--max-memory=1G` - when specifying a shared memory this must also
+        //   be specified. We conservatively choose 1GB but users should be able
+        //   to override this with `-C link-arg`.
+        //
+        // * `--import-memory` - it doesn't make much sense for memory to be
+        //   exported in a threaded module because typically you're
+        //   sharing memory and instantiating the module multiple times. As a
+        //   result if it were exported then we'd just have no sharing.
+        //
+        // * `--passive-segments` - all memory segments should be passive to
+        //   prevent each module instantiation from reinitializing memory.
+        //
+        // * `--export=__wasm_init_memory` - when using `--passive-segments` the
+        //   linker will synthesize this function, and so we need to make sure
+        //   that our usage of `--export` below won't accidentally cause this
+        //   function to get deleted.
+        //
+        // * `--export=*tls*` - when `#[thread_local]` symbols are used these
+        //   symbols are how the TLS segments are initialized and configured.
+        let atomics = sess.opts.cg.target_feature.contains("+atomics") ||
+            sess.target.target.options.features.contains("+atomics");
+        if atomics {
+            cmd.arg("--shared-memory");
+            cmd.arg("--max-memory=1073741824");
+            cmd.arg("--import-memory");
+            cmd.arg("--passive-segments");
+            cmd.arg("--export=__wasm_init_memory");
+            cmd.arg("--export=__wasm_init_tls");
+            cmd.arg("--export=__tls_size");
+            cmd.arg("--export=__tls_align");
+            cmd.arg("--export=__tls_base");
+        }
         WasmLd { cmd, sess, info }
     }
 }
@@ -1004,6 +1042,13 @@ impl<'a> Linker for WasmLd<'a> {
         for sym in self.info.exports[&crate_type].iter() {
             self.cmd.arg("--export").arg(&sym);
         }
+
+        // LLD will hide these otherwise-internal symbols since our `--export`
+        // list above is a whitelist of what to export. Various bits and pieces
+        // of tooling use this, so be sure these symbols make their way out of
+        // the linker as well.
+        self.cmd.arg("--export=__heap_base");
+        self.cmd.arg("--export=__data_end");
     }
 
     fn subsystem(&mut self, _subsystem: &str) {
diff --git a/src/librustc_codegen_ssa/back/mod.rs b/src/librustc_codegen_ssa/back/mod.rs
index a16d099ee3e..901891d85a4 100644
--- a/src/librustc_codegen_ssa/back/mod.rs
+++ b/src/librustc_codegen_ssa/back/mod.rs
@@ -6,4 +6,3 @@ pub mod command;
 pub mod symbol_export;
 pub mod archive;
 pub mod rpath;
-pub mod wasm;
diff --git a/src/librustc_codegen_ssa/back/wasm.rs b/src/librustc_codegen_ssa/back/wasm.rs
deleted file mode 100644
index 2a9e81a788e..00000000000
--- a/src/librustc_codegen_ssa/back/wasm.rs
+++ /dev/null
@@ -1,191 +0,0 @@
-use std::fs;
-use std::path::Path;
-use std::str;
-
-use rustc_serialize::leb128;
-
-// https://webassembly.github.io/spec/core/binary/modules.html#binary-importsec
-const WASM_CUSTOM_SECTION_ID: u8 = 0;
-
-/// Adds or augment the existing `producers` section to encode information about
-/// the Rust compiler used to produce the wasm file.
-pub fn add_producer_section(
-    path: &Path,
-    rust_version: &str,
-    rustc_version: &str,
-) {
-    struct Field<'a> {
-        name: &'a str,
-        values: Vec<FieldValue<'a>>,
-    }
-
-    #[derive(Copy, Clone)]
-    struct FieldValue<'a> {
-        name: &'a str,
-        version: &'a str,
-    }
-
-    let wasm = fs::read(path).expect("failed to read wasm output");
-    let mut ret = WasmEncoder::new();
-    ret.data.extend(&wasm[..8]);
-
-    // skip the 8 byte wasm/version header
-    let rustc_value = FieldValue {
-        name: "rustc",
-        version: rustc_version,
-    };
-    let rust_value = FieldValue {
-        name: "Rust",
-        version: rust_version,
-    };
-    let mut fields = Vec::new();
-    let mut wrote_rustc = false;
-    let mut wrote_rust = false;
-
-    // Move all sections from the original wasm file to our output, skipping
-    // everything except the producers section
-    for (id, raw) in WasmSections(WasmDecoder::new(&wasm[8..])) {
-        if id != WASM_CUSTOM_SECTION_ID {
-            ret.byte(id);
-            ret.bytes(raw);
-            continue
-        }
-        let mut decoder = WasmDecoder::new(raw);
-        if decoder.str() != "producers" {
-            ret.byte(id);
-            ret.bytes(raw);
-            continue
-        }
-
-        // Read off the producers section into our fields outside the loop,
-        // we'll re-encode the producers section when we're done (to handle an
-        // entirely missing producers section as well).
