diff options
| author | Michael Matz <matz@suse.de> | 2010-08-22 16:53:24 +0200 |
|---|---|---|
| committer | Petr Baudis <pasky@suse.cz> | 2010-11-16 02:50:19 +0100 |
| commit | 31eb5a4c4986367538ccb154dfce0c84276ba151 (patch) | |
| tree | d5fdda573b28dd3c55ddd3acd0a580262ba97b92 | |
| parent | 9fcff7a79b60cd9202fdacafdbd5fb369933c64b (diff) | |
| download | glibc-31eb5a4c4986367538ccb154dfce0c84276ba151.tar.gz glibc-31eb5a4c4986367538ccb154dfce0c84276ba151.tar.xz glibc-31eb5a4c4986367538ccb154dfce0c84276ba151.zip | |
Add proper unwind information for x86_64 _fini
It is impossible to reliably unwind the stack above _fini() on x86_64 since no unwind information is provided for it and it modifies a stack register. This matters for gdb backtracing - if a process crashes within a destructor, it can frequently be essential to look at why the program began terminating in the first place.
| -rw-r--r-- | ChangeLog | 5 | ||||
| -rw-r--r-- | sysdeps/x86_64/elf/initfini.c | 33 |
2 files changed, 37 insertions, 1 deletions
diff --git a/ChangeLog b/ChangeLog index 122a2b8743..d253fde600 100644 --- a/ChangeLog +++ b/ChangeLog @@ -1,3 +1,8 @@ +2010-05-17 Michael Matz <matz@suse.de> + + [BZ #11610] + * sysdeps/x86_64/elf/initfini.c (_fini): Add unwind information. + 2010-05-12 Petr Baudis <pasky@suse.cz> [BZ #11589] diff --git a/sysdeps/x86_64/elf/initfini.c b/sysdeps/x86_64/elf/initfini.c index 30161d52ed..dd1a4cedc9 100644 --- a/sysdeps/x86_64/elf/initfini.c +++ b/sysdeps/x86_64/elf/initfini.c @@ -44,6 +44,25 @@ * crtn.s puts the corresponding function epilogues in the .init and .fini sections. */ +/* The unwind annotation for _fini is peculiar for good reasons: + (a) We need a real function that isn't constructed separately + (i.e. one which has a .size directive) in order to attach unwind + info to it. Hence _fini is a wrapper around _real_fini, the + former being a normal function, the latter being the first + instruction of the traditional _fini. + (b) We must not fiddle with the stack pointer in _real_fini, + as we wouldn't be able to describe the effects in unwind info + (c) some versions of GCC have no correct unwind info for + __do_global_dtors_aux, meaning they can't properly restore %rbp + (unwinding through it is possible but later up when we next + need %rbp we can't access it anymore) + Therefore we save/restore it in _fini for uses later up the call chain. + But we don't make the CFA use that register (that would lead to + the above problem) + (d) We want an 16-aligned stack pointer at _real_fini. Because of (a) + we can't align it in _real_fini, hence we do it in the caller by + subtracting 8, making in 8mod16 which the call then make 0mod16 + again. */ __asm__ ("\n\ #include \"defs.h\"\n\ \n\ @@ -88,16 +107,28 @@ _init:\n\ .globl _fini\n\ .type _fini,@function\n\ _fini:\n\ + .cfi_startproc\n\ + push %rbp\n\ + .cfi_def_cfa_offset 16\n\ + .cfi_offset 6,-16\n\ subq $8, %rsp\n\ + .cfi_def_cfa_offset 24\n\ + call _real_fini\n\ + addq $8, %rsp\n\ + .cfi_def_cfa_offset 16\n\ + pop %rbp\n\ + ret\n\ + .cfi_endproc\n\ ALIGN\n\ END_FINI\n\ +.size _fini, .-_fini\n\ +_real_fini:\n\ \n\ /*@_fini_PROLOG_ENDS*/\n\ call i_am_not_a_leaf@PLT\n\ \n\ /*@_fini_EPILOG_BEGINS*/\n\ .section .fini\n\ - addq $8, %rsp\n\ ret\n\ END_FINI\n\ \n\ |