core: Initialize cu->node with INIT_LIST_HEAD()

In cu__new() zalloc() is used defensively, and that helped catch this
problem where we assume that a cu us in the cus list of cu instances,
but that is not the case when we use cus__merge_and_process_cu(), for
instance when loading files created by clang with LTO, as reported by
Peter Jung and narrowed down by Nathan Chancellor.

If we use INIT_LIST_HEAD() in cu__new() to initialize cu->node, which is
what we do with other lists and nodes there, then the unconditional
removal using list_del_init() will be a no-op and removing something not
on the cus list of cu instances will not cause problems, just keep an
unconsistent cus->nr_entries field.

So lets just have this fix in first, keeping Nathan's Tested-by and then
do the a bit more involved fix of either adding that cu to the cus list
or checking at removal time if it is there.

  Program received signal SIGSEGV, Segmentation fault.
  0x00007ffff7f1e13e in __list_del (prev=0x0, next=0x0) at /home/acme/git/pahole/list.h:106
  106		next->prev = prev;
  (gdb) bt
  #0  0x00007ffff7f1e13e in __list_del (prev=0x0, next=0x0) at /home/acme/git/pahole/list.h:106
  #1  0x00007ffff7f1e176 in list_del_init (entry=0x417980) at /home/acme/git/pahole/list.h:165
  #2  0x00007ffff7f1f8f9 in __cus__remove (cus=0x4142a0, cu=0x417980) at /home/acme/git/pahole/dwarves.c:527
  #3  0x00007ffff7f1f92b in cus__remove (cus=0x4142a0, cu=0x417980) at /home/acme/git/pahole/dwarves.c:533
  #4  0x00007ffff7f3d01c in cus__finalize (cus=0x4142a0, cu=0x417980, conf=0x4133c0 <conf_load>, thr_data=0x0)
      at /home/acme/git/pahole/dwarf_loader.c:3040
  #5  0x00007ffff7f3e05c in cus__merge_and_process_cu (cus=0x4142a0, conf=0x4133c0 <conf_load>, mod=0x415cf0, dw=0x416110, elf=0x414380,
      filename=0x7fffffffe3f7 "cast_common.ko", build_id=0x416680 "\265D\371U\213\373u|\037\250\242\032\271\365⒜]y\023", build_id_len=20,
      type_dcu=0x0) at /home/acme/git/pahole/dwarf_loader.c:3482
  #6  0x00007ffff7f3e218 in cus__load_module (cus=0x4142a0, conf=0x4133c0 <conf_load>, mod=0x415cf0, dw=0x416110, elf=0x414380,
      filename=0x7fffffffe3f7 "cast_common.ko") at /home/acme/git/pahole/dwarf_loader.c:3521
  #7  0x00007ffff7f3e396 in cus__process_dwflmod (dwflmod=0x415cf0, userdata=0x415d00, name=0x415ea0 "cast_common.ko", base=65536,
      arg=0x7fffffffde40) at /home/acme/git/pahole/dwarf_loader.c:3581
  #8  0x00007ffff7eb4609 in dwfl_getmodules (dwfl=0x414300, callback=0x7ffff7f3e2ec <cus__process_dwflmod>, arg=0x7fffffffde40, offset=0)
      at ../libdwfl/dwfl_getmodules.c:86
  #9  0x00007ffff7f3e4c5 in cus__process_file (cus=0x4142a0, conf=0x4133c0 <conf_load>, fd=3, filename=0x7fffffffe3f7 "cast_common.ko")
      at /home/acme/git/pahole/dwarf_loader.c:3647
  #10 0x00007ffff7f3e5cd in dwarf__load_file (cus=0x4142a0, conf=0x4133c0 <conf_load>, filename=0x7fffffffe3f7 "cast_common.ko")
      at /home/acme/git/pahole/dwarf_loader.c:3684
  #11 0x00007ffff7f232df in cus__load_file (cus=0x4142a0, conf=0x4133c0 <conf_load>, filename=0x7fffffffe3f7 "cast_common.ko")
      at /home/acme/git/pahole/dwarves.c:2134
  #12 0x00007ffff7f23e8b in cus__load_files (cus=0x4142a0, conf=0x4133c0 <conf_load>, filenames=0x7fffffffe0f0)
      at /home/acme/git/pahole/dwarves.c:2637
  #13 0x000000000040aec0 in main (argc=2, argv=0x7fffffffe0e8) at /home/acme/git/pahole/pahole.c:3805
  (gdb) fr 1
  #1  0x00007ffff7f1e176 in list_del_init (entry=0x417980) at /home/acme/git/pahole/list.h:165
  165		__list_del(entry->prev, entry->next);
  (gdb) p entry
  $1 = (struct list_head *) 0x417980
  (gdb) p entry->next
  $2 = (struct list_head *) 0x0
  (gdb) p entry->prev
  $3 = (struct list_head *) 0x0

Closes: #53
Closes: https://gitlab.archlinux.org/archlinux/packaging/packages/pahole/-/issues/1
Tested-by: Nathan Chancellor <nathan@kernel.org>
Link: https://lore.kernel.org/all/20240617210810.GA1877676@thelio-3990X
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

Author: acmel

dwarves.c

@@ -739,6 +739,7 @@ struct cu *cu__new(const char *name, uint8_t addr_size,
 		cu->dfops	= NULL;
 		INIT_LIST_HEAD(&cu->tags);
 		INIT_LIST_HEAD(&cu->tool_list);
+		INIT_LIST_HEAD(&cu->node);
 
 		cu->addr_size = addr_size;
 		cu->extra_dbg_info = 0;