Commit d4d47032 authored by ahaas's avatar ahaas Committed by Commit bot

[wasm] Move the semaphore for parallel compilation to the wasm module.

If the semaphore is stored as a local variable in {CompileInParallel},
then the semaphore was sometimes deallocated too early and caused
the compilation tasks to crash. This only happens with libc-2.19,
libc-2.21 fixes the problem.

R=mlippautz@chromium.org, rossberg@chromium.org

Review-Url: https://codereview.chromium.org/2080223006
Cr-Commit-Position: refs/heads/master@{#37183}
parent bbbf21c2
......@@ -101,6 +101,9 @@ Semaphore::~Semaphore() {
void Semaphore::Signal() {
int result = sem_post(&native_handle_);
// This check may fail with <libc-2.21, which we use on the try bots, if the
// semaphore is destroyed while sem_post is still executed. A work around is
// to extend the lifetime of the semaphore.
CHECK_EQ(0, result);
}
......
......@@ -348,7 +348,8 @@ WasmModule::WasmModule()
start_function_index(-1),
origin(kWasmOrigin),
globals_size(0),
indirect_table_size(0) {}
indirect_table_size(0),
pending_tasks(new base::Semaphore(0)) {}
static MaybeHandle<JSFunction> ReportFFIError(ErrorThrower& thrower,
const char* error, uint32_t index,
......@@ -614,16 +615,15 @@ void CompileInParallel(Isolate* isolate, const WasmModule* module,
*module_env, *thrower);
// Objects for the synchronization with the background threads.
base::SmartPointer<base::Semaphore> pending_tasks(new base::Semaphore(0));
base::Mutex result_mutex;
base::AtomicNumber<size_t> next_unit(
static_cast<size_t>(FLAG_skip_compiling_wasm_funcs));
// 2) The main thread spawns {WasmCompilationTask} instances which run on
// the background threads.
base::SmartArrayPointer<uint32_t> task_ids(
StartCompilationTasks(isolate, compilation_units, executed_units,
pending_tasks.get(), result_mutex, next_unit));
base::SmartArrayPointer<uint32_t> task_ids(StartCompilationTasks(
isolate, compilation_units, executed_units, module->pending_tasks.get(),
result_mutex, next_unit));
// 3.a) The background threads and the main thread pick one compilation
// unit at a time and execute the parallel phase of the compilation
......@@ -640,7 +640,7 @@ void CompileInParallel(Isolate* isolate, const WasmModule* module,
}
// 4) After the parallel phase of all compilation units has started, the
// main thread waits for all {WasmCompilationTask} instances to finish.
WaitForCompilationTasks(isolate, task_ids.get(), pending_tasks.get());
WaitForCompilationTasks(isolate, task_ids.get(), module->pending_tasks.get());
// Finish the compilation of the remaining compilation units.
FinishCompilationUnits(executed_units, functions, result_mutex);
}
......
......@@ -178,6 +178,14 @@ struct WasmModule {
std::vector<uint16_t> function_table; // function table.
std::vector<WasmImport> import_table; // import table.
std::vector<WasmExport> export_table; // export table.
// We store the semaphore here to extend its lifetime. In <libc-2.21, which we
// use on the try bots, semaphore::Wait() can return while some compilation
// tasks are still executing semaphore::Signal(). If the semaphore is cleaned
// up right after semaphore::Wait() returns, then this can cause an
// invalid-semaphore error in the compilation tasks.
// TODO(wasm): Move this semaphore back to CompileInParallel when the try bots
// switch to libc-2.21 or higher.
base::SmartPointer<base::Semaphore> pending_tasks;
WasmModule();
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment