[gdb/symtab] Fix race condition in just_read_cus

Race condition between:
...
  Read of size 8 at 0x7b640001f2c8 by thread T1:
    #0 std::vector<dwarf2_per_cu_data*, std::allocator<dwarf2_per_cu_data*>
    >::push_back(dwarf2_per_cu_data* const&)
    /usr/include/c++/12/bits/stl_vector.h:1278 (gdb+0x888ad6)
    #1 process_full_comp_unit gdb/dwarf2/read.c:8451 (gdb+0x8397d0)
    #2 process_queue_item gdb/dwarf2/read.c:7592 (gdb+0x8359f5)
...
and:
...
  Previous write of size 8 at 0x7b640001f2c8 by main thread:
    #0 std::vector<dwarf2_per_cu_data*, std::allocator<dwarf2_per_cu_data*>
    >::push_back(dwarf2_per_cu_data* const&)
    /usr/include/c++/12/bits/stl_vector.h:1283 (gdb+0x888b52)
    #1 process_full_comp_unit gdb/dwarf2/read.c:8451 (gdb+0x8397d0)
    #2 process_queue_item gdb/dwarf2/read.c:7592 (gdb+0x8359f5)
...

Fix this by doing the just_read_cus push_back in a lock.
This commit is contained in:
Tom de Vries
2022-07-15 18:46:03 +02:00
parent a4f4bbc353
commit c2ae16cc58

View File

@@ -8463,8 +8463,14 @@ process_full_comp_unit (dwarf2_cu *cu, enum language pretend_language)
per_objfile->set_symtab (cu->per_cu, cust);
/* Push it for inclusion processing later. */
per_objfile->per_bfd->just_read_cus.push_back (cu->per_cu);
#if CXX_STD_THREAD
{
static std::mutex just_read_cus_lock;
std::lock_guard<std::mutex> guard (just_read_cus_lock);
#endif
/* Push it for inclusion processing later. */
per_objfile->per_bfd->just_read_cus.push_back (cu->per_cu);
}
/* Not needed any more. */
cu->reset_builder ();