It's been awhile since we last updated jemalloc, and there's likely some bugs
that have been fixed since the last version we're using, so let's try to update
again.
Fix formatting
Remove unused imports
Refactor
Fix msvc build
Fix line lengths
Formatting
Enable backtrace tests
Fix using directive on mac
pwd info
Work-around buildbot PWD bug, and fix libbacktrace configuration
Use alternative to `env -u` which is not supported on bitrig
Disable tests on 32-bit windows gnu
This commit is an implementation of [RFC 1183][rfc] which allows swapping out
the default allocator on nightly Rust. No new stable surface area should be
added as a part of this commit.
[rfc]: https://github.com/rust-lang/rfcs/pull/1183
Two new attributes have been added to the compiler:
* `#![needs_allocator]` - this is used by liballoc (and likely only liballoc) to
indicate that it requires an allocator crate to be in scope.
* `#![allocator]` - this is a indicator that the crate is an allocator which can
satisfy the `needs_allocator` attribute above.
The ABI of the allocator crate is defined to be a set of symbols that implement
the standard Rust allocation/deallocation functions. The symbols are not
currently checked for exhaustiveness or typechecked. There are also a number of
restrictions on these crates:
* An allocator crate cannot transitively depend on a crate that is flagged as
needing an allocator (e.g. allocator crates can't depend on liballoc).
* There can only be one explicitly linked allocator in a final image.
* If no allocator is explicitly requested one will be injected on behalf of the
compiler. Binaries and Rust dylibs will use jemalloc by default where
available and staticlibs/other dylibs will use the system allocator by
default.
Two allocators are provided by the distribution by default, `alloc_system` and
`alloc_jemalloc` which operate as advertised.
Closes#27389
This commit removes all morestack support from the compiler which entails:
* Segmented stacks are no longer emitted in codegen.
* We no longer build or distribute libmorestack.a
* The `stack_exhausted` lang item is no longer required
The only current use of the segmented stack support in LLVM is to detect stack
overflow. This is no longer really required, however, because we already have
guard pages for all threads and registered signal handlers watching for a
segfault on those pages (to print out a stack overflow message). Additionally,
major platforms (aka Windows) already don't use morestack.
This means that Rust is by default less likely to catch stack overflows because
if a function takes up more than one page of stack space it won't hit the guard
page. This is what the purpose of morestack was (to catch this case), but it's
better served with stack probes which have more cross platform support and no
runtime support necessary. Until LLVM supports this for all platform it looks
like morestack isn't really buying us much.
cc #16012 (still need stack probes)
Closes#26458 (a drive-by fix to help diagnostics on stack overflow)
This commit moves the IR files in the distribution, rust_try.ll,
rust_try_msvc_64.ll, and rust_try_msvc_32.ll into the compiler from the main
distribution. There's a few reasons for this change:
* LLVM changes its IR syntax from time to time, so it's very difficult to
have these files build across many LLVM versions simultaneously. We'll likely
want to retain this ability for quite some time into the future.
* The implementation of these files is closely tied to the compiler and runtime
itself, so it makes sense to fold it into a location which can do more
platform-specific checks for various implementation details (such as MSVC 32
vs 64-bit).
* This removes LLVM as a build-time dependency of the standard library. This may
end up becoming very useful if we move towards building the standard library
with Cargo.
In the immediate future, however, this commit should restore compatibility with
LLVM 3.5 and 3.6.
This commit modifies the configure script and our makefiles to support building
32-bit MSVC targets. The MSVC toolchain is now parameterized over whether it can
produce a 32-bit or 64-bit binary. The configure script was updated to export
more variables at configure time, and the makefiles were rejiggered to
selectively reexport the relevant environment variables for the applicable
targets they're going to run for.
Now that LLVM has been updated, the only remaining roadblock to implementing
unwinding for MSVC is to fill out the runtime support in `std::rt::unwind::seh`.
This commit does precisely that, fixing up some other bits and pieces along the
way:
* The `seh` unwinding module now uses `RaiseException` to initiate a panic.
