Re: [PATCH] Fix debugging of stripped PIE executables with padded PT_TLS

[Simon Marchi <simark@simark.ca>]

On 2018-08-19 11:04, Simon Marchi wrote:
> On 2018-08-16 2:35 p.m., Michael Spang via gdb-patches wrote:
>> From: Michael Spang <spang@google.com>
>> 
>> Certain PIE executables produced by gold cannot be debugged by gdb 
>> after
>> being stripped. GDB requires program headers of PIE executables to 
>> match,
>> and those checks may fail due to adjustments made during stripping.
>> 
>> One case of this occurs because strip recomputes the memsz of PT_TLS 
>> and
>> does not add alignment, while gold does. This is another variant of PR
>> 11786, so apply the same fix of relaxing the program header matching.
> 
> Thanks, as far as I understand this patch is correct.  I pushed it with 
> some
> minor changes outlined below.

Hmm, the new test fails on the buildbot, see here for example:

https://gdb-build.sergiodj.net/builders/Fedora-x86_64-m64/builds/10711

On the buildbot:

x/i $pc^M
=> 0x55555555467d:      mov    $0x0,%eax^M
(gdb) FAIL: gdb.base/gcore-tls-pie.exp: x/i $pc
frame^M
#0  0x000055555555467d in ?? ()^M
(gdb) FAIL: gdb.base/gcore-tls-pie.exp: unstripped + core ok


on my machine:

x/i $pc^M
=> 0x55555555460d <break_here+4>:       mov    $0x0,%eax^M
(gdb) PASS: gdb.base/gcore-tls-pie.exp: x/i $pc
frame^M
#0  break_here () at 
/home/simark/src/binutils-gdb/gdb/testsuite/gdb.base/gcore-tls-pie.c:28^M
28        *(int *) 0 = 0;^M
(gdb) PASS: gdb.base/gcore-tls-pie.exp: unstripped + core ok


Does anybody see the failure locally?

Simon