rust/src/librustrt/stack.rs

344 lines
15 KiB
Rust
Raw Normal View History

// Copyright 2013 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.
//! Rust stack-limit management
//!
//! Currently Rust uses a segmented-stack-like scheme in order to detect stack
//! overflow for rust tasks. In this scheme, the prologue of all functions are
//! preceded with a check to see whether the current stack limits are being
//! exceeded.
//!
//! This module provides the functionality necessary in order to manage these
//! stack limits (which are stored in platform-specific locations). The
//! functions here are used at the borders of the task lifetime in order to
//! manage these limits.
//!
//! This function is an unstable module because this scheme for stack overflow
//! detection is not guaranteed to continue in the future. Usage of this module
//! is discouraged unless absolutely necessary.
2014-05-05 02:08:34 -05:00
// iOS related notes
//
// It is possible to implement it using idea from
// http://www.opensource.apple.com/source/Libc/Libc-825.40.1/pthreads/pthread_machdep.h
//
// In short: _pthread_{get,set}_specific_direct allows extremely fast
// access, exactly what is required for segmented stack
// There is a pool of reserved slots for Apple internal use (0..119)
// First dynamic allocated pthread key starts with 257 (on iOS7)
// So using slot 149 should be pretty safe ASSUMING space is reserved
// for every key < first dynamic key
//
// There is also an opportunity to steal keys reserved for Garbage Collection
// ranges 80..89 and 110..119, especially considering the fact Garbage Collection
// never supposed to work on iOS. But as everybody knows it - there is a chance
// that those slots will be re-used, like it happened with key 95 (moved from
// JavaScriptCore to CoreText)
//
// Unfortunately Apple rejected patch to LLVM which generated
// corresponding prolog, decision was taken to disable segmented
// stack support on iOS.
pub static RED_ZONE: uint = 20 * 1024;
/// This function is invoked from rust's current __morestack function. Segmented
/// stacks are currently not enabled as segmented stacks, but rather one giant
/// stack segment. This means that whenever we run out of stack, we want to
/// truly consider it to be stack overflow rather than allocating a new stack.
2014-05-29 19:32:50 -05:00
#[cfg(not(test))] // in testing, use the original libstd's version
rustc: Add official support for weak failure This commit is part of the ongoing libstd facade efforts (cc #13851). The compiler now recognizes some language items as "extern { fn foo(...); }" and will automatically perform the following actions: 1. The foreign function has a pre-defined name. 2. The crate and downstream crates can only be built as rlibs until a crate defines the lang item itself. 3. The actual lang item has a pre-defined name. This is essentially nicer compiler support for the hokey core-depends-on-std-failure scheme today, but it is implemented the same way. The details are a little more hidden under the covers. In addition to failure, this commit promotes the eh_personality and rust_stack_exhausted functions to official lang items. The compiler can generate calls to these functions, causing linkage errors if they are left undefined. The checking for these items is not as precise as it could be. Crates compiling with `-Z no-landing-pads` will not need the eh_personality lang item, and crates compiling with no split stacks won't need the stack exhausted lang item. For ease, however, these items are checked for presence in all final outputs of the compiler. It is quite easy to define dummy versions of the functions necessary: #[lang = "stack_exhausted"] extern fn stack_exhausted() { /* ... */ } #[lang = "eh_personality"] extern fn eh_personality() { /* ... */ } cc #11922, rust_stack_exhausted is now a lang item cc #13851, libcollections is blocked on eh_personality becoming weak
2014-05-19 11:30:09 -05:00
#[lang = "stack_exhausted"]
extern fn stack_exhausted() {
std: Extract librustrt out of libstd As part of the libstd facade efforts, this commit extracts the runtime interface out of the standard library into a standalone crate, librustrt. This crate will provide the following services: * Definition of the rtio interface * Definition of the Runtime interface * Implementation of the Task structure * Implementation of task-local-data * Implementation of task failure via unwinding via libunwind * Implementation of runtime initialization and shutdown * Implementation of thread-local-storage for the local rust Task Notably, this crate avoids the following services: * Thread creation and destruction. The crate does not require the knowledge of an OS threading system, and as a result it seemed best to leave out the `rt::thread` module from librustrt. The librustrt module does depend on