https://sourceware.org/bugzilla/show_bug.cgi?id=34366
Bug ID: 34366
Summary: readelf: spurious "Hole and overlap detection" warning
for DW_LLE_startx_endx/startx_length loclists with
DW_AT_GNU_locviews
Product: binutils
Version: 2.46
Status: UNCONFIRMED
Severity: normal
Priority: P2
Component: binutils
Assignee: unassigned at sourceware dot org
Reporter: kamil.bucki at nokia dot com
Target Milestone: ---
Created attachment 16823
--> https://sourceware.org/bugzilla/attachment.cgi?id=16823&action=edit
gas-only reproducer (locview-startx.s + build/reproduce scripts)
SUMMARY
=======
When readelf/objdump dumps a DWARF 5 .debug_loclists location list that uses
an indexed bounded entry -- DW_LLE_startx_endx (0x02) or DW_LLE_startx_length
(0x03) -- and the variable also has a GNU location-view list
(DW_AT_GNU_locviews), a spurious warning is printed:
readelf: Warning: Hole and overlap detection requires adjacent view lists
and loclists.
The DWARF is well-formed; this is a false positive. The range is printed
WITHOUT its "views at ... for:" annotation (unlike offset_pair / start_end /
start_length entries), and the preceding view pair is left unread, so the
adjacency heuristic at the end of the per-list loop reports a view/loclist
mismatch.
ROOT CAUSE
==========
In display_loclists_list (binutils/dwarf.c, ~line 8177), the block that reads
and attaches the location-view pair is guarded by a whitelist of entry kinds
that only lists DW_LLE_offset_pair, DW_LLE_start_end and DW_LLE_start_length --
even though the switch in the same function already decodes DW_LLE_startx_endx
and DW_LLE_startx_length as bounded ranges (since PR 28981):
/* binutils/dwarf.c, display_loclists_list() */
if (vstart && (llet == DW_LLE_offset_pair
|| llet == DW_LLE_start_end
|| llet == DW_LLE_start_length))
{
off = offset + (vstart - *start_ptr);
...
}
Because DW_LLE_startx_endx / DW_LLE_startx_length are not in this guard, their
view pair is never consumed nor printed, and the adjacency heuristic at the end
of the per-list loop then reports a view/loclist mismatch.
Adding the two indexed bounded forms to that guard fixes it:
if (vstart && (llet == DW_LLE_offset_pair
|| llet == DW_LLE_start_end
|| llet == DW_LLE_start_length
|| llet == DW_LLE_startx_endx
|| llet == DW_LLE_startx_length))
This is not target specific: it reproduces on any DWARF 5 object whose producer
emits startx_* loclist entries together with location views.
REPRODUCTION
============
Minimal hand-crafted DWARF 5 reproducer assembled with gas only (no GCC,
no BOLT). One DW_LLE_startx_endx entry plus a DW_AT_GNU_locviews view list.
as -o locview-startx.o locview-startx.s
readelf --debug-dump=loc locview-startx.o
OBSERVED (before fix, readelf 2.46)
===================================
A standalone view pair at 0x0c, followed by a startx_endx range at 0x0e with
NO "views at" annotation, then the warning:
Contents of the .debug_loclists section:
Table at Offset 0
Length: 0x10
DWARF version: 5
Address size: 4
Segment size: 0
Offset entries: 0
Offset Begin End Expression
0000000c v0000000 v0000000 location view pair
0000000e 00000000 00000000 (DW_OP_reg1 (rdx)) (start == end)
00000013 <End of list>
readelf: Warning: Hole and overlap detection requires adjacent view lists
and loclists.
EXPECTED (after the attached patch)
===================================
The view pair is attached and no warning is printed:
0000000c v0000000 v0000000 location view pair
0000000e v0000000 v0000000 views at 0000000c for:
00000000 00000000 (DW_OP_reg1 (rdx))
00000013 <End of list>
PATCH
=====
diff --git a/binutils/dwarf.c b/binutils/dwarf.c
--- a/binutils/dwarf.c
+++ b/binutils/dwarf.c
@@ -8177,7 +8177,9 @@ display_loclists_list (struct dwarf_section * section,
if (vstart && (llet == DW_LLE_offset_pair
|| llet == DW_LLE_start_end
- || llet == DW_LLE_start_length))
+ || llet == DW_LLE_start_length
+ || llet == DW_LLE_startx_endx
+ || llet == DW_LLE_startx_length))
{
off = offset + (vstart - *start_ptr);
VERSIONS AFFECTED
=================
- Released readelf 2.46 prints the warning on the attached reproducer.
--
You are receiving this mail because:
You are on the CC list for the bug.
