Re: Buildbot failure in Wildebeest Builder on whole buildset

[Mark Wielaard <mark@klomp.org>]

[-- Attachment #1: Type: text/plain, Size: 1941 bytes --]

Hi,

On Sat, Nov 28, 2020 at 04:12:25PM +0100, Mark Wielaard wrote:
> On Sat, 2020-11-28 at 14:41 +0100, Mark Wielaard wrote:
> > That is interesting, it only fails on the two 32bit systems.
> > The failure is about the specific error message returned.
> > 
> > FAIL: run-dwflsyms.sh
> > =====================
> > --- dwflsyms.out	2020-11-28 01:17:44.130295202 +0000
> > +++ -	2020-11-28 01:17:44.140327188 +0000
> > @@ -10,7 +10,7 @@
> >     9: NOTYPE	GLOBAL	__kernel_sigtramp_rt64 (12) 0xfffb1af0418
> >    10: NOTYPE	GLOBAL	__kernel_clock_gettime (152) 0xfffb1af0494
> >    11: NOTYPE	GLOBAL	__kernel_get_syscall_map (44) 0xfffb1af05f4
> > -ld64.so.1: No symbol table found
> > +ld64.so.1: Callback returned failure
> >     0: NOTYPE	LOCAL	 (0) 0
> >     1: SECTION	LOCAL	 (0) 0x461b0190
> >     2: SECTION	LOCAL	 (0) 0x461b01a4
> > FAIL run-dwflsyms.sh (exit status: 1)
> > 
> > So there is some subtle difference in the behavior of the
> > 'segment_report_module: Inline consider_notes() into only caller'
> > patch.
> > 
> > I haven't spotted it yet, but I suspect some 'return' from the original
> > function got mistranslated as a continue, break or goto out in the
> > inlined variant.
> > 
> > The specific testcase that fails is:
> > testrun_compare ${abs_builddir}/dwflsyms -e testfile66 --core=testfile66.core
> > 
> > Which is a big endian ppc64 executable and core file.
> 
> The issue can be replicated on x86_64 with:
> 
> $ CXX="g++ -m32" CC="gcc -m32" ~/src/elfutils/configure --enable-maintainer-mode
> $ make -j4
> $ make check TESTS=run-dwflsyms.sh

Found it. There was a small change to calculate note_vaddr (so it
wouldn't clash with the local vaddr) as:

  const size_t note_vaddr = start + offset;

But size_t is only 32bit on the failing systems, which is obviously
not enough for analyzing addresses from a 64bit core file. The fix is
simply to use GElf_Addr instead. Pushed the attached to fix it.

Cheers,

Mark

[-- Attachment #2: 0001-libdwfl-Use-64bit-GElf_Addr-instead-of-size_t-to-cal.patch --]
[-- Type: text/x-diff, Size: 1618 bytes --]

From 609290a61d4f900c65b7e0e273981022a826e4c0 Mon Sep 17 00:00:00 2001
From: Mark Wielaard <mark@klomp.org>
Date: Sun, 29 Nov 2020 01:57:53 +0100
Subject: [PATCH] libdwfl: Use 64bit GElf_Addr instead of size_t to calculate
 address.

size_t is too small on 32 bit systems to analyze a 64 bit core file.

Signed-off-by: Mark Wielaard <mark@klomp.org>
---
 libdwfl/ChangeLog                    | 5 +++++
 libdwfl/dwfl_segment_report_module.c | 2 +-
 2 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/libdwfl/ChangeLog b/libdwfl/ChangeLog
index a5cffc49..67a4d743 100644
--- a/libdwfl/ChangeLog
+++ b/libdwfl/ChangeLog
@@ -1,3 +1,8 @@
+2020-11-28  Mark Wielaard  <mark@klomp.org>
+
+	* dwfl_segment_report_module.c (dwfl_segment_report_module):
+	Use GElf_Addr to calculate note_vaddr instead of size_t.
+
 2020-11-26  Timm Bäder  <tbaeder@redhat.com>
 
 	* dwfl_segment_report_module.c (dwfl_segment_report_module):
diff --git a/libdwfl/dwfl_segment_report_module.c b/libdwfl/dwfl_segment_report_module.c
index 8d99e3bb..ee9cfa2e 100644
--- a/libdwfl/dwfl_segment_report_module.c
+++ b/libdwfl/dwfl_segment_report_module.c
@@ -501,7 +501,7 @@ dwfl_segment_report_module (Dwfl *dwfl, int ndx, const char *name,
 
               /* We calculate from the p_offset of the note segment,
                because we don't yet know the bias for its p_vaddr.  */
-              const size_t note_vaddr = start + offset;
+              const GElf_Addr note_vaddr = start + offset;
               void *data;
               size_t data_size;
               if (read_portion (&read_state, &data, &data_size,
-- 
2.18.4