follow up to fix for 339721 assertion 'check_sibling == sibling' failed in readdwarf3.c ...
The fix committed in revision 14603 is properly fixing the bug 339721.
However, when enabling the dwarf tracing, the DW_FORM_ref_sig8 causes
a segmentation violation, as the tracing code is shared with the
reading code. But the DW_FORM_ref_sig8 reading code is dereferencing
some data structure that is only initialised when --read-var-info=yes.
So, in case DW_FORM_ref_sig8 form reading is done and --read-var-info=no,
then check that we are tracing, and avoid dereferencing the (not initialised)
signature hash table.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@14610 a5019735-40e9-0310-863c-91ae7b9d1cf9
diff --git a/NEWS b/NEWS
index 8c3320b..de86c41 100644
--- a/NEWS
+++ b/NEWS
@@ -35,7 +35,7 @@
339182 ppc64: AvSplat ought to load destination vector register with 16/16
bytes stored prior
339433 ppc64 lxvw4x instruction uses four 32-byte loads
-(not fully fixed yet) 339721 assertion 'check_sibling == sibling' failed in readdwarf3.c ...
+339721 assertion 'check_sibling == sibling' failed in readdwarf3.c ...
n-i-bz Old STABS code is still being compiled, but never used. Remove it.
n-i-bz Fix compilation on distros with glibc < 2.5
diff --git a/coregrind/m_debuginfo/readdwarf3.c b/coregrind/m_debuginfo/readdwarf3.c
index 1a933d9..db924d7 100644
--- a/coregrind/m_debuginfo/readdwarf3.c
+++ b/coregrind/m_debuginfo/readdwarf3.c
@@ -1371,11 +1371,29 @@
TRACE_D3("%x ", (UInt)u8);
work >>= 8;
}
- /* Due to the way that the hash table is constructed, the
- resulting DIE offset here is already "cooked". See
- cook_die_using_form. */
- cts->u.val = lookup_signatured_type (cc->signature_types, signature,
- c->barf);
+
+ /* cc->signature_types is only built/initialised when
+ VG_(clo_read_var_info) is set. In this case,
+ the DW_FORM_ref_sig8 can be looked up.
+ But we can also arrive here when only reading inline info
+ and VG_(clo_trace_symtab) is set. In such a case,
+ we cannot lookup the DW_FORM_ref_sig8, we rather assign
+ a dummy value. This is a kludge, but otherwise,
+ the 'dwarf inline info reader' tracing would have to
+ do type processing/reading. It is better to avoid
+ adding significant 'real' processing only due to tracing. */
+ if (VG_(clo_read_var_info)) {
+ /* Due to the way that the hash table is constructed, the
+ resulting DIE offset here is already "cooked". See
+ cook_die_using_form. */
+ cts->u.val = lookup_signatured_type (cc->signature_types, signature,
+ c->barf);
+ } else {
+ vg_assert (td3);
+ vg_assert (VG_(clo_read_inline_info));
+ TRACE_D3("<not dereferencing signature type>");
+ cts->u.val = 0; /* Assign a dummy/rubbish value */
+ }
cts->szB = sizeof(UWord);
break;
}