* The `rust_try.ll` file was rewritten for MSVC (as it's quite different) and is
located at `rust_try_msvc_64.ll`, only included on MSVC builds for now.
* The personality function for all landing pads generated by LLVM is hard-wired
to `__C_specific_handler` instead of the standard `rust_eh_personality` lang
item. This is required to get LLVM to emit SEH unwinding information instead
of DWARF unwinding information. This also means that on MSVC the
`rust_eh_personality` function is entirely unused (but is defined as it's a
lang item).
More details about how panicking works on SEH can be found in the
`rust_try_msvc_64.ll` or `seh.rs` files, but I'm always open to adding more
comments!
A key aspect of this PR is missing, however, which is that **unwinding is still
turned off by default for MSVC**. There is a [bug in llvm][llvm-bug] which
causes optimizations to inline enough landing pads that LLVM chokes. If the
compiler is optimized at `-O1` (where inlining isn't enabled) then it can
bootstrap with unwinding enabled, but when optimized at `-O2` (inlining is
enabled) then it hits a fatal LLVM error.
[llvm-bug]: https://llvm.org/bugs/show_bug.cgi?id=23884
It looks like compiler-rt has a cmake build sytem inside its source, but I have
been unable to figure out how to use it and actually build the right library.
For now this commit hard-wires MSVC-targeting builds of libcompiler-rt to
continue using `make` as the primary bulid system, but some frobbing of the
flags are necessary to ensure that the right compiler is used.
We have a number of support C/C++ files in Rust that we link into the standard
library and other various locations, and these all need to be built with cl.exe
instead of gcc.exe when targeting MSVC. This commit adds helper macros for this
functionality to use different sets of programs/flags/invocations on MSVC than
on GNU-like platforms.
Previously libmorestack.a and libcompiler-rt.a were installed, but link.exe
looks for morestack.lib and compiler-rt.lib by default, so we need to install
these with the correct name
There were a few test cases to fix:
* Dynamic libraries are not supported with MUSL right now, so all of those
related test which force or require dylibs are ignored.
* Looks like the default stack for MUSL is smaller than glibc, so a few stack
allocations in benchmarks were boxed up (shouldn't have a perf impact).
* Some small linkage tweaks here and there
* Out-of-stack detection does not currently work with MUSL
This commit adds support to the makefiles, configuration script, and build
system to understand MUSL. This is broken up into a few parts:
* Any target of the form `*-musl` requires the `--musl-root` option to
`./configure` which will indicate the root of the MUSL installation. It is
also expected that there is a libunwind build inside of that installation
built against that MUSL.
* Objects from MUSL are copied into the build tree for Rust to be statically
linked into the appropriate Rust library.
* Objects for binary startup and shutdown are included in each Rust installation
by default for MUSL. This requires MUSL to only be installed on the machine
compiling rust. Only a linker will be necessary for compiling against MUSL on
a target machine.
Eventually a MUSL and/or libunwind build may be integrated by default into the
build but for now they are just always assumed to exist externally.
It's quite possible that small programs don't use all of jemalloc, and building
with -ffunction-sections and -fdata-sections allows the linker (via
--gc-sections) to strip out all unused code at link time. This decreases the
size of a "hello world" executable for me from 716K to 482K with no measurable
impact on link time. After this patch jemalloc is still the largest portion of
our hello world executables, but this helps cut down on the size at least
somewhat!
We add CFG_LLVM_TARGET_$(target) (which can be defined in any of the
mk/cfg/* files) and supply a default to the plain target name
CFG_LLVM_TARGET mirrors the value of llvm_target (aka llvm-target) in
the librustc_back runtime target specification.
- CFG_CFLAGS is gone (it was previously only used by jemalloc anyhow).
- CFG_JEMALLOC_CFLAGS may contain flags needed for the compiler to
function (produce a binary output).
- jemalloc's configure runs $(CC) without EXTRA_CFLAGS, and (without
this change) will fail if any flags are required for CC to work.