-        info!("rewriting existing producers section");
-
-        for _ in 0..decoder.u32() {
-            let name = decoder.str();
-            let mut values = Vec::new();
-            for _ in 0..decoder.u32() {
-                let name = decoder.str();
-                let version = decoder.str();
-                values.push(FieldValue { name, version });
-            }
-
-            if name == "language" {
-                values.push(rust_value);
-                wrote_rust = true;
-            } else if name == "processed-by" {
-                values.push(rustc_value);
-                wrote_rustc = true;
-            }
-            fields.push(Field { name, values });
-        }
-    }
-
-    if !wrote_rust {
-        fields.push(Field {
-            name: "language",
-            values: vec![rust_value],
-        });
-    }
-    if !wrote_rustc {
-        fields.push(Field {
-            name: "processed-by",
-            values: vec![rustc_value],
-        });
-    }
-
-    // Append the producers section to the end of the wasm file.
-    let mut section = WasmEncoder::new();
-    section.str("producers");
-    section.u32(fields.len() as u32);
-    for field in fields {
-        section.str(field.name);
-        section.u32(field.values.len() as u32);
-        for value in field.values {
-            section.str(value.name);
-            section.str(value.version);
-        }
-    }
-    ret.byte(WASM_CUSTOM_SECTION_ID);
-    ret.bytes(&section.data);
-
-    fs::write(path, &ret.data).expect("failed to write wasm output");
-}
-
-struct WasmSections<'a>(WasmDecoder<'a>);
-
-impl<'a> Iterator for WasmSections<'a> {
-    type Item = (u8, &'a [u8]);
-
-    fn next(&mut self) -> Option<(u8, &'a [u8])> {
-        if self.0.data.is_empty() {
-            return None
-        }
-
-        // see https://webassembly.github.io/spec/core/binary/modules.html#sections
-        let id = self.0.byte();
-        let section_len = self.0.u32();
-        info!("new section {} / {} bytes", id, section_len);
-        let section = self.0.skip(section_len as usize);
-        Some((id, section))
-    }
-}
-
-struct WasmDecoder<'a> {
-    data: &'a [u8],
-}
-
-impl<'a> WasmDecoder<'a> {
-    fn new(data: &'a [u8]) -> WasmDecoder<'a> {
-        WasmDecoder { data }
-    }
-
-    fn byte(&mut self) -> u8 {
-        self.skip(1)[0]
-    }
-
-    fn u32(&mut self) -> u32 {
-        let (n, l1) = leb128::read_u32_leb128(self.data);
-        self.data = &self.data[l1..];
-        return n
-    }
-
-    fn skip(&mut self, amt: usize) -> &'a [u8] {
-        let (data, rest) = self.data.split_at(amt);
-        self.data = rest;
-        data
-    }
-
-    fn str(&mut self) -> &'a str {
-        let len = self.u32();
-        str::from_utf8(self.skip(len as usize)).unwrap()
-    }
-}
-
-struct WasmEncoder {
-    data: Vec<u8>,
-}
-
-impl WasmEncoder {
-    fn new() -> WasmEncoder {
-        WasmEncoder { data: Vec::new() }
-    }
-
-    fn u32(&mut self, val: u32) {
-        leb128::write_u32_leb128(&mut self.data, val);
-    }
-
-    fn byte(&mut self, val: u8) {
-        self.data.push(val);
-    }
-
-    fn bytes(&mut self, val: &[u8]) {
-        self.u32(val.len() as u32);
-        self.data.extend_from_slice(val);
-    }
-
-    fn str(&mut self, val: &str) {
-        self.bytes(val.as_bytes())
-    }
-}