mutexes, however. * Implementation of backtraces. There is no inherent requirement for the runtime to be able to generate backtraces. As will be discussed later, this functionality continues to live in libstd rather than librustrt. As usual, a number of architectural changes were required to make this crate possible. Users of "stable" functionality will not be impacted by this change, but users of the `std::rt` module will likely note the changes. A list of architectural changes made is: * The stdout/stderr handles no longer live directly inside of the `Task` structure. This is a consequence of librustrt not knowing about `std::io`. These two handles are now stored inside of task-local-data. The handles were originally stored inside of the `Task` for perf reasons, and TLD is not currently as fast as it could be. For comparison, 100k prints goes from 59ms to 68ms (a 15% slowdown). This appeared to me to be an acceptable perf loss for the successful extraction of a librustrt crate. * The `rtio` module was forced to duplicate more functionality of `std::io`. As the module no longer depends on `std::io`, `rtio` now defines structures such as socket addresses, addrinfo fiddly bits, etc. The primary change made was that `rtio` now defines its own `IoError` type. This type is distinct from `std::io::IoError` in that it does not have an enum for what error occurred, but rather a platform-specific error code. The native and green libraries will be updated in later commits for this change, and the bulk of this effort was put behind updating the two libraries for this change (with `rtio`). * Printing a message on task failure (along with the backtrace) continues to live in libstd, not in librustrt. This is a consequence of the above decision to move the stdout/stderr handles to TLD rather than inside the `Task` itself. The unwinding API now supports registration of global callback functions which will be invoked when a task fails, allowing for libstd to register a function to print a message and a backtrace. The API for registering a callback is experimental and unsafe, as the ramifications of running code on unwinding is pretty hairy. * The `std::unstable::mutex` module has moved to `std::rt::mutex`. * The `std::unstable::sync` module has been moved to `std::rt::exclusive` and the type has been rewritten to not internally have an Arc and to have an RAII guard structure when locking. Old code should stop using `Exclusive` in favor of the primitives in `libsync`, but if necessary, old code should port to `Arc<Exclusive<T>>`. * The local heap has been stripped down to have fewer debugging options. None of these were tested, and none of these have been used in a very long time. [breaking-change]
2014-06-03 21:11:49 -05:00
use core::prelude::*;
use alloc::boxed::Box;
std: Extract librustrt out of libstd As part of the libstd facade efforts, this commit extracts the runtime interface out of the standard library into a standalone crate, librustrt. This crate will provide the following services: * Definition of the rtio interface * Definition of the Runtime interface * Implementation of the Task structure * Implementation of task-local-data * Implementation of task failure via unwinding via libunwind * Implementation of runtime initialization and shutdown * Implementation of thread-local-storage for the local rust Task Notably, this crate avoids the following services: * Thread creation and destruction. The crate does not require the knowledge of an OS threading system, and as a result it seemed best to leave out the `rt::thread` module from librustrt. The librustrt module does depend on mutexes, however. * Implementation of backtraces. There is no inherent requirement for the runtime to be able to generate backtraces. As will be discussed later, this functionality continues to live in libstd rather than librustrt. As usual, a number of architectural changes were required to make this crate possible. Users of "stable" functionality will not be impacted by this change, but users of the `std::rt` module will likely note the changes. A list of architectural changes made is: * The stdout/stderr handles no longer live directly inside of the `Task` structure. This is a consequence of librustrt not knowing about `std::io`. These two handles are now stored inside of task-local-data. The handles were originally stored inside of the `Task` for perf reasons, and TLD is not currently as fast as it could be. For comparison, 100k prints goes from 59ms to 68ms (a 15% slowdown). This appeared to me to be an acceptable perf loss for the successful extraction of a librustrt crate. * The `rtio` module was forced to duplicate more functionality of `std::io`. As the module no longer depends on `std::io`, `rtio` now defines structures such as socket addresses, addrinfo fiddly bits, etc. The primary change made was