If I modify tests/Gtest-bt.c with this patch:
diff --git a/tests/Gtest-bt.c b/tests/Gtest-bt.c
index 802dd08..82d8317 100644
--- a/tests/Gtest-bt.c
+++ b/tests/Gtest-bt.c
@@ -86,9 +86,10 @@ do_backtrace (void)
if (unw_get_proc_info (&cursor, &pi) == 0)
{
- printf ("\tproc=0x%lx-0x%lx\n\thandler=0x%lx lsda=0x%lx gp=0x%lx",
+ printf ("\tproc=0x%lx-0x%lx\n\thandler=0x%lx lsda=0x%lx gp=0x%lx
uw=0x%lx",
(long) pi.start_ip, (long) pi.end_ip,
- (long) pi.handler, (long) pi.lsda, (long) pi.gp);
+ (long) pi.handler, (long) pi.lsda, (long) pi.gp,
+ (long) pi.unwind_info);
}
#if UNW_TARGET_IA64
and then I run
./Gtest-bt foo | grep uw
with ‘foo’ for verbose output, and looking for the unwind info added by the
patch, I see that in every case, the unwind info is 0x0.
Doug
> On Apr 11, 2017, at 9:32 AM, Dave Watson <davejwat...@fb.com> wrote:
>
> On 04/11/17 12:16 AM, Doug Moore wrote:
>>
>> For x86_64, specifically, and for dwarf generally, unw_get_proc_info does
>> not set the unwind_info field of the unw_proc_info_t to any non-NULL value,
>> ever. That is because dwarf_make_proc_info passes 0 as the last argument to
>> fetch_proc_info, so that dwarf_extract_proc_info_from_fde ends up not
>> copying unwind info.
>>
>> How can one get the actual unwind info out of unw_get_proc_info?
>
> On current master, tdep_find_proc_info is always passed 1, regardless
> of whatever need_unwind_info is passed:
>
> http://git.savannah.gnu.org/gitweb/?p=libunwind.git;a=blob;f=src/dwarf/Gparser.c;h=ede856e053e5d1716058d2e56165e9feb4a670e6;hb=HEAD#l441
>
> Can you provide a unittest for the issue? It looks like
> unw_get_proc_info is called from both test-ptrace and test-coredump
> already, and at least appears to pass on my machine (excepting the
> -m32 issue just reported).
>
> !DSPAM:10223,58ece8fa39345642720838!
>
>
_______________________________________________
Libunwind-devel mailing list
Libunwind-devel@nongnu.org
https://lists.nongnu.org/mailman/listinfo/libunwind-devel