Removes all target-specific knowledge from rustc. Some targets have changed
during this, but none of these should be very visible outside of
cross-compilation. The changes make our targets more consistent.
iX86-unknown-linux-gnu is now only available as i686-unknown-linux-gnu. We
used to accept any value of X greater than 1. i686 was released in 1995, and
should encompass the bare minimum of what Rust supports on x86 CPUs.
The only two windows targets are now i686-pc-windows-gnu and
x86_64-pc-windows-gnu.
The iOS target has been renamed from arm-apple-ios to arm-apple-darwin.
A complete list of the targets we accept now:
arm-apple-darwin
arm-linux-androideabi
arm-unknown-linux-gnueabi
arm-unknown-linux-gnueabihf
i686-apple-darwin
i686-pc-windows-gnu
i686-unknown-freebsd
i686-unknown-linux-gnu
mips-unknown-linux-gnu
mipsel-unknown-linux-gnu
x86_64-apple-darwin
x86_64-unknown-freebsd
x86_64-unknown-linux-gnu
x86_64-pc-windows-gnu
Closes#16093
[breaking-change]
This commit removes the libuv and gyp submodules, as well as all build
infrastructure related to them.
For more context, see the [runtime removal
RFC](https://github.com/rust-lang/rfcs/pull/230)
[breaking-change]
The performance hit from these checks is significant, but unoptimized
builds are already incredibly slow. Enabling these checks results in
better test coverage since there are bots doing unoptimized builds, and
the cost is relatively small in the context of an unoptimized build.
This also allows using `JEMALLOC_FLAGS` to override the default
configure flags.
Not included are two required patches:
* LLVM: segmented stack support for DragonFly [1]
* jemalloc: simple configure patches
[1]: http://reviews.llvm.org/D4705
It seems in one of rebases I’ve resolved conflicts wrong and left one redundant line, it is absent in current master and it might cause compilation failure by copying file into itself.
This adds a new configure option, --jemalloc-root, which will specify a location
at which libjemalloc_pic.a must live. This library is then used for the build
triple as the jemalloc library to link.
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]
By default, jemalloc is building itself with -g3 if the local compiler supports
it. It looks like this is generating a good deal of debug info that windows
isn't optimizing out (on the order of 18MB). Windows gcc/ld is also not
optimizing this data away, causing hello world to be 18MB in size.
There's no current real need for debugging jemalloc to a great extent, so this
commit manually passes -g1 to override -g3 which jemalloc is using. This is
confirmed to drop the size of executables on windows back to a more reasonable
size (2.0MB, as they were before).
Closes#14144
This adds a `std::rt::heap` module with a nice allocator API. It's a
step towards fixing #13094 and is a starting point for working on a
generic allocator trait.
The revision used for the jemalloc submodule is the stable 3.6.0 release.
Closes#11807
This primary fix brought on by this upgrade is the proper matching of the ```
and ~~~ doc blocks. This also moves hoedown to a git submodule rather than a
bundled repository.
Additionally, hoedown is stricter about code blocks, so this ended up fixing a
lot of invalid code blocks (ending with " ```" instead of "```", or ending with
"~~~~" instead of "~~~").
Closes#12776
Whenever a failure happens, if a program is run with
`RUST_LOG=std::rt::backtrace` a backtrace will be printed to the task's stderr
handle. Stack traces are uncondtionally printed on double-failure and
rtabort!().
This ended up having a nontrivial implementation, and here's some highlights of
it:
* We're bundling libbacktrace for everything but OSX and Windows
* We use libgcc_s and its libunwind apis to get a backtrace of instruction
pointers
* On OSX we use dladdr() to go from an instruction pointer to a symbol
* On unix that isn't OSX, we use libbacktrace to get symbols
* Windows, as usual, has an entirely separate implementation
Lots more fun details and comments can be found in the source itself.
Closes#10128
Currently when you run `make -jN` it's likely that you'll remove compiler-rt and
then it won't get cp'd back into the right place. I believe the reason for this
is that the compiler-rt library target never got updated so make decided it
never needed to copy the files back into place. The files were all there at the
beginning of `make`, but then we may clean out the stage0 versions if we unzip
the snapshot again.