that `rtio` now defines its own `IoError` type. This type is distinct from `std::io::IoError` in that it does not have an enum for what error occurred, but rather a platform-specific error code. The native and green libraries will be updated in later commits for this change, and the bulk of this effort was put behind updating the two libraries for this change (with `rtio`). * Printing a message on task failure (along with the backtrace) continues to live in libstd, not in librustrt. This is a consequence of the above decision to move the stdout/stderr handles to TLD rather than inside the `Task` itself. The unwinding API now supports registration of global callback functions which will be invoked when a task fails, allowing for libstd to register a function to print a message and a backtrace. The API for registering a callback is experimental and unsafe, as the ramifications of running code on unwinding is pretty hairy. * The `std::unstable::mutex` module has moved to `std::rt::mutex`. * The `std::unstable::sync` module has been moved to `std::rt::exclusive` and the type has been rewritten to not internally have an Arc and to have an RAII guard structure when locking. Old code should stop using `Exclusive` in favor of the primitives in `libsync`, but if necessary, old code should port to `Arc<Exclusive<T>>`. * The local heap has been stripped down to have fewer debugging options. None of these were tested, and none of these have been used in a very long time. [breaking-change]
2014-06-03 21:11:49 -05:00
use local::Local;
use task::Task;
use core::intrinsics;
unsafe {
// We're calling this function because the stack just ran out. We need
// to call some other rust functions, but if we invoke the functions
// right now it'll just trigger this handler being called again. In
// order to alleviate this, we move the stack limit to be inside of the
// red zone that was allocated for exactly this reason.
let limit = get_sp_limit();
record_sp_limit(limit - RED_ZONE / 2);
// This probably isn't the best course of action. Ideally one would want
// to unwind the stack here instead of just aborting the entire process.
// This is a tricky problem, however. There's a few things which need to
// be considered:
//
// 1. We're here because of a stack overflow, yet unwinding will run
// destructors and hence arbitrary code. What if that code overflows
// the stack? One possibility is to use the above allocation of an
// extra 10k to hope that we don't hit the limit, and if we do then
// abort the whole program. Not the best, but kind of hard to deal
// with unless we want to switch stacks.
//
// 2. LLVM will optimize functions based on whether they can unwind or
// not. It will flag functions with 'nounwind' if it believes that
// the function cannot trigger unwinding, but if we do unwind on
// stack overflow then it means that we could unwind in any function
// anywhere. We would have to make sure that LLVM only places the
// nounwind flag on functions which don't call any other functions.
//
// 3. The function that overflowed may have owned arguments. These
// arguments need to have their destructors run, but we haven't even
// begun executing the function yet, so unwinding will not run the
// any landing pads for these functions. If this is ignored, then
// the arguments will just be leaked.
//
// Exactly what to do here is a very delicate topic, and is possibly
// still up in the air for what exactly to do. Some relevant issues:
//
// #3555 - out-of-stack failure leaks arguments
// #3695 - should there be a stack limit?
// #9855 - possible strategies which could be taken
// #9854 - unwinding on windows through __morestack has never worked
// #2361 - possible implementation of not using landing pads
let task: Option<Box<Task>> = Local::try_take();
let name = match task {
Some(ref task) => {
task.name.as_ref().map(|n| n.as_slice())
}
None => None
};
let name = name.unwrap_or("<unknown>");
// See the message below for why this is not emitted to the
// task's logger. This has the additional conundrum of the
// logger may not be initialized just yet, meaning that an FFI
// call would happen to initialized it (calling out to libuv),
// and the FFI call needs 2MB of stack when we just ran out.
rterrln!("task '{}' has overflowed its stack", name);
intrinsics::abort();
}
}
// Windows maintains a record of upper and lower stack bounds in the Thread Information
// Block (TIB), and some syscalls do check that addresses which are supposed to be in
// the stack, indeed lie between these two values.
// (See https://github.com/rust-lang/rust/issues/3445#issuecomment-26114839)
//
// When using Rust-managed stacks (libgreen), we must maintain these values accordingly.
// For OS-managed stacks (libnative), we let the OS manage them for us.
//
// On all other platforms both variants behave identically.
