提交 b477f7a7 编写于 作者: B bors

auto merge of #10013 : bdash/rust/mavericks-unwinding, r=alexcrichton

OS X 10.9's linker has a bug that results in it failing to preserve
DWARF unwind information when passed the -no_compact_unwind flag.
This flag is passed on OS X because the unwind information for
__morestack cannot be represented by the compact unwind format.

We can work around this problem by using a more targeted approach
to disabling compact unwind information. The OS X linker looks for
a particular pattern in the DWARF unwind information and will not
attempt to convert the unwind information to the compact format.
The pattern in question is the return address register being saved
twice to the same location.

Fixes #6849.
......@@ -204,7 +204,7 @@ CFG_LIB_GLOB_x86_64-apple-darwin=lib$(1)-*.dylib
CFG_LIB_DSYM_GLOB_x86_64-apple-darwin=lib$(1)-*.dylib.dSYM
CFG_GCCISH_CFLAGS_x86_64-apple-darwin := -Wall -Werror -g -fPIC -m64 -arch x86_64
CFG_GCCISH_CXXFLAGS_x86_64-apple-darwin := -fno-rtti
CFG_GCCISH_LINK_FLAGS_x86_64-apple-darwin := -dynamiclib -pthread -framework CoreServices -Wl,-no_compact_unwind -m64
CFG_GCCISH_LINK_FLAGS_x86_64-apple-darwin := -dynamiclib -pthread -framework CoreServices -m64
CFG_GCCISH_DEF_FLAG_x86_64-apple-darwin := -Wl,-exported_symbols_list,
CFG_GCCISH_PRE_LIB_FLAGS_x86_64-apple-darwin :=
CFG_GCCISH_POST_LIB_FLAGS_x86_64-apple-darwin :=
......@@ -230,7 +230,7 @@ CFG_LIB_GLOB_i686-apple-darwin=lib$(1)-*.dylib
CFG_LIB_DSYM_GLOB_i686-apple-darwin=lib$(1)-*.dylib.dSYM
CFG_GCCISH_CFLAGS_i686-apple-darwin := -Wall -Werror -g -fPIC -m32 -arch i386
CFG_GCCISH_CXXFLAGS_i686-apple-darwin := -fno-rtti
CFG_GCCISH_LINK_FLAGS_i686-apple-darwin := -dynamiclib -pthread -framework CoreServices -Wl,-no_compact_unwind -m32
CFG_GCCISH_LINK_FLAGS_i686-apple-darwin := -dynamiclib -pthread -framework CoreServices -m32
CFG_GCCISH_DEF_FLAG_i686-apple-darwin := -Wl,-exported_symbols_list,
CFG_GCCISH_PRE_LIB_FLAGS_i686-apple-darwin :=
CFG_GCCISH_POST_LIB_FLAGS_i686-apple-darwin :=
......
......@@ -1112,14 +1112,6 @@ fn unlib(config: @session::config, stem: ~str) -> ~str {
~"-Wl,-rpath,/usr/local/lib/gcc44"]);
}
// OS X 10.6 introduced 'compact unwind info', which is produced by the
// linker from the dwarf unwind info. Unfortunately, it does not seem to
// understand how to unwind our __morestack frame, so we have to turn it
// off. This has impacted some other projects like GHC.
if sess.targ_cfg.os == session::OsMacos {
args.push(~"-Wl,-no_compact_unwind");
}
// Stack growth requires statically linking a __morestack function
args.push(~"-lmorestack");
......
......@@ -62,9 +62,9 @@
is that OS X 10.6+ uses its own 'compact unwind info',
an undocumented format generated by the linker from
the DWARF CFI. This compact unwind info doesn't correctly
capture the nuance of the __morestack frame, and as a
result all of our linking on OS X uses the -no_compact_unwind
flag.
capture the nuance of the __morestack frame, so we need to
prevent the linker from attempting to convert its DWARF unwind
information.
*/
.text
......@@ -118,6 +118,15 @@ MORESTACK:
// FIXME(#9854) these cfi directives don't work on windows.
pushl %ebp
#if defined(__APPLE__)
// The pattern of the return address being saved twice to the same location
// tells the OS X linker that it should not attempt to convert the DWARF
// unwind information to the compact format.
.cfi_offset %eip, -4
.cfi_offset %eip, -4
#endif
// The CFA is 20 bytes above the register that it is
// associated with for this frame (which will be %ebp)
.cfi_def_cfa_offset 20
......
......@@ -43,6 +43,15 @@ MORESTACK:
// bytes greater than a normal frame, to allow the unwinder
// to skip the partial frame of the original function.
.cfi_def_cfa_offset 24
#if defined(__APPLE__)
// The pattern of the return address being saved twice to the same location
// tells the OS X linker that it should not attempt to convert the DWARF
// unwind information to the compact format.
.cfi_offset %rip, -8
.cfi_offset %rip, -8
#endif
// %rbp is -24 bytes from the CFA
.cfi_offset %rbp, -24
movq %rsp, %rbp
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册