#[inline(always)]
pub unsafe fn record_os_managed_stack_bounds(stack_lo: uint, _stack_hi: uint) {
record_sp_limit(stack_lo + RED_ZONE);
}
#[inline(always)]
pub unsafe fn record_rust_managed_stack_bounds(stack_lo: uint, stack_hi: uint) {
// When the old runtime had segmented stacks, it used a calculation that was
// "limit + RED_ZONE + FUDGE". The red zone was for things like dynamic
// symbol resolution, llvm function calls, etc. In theory this red zone
// value is 0, but it matters far less when we have gigantic stacks because
// we don't need to be so exact about our stack budget. The "fudge factor"
// was because LLVM doesn't emit a stack check for functions < 256 bytes in
// size. Again though, we have giant stacks, so we round all these
// calculations up to the nice round number of 20k.
record_sp_limit(stack_lo + RED_ZONE);
return target_record_stack_bounds(stack_lo, stack_hi);
#[cfg(not(windows))] #[inline(always)]
unsafe fn target_record_stack_bounds(_stack_lo: uint, _stack_hi: uint) {}
2014-09-29 01:38:24 -05:00
#[cfg(all(windows, target_arch = "x86"))] #[inline(always)]
unsafe fn target_record_stack_bounds(stack_lo: uint, stack_hi: uint) {
// stack range is at TIB: %fs:0x04 (top) and %fs:0x08 (bottom)
asm!("mov $0, %fs:0x04" :: "r"(stack_hi) :: "volatile");
asm!("mov $0, %fs:0x08" :: "r"(stack_lo) :: "volatile");
}
2014-09-29 01:38:24 -05:00
#[cfg(all(windows, target_arch = "x86_64"))] #[inline(always)]
unsafe fn target_record_stack_bounds(stack_lo: uint, stack_hi: uint) {
// stack range is at TIB: %gs:0x08 (top) and %gs:0x10 (bottom)
asm!("mov $0, %gs:0x08" :: "r"(stack_hi) :: "volatile");
asm!("mov $0, %gs:0x10" :: "r"(stack_lo) :: "volatile");
}
}
/// Records the current limit of the stack as specified by `end`.
///
/// This is stored in an OS-dependent location, likely inside of the thread
/// local storage. The location that the limit is stored is a pre-ordained
/// location because it's where LLVM has emitted code to check.
///
/// Note that this cannot be called under normal circumstances. This function is
/// changing the stack limit, so upon returning any further function calls will
/// possibly be triggering the morestack logic if you're not careful.
///
/// Also note that this and all of the inside functions are all flagged as
/// "inline(always)" because they're messing around with the stack limits. This
/// would be unfortunate for the functions themselves to trigger a morestack
/// invocation (if they were an actual function call).
#[inline(always)]
pub unsafe fn record_sp_limit(limit: uint) {
return target_record_sp_limit(limit);
// x86-64
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86_64",
any(target_os = "macos", target_os = "ios")))]
#[inline(always)]
unsafe fn target_record_sp_limit(limit: uint) {
asm!("movq $$0x60+90*8, %rsi
movq $0, %gs:(%rsi)" :: "r"(limit) : "rsi" : "volatile")
}
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86_64", target_os = "linux"))] #[inline(always)]
unsafe fn target_record_sp_limit(limit: uint) {
asm!("movq $0, %fs:112" :: "r"(limit) :: "volatile")
}
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86_64", target_os = "windows"))] #[inline(always)]
unsafe fn target_record_sp_limit(_: uint) {
}
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86_64", target_os = "freebsd"))] #[inline(always)]
unsafe fn target_record_sp_limit(limit: uint) {
asm!("movq $0, %fs:24" :: "r"(limit) :: "volatile")
}
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86_64", target_os = "dragonfly"))] #[inline(always)]
unsafe fn target_record_sp_limit(limit: uint) {
asm!("movq $0, %fs:32" :: "r"(limit) :: "volatile")
}
// x86
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86",
any(target_os = "macos", target_os = "ios")))]
#[inline(always)]
unsafe fn target_record_sp_limit(limit: uint) {
asm!("movl $$0x48+90*4, %eax
movl $0, %gs:(%eax)" :: "r"(limit) : "eax" : "volatile")
}
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86",
any(target_os = "linux", target_os = "freebsd")))]
#[inline(always)]
unsafe fn target_record_sp_limit(limit: uint) {
asm!("movl $0, %gs:48" :: "r"(limit) :: "volatile")
}
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86", target_os = "windows"))] #[inline(always)]
unsafe fn target_record_sp_limit(_: uint) {
}
// mips, arm - Some brave soul can port these to inline asm, but it's over
// my head personally
2014-09-29 01:38:24 -05:00
#[cfg(any(target_arch = "mips",
target_arch = "mipsel",
all(target_arch = "arm", not(target_os = "ios"))))]
#[inline(always)]
unsafe fn target_record_sp_limit(limit: uint) {
use libc::c_void;
2014-06-25 14:47:34 -05:00
return record_sp_limit(limit as *const c_void);
extern {
2014-06-25 14:47:34 -05:00
fn record_sp_limit(limit: *const c_void);
}
}
2014-05-05 02:08:34 -05:00
// iOS segmented stack is disabled for now, see related notes
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "arm", target_os = "ios"))] #[inline(always)]
2014-05-05 02:08:34 -05:00
unsafe fn target_record_sp_limit(_: uint) {
}
}
/// The counterpart of the function above, this function will fetch the current
/// stack limit stored in TLS.
///
/// Note that all of these functions are meant to be exact counterparts of their
/// brethren above, except that the operands are reversed.
///
/// As with the setter, this function does not have a __morestack header and can
/// therefore be called in a "we're out of stack" situation.
#[inline(always)]
pub unsafe fn get_sp_limit() -> uint {
return target_get_sp_limit();
// x86-64
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86_64",
any(target_os = "macos", target_os = "ios")))]
#[inline(always)]
unsafe fn target_get_sp_limit() -> uint {
let limit;
asm!("movq $$0x60+90*8, %rsi
movq %gs:(%rsi), $0" : "=r"(limit) :: "rsi" : "volatile");
return limit;
}
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86_64", target_os = "linux"))] #[inline(always)]
unsafe fn target_get_sp_limit() -> uint {
let limit;
asm!("movq %fs:112, $0" : "=r"(limit) ::: "volatile");
return limit;
}
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86_64", target_os = "windows"))] #[inline(always)]
unsafe fn target_get_sp_limit() -> uint {
return 1024;
}
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86_64", target_os = "freebsd"))] #[inline(always)]
unsafe fn target_get_sp_limit() -> uint {
let limit;
asm!("movq %fs:24, $0" : "=r"(limit) ::: "volatile");
return limit;
}
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86_64", target_os = "dragonfly"))] #[inline(always)]
unsafe fn target_get_sp_limit() -> uint {
let limit;
asm!("movq %fs:32, $0" : "=r"(limit) ::: "volatile");
return limit;
}
// x86
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86",
any(target_os = "macos", target_os = "ios")))]
#[inline(always)]
unsafe fn target_get_sp_limit() -> uint {
let limit;
asm!("movl $$0x48+90*4, %eax
movl %gs:(%eax), $0" : "=r"(limit) :: "eax" : "volatile");
return limit;
}
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86",
any(target_os = "linux", target_os = "freebsd")))]
#[inline(always)]
unsafe fn target_get_sp_limit() -> uint {
let limit;
asm!("movl %gs:48, $0" : "=r"(limit) ::: "volatile");
return limit;
}
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "x86", target_os = "windows"))] #[inline(always)]
unsafe fn target_get_sp_limit() -> uint {
return 1024;
}
// mips, arm - Some brave soul can port these to inline asm, but it's over
// my head personally
2014-09-29 01:38:24 -05:00
#[cfg(any(target_arch = "mips",
target_arch = "mipsel",
all(target_arch = "arm", not(target_os = "ios"))))]
#[inline(always)]
unsafe fn target_get_sp_limit() -> uint {
use libc::c_void;
return get_sp_limit() as uint;
extern {
2014-06-25 14:47:34 -05:00
fn get_sp_limit() -> *const c_void;
}
}
2014-05-05 02:08:34 -05:00
// iOS doesn't support segmented stacks yet. This function might
// be called by runtime though so it is unsafe to mark it as
// unreachable, let's return a fixed constant.
2014-09-29 01:38:24 -05:00
#[cfg(all(target_arch = "arm", target_os = "ios"))] #[inline(always)]
2014-05-05 02:08:34 -05:00
unsafe fn target_get_sp_limit() -> uint {
1024
}
}