[PATCH 5/5] [gdb/symtab] Fix data race on per_cu->lang

[Tom de Vries <tdevries@suse.de>]

When building gdb with -fsanitize=thread and gcc 12, and running test-case
gdb.dwarf2/inlined_subroutine-inheritance.exp, we run into a data race
between thread T4 and the main thread:
...
  Write of size 1 at 0x7b200000308d:
    #0 prepare_one_comp_unit gdb/dwarf2/read.c:23586 (gdb+0x86f859)
...
which is here:
...
    cu->per_cu->lang = dwarf_lang_to_enum_language (attr->constant_value (0));
...

Both writes are called from the parallel for in dwarf2_build_psymtabs_hard,
this one directly:
...
    #1 process_psymtab_comp_unit gdb/dwarf2/read.c:6812 (gdb+0x830912)
    #2 operator() gdb/dwarf2/read.c:7102 (gdb+0x831902)
    #3 operator() gdb/../gdbsupport/parallel-for.h:171 (gdb+0x8723a8)
...
and this one when handling cross-CU refs:
...
    #1 cooked_indexer::ensure_cu_exists(cutu_reader*, dwarf2_per_objfile*, \
    sect_offset, bool, bool) gdb/dwarf2/read.c:17973 (gdb+0x85c522)
    #2 cooked_indexer::scan_attributes(dwarf2_per_cu_data*, cutu_reader*, \
    unsigned char const*, unsigned char const*, abbrev_info const*, \
    char const**, char const**, enum_flags<cooked_index_flag_enum>*, \
    sect_offset*, cooked_index_entry const**, unsigned long*, bool) \
    gdb/dwarf2/read.c:18148 (gdb+0x85d079)
    #3 cooked_indexer::index_dies(cutu_reader*, unsigned char const*, \
    cooked_index_entry const*, bool) gdb/dwarf2/read.c:18327 (gdb+0x85df65)
    #4 cooked_indexer::make_index(cutu_reader*) gdb/dwarf2/read.c:18450 \
    (gdb+0x85e72c)
    #5 process_psymtab_comp_unit gdb/dwarf2/read.c:6816 (gdb+0x8309c1)
    #6 operator() gdb/dwarf2/read.c:7102 (gdb+0x831902)
    #7 operator() gdbsupport/parallel-for.h:163 (gdb+0x872346)
...

Fix this by guarding the write with a lock.
---
 gdb/dwarf2/read.c | 31 ++++++++++++++++++++++++++-----
 1 file changed, 26 insertions(+), 5 deletions(-)

diff --git a/gdb/dwarf2/read.c b/gdb/dwarf2/read.c
index 67702d43aff..a36f25f4e62 100644
--- a/gdb/dwarf2/read.c
+++ b/gdb/dwarf2/read.c
@@ -6766,6 +6766,10 @@ class cooked_indexer
   std::vector<deferred_entry> m_deferred_entries;
 };
 
+#if CXX_STD_THREAD
+static std::mutex per_cu_long_lock;
+#endif
+
 /* Subroutine of dwarf2_build_psymtabs_hard to simplify it.
    Process compilation unit THIS_CU for a psymtab.  */
 
@@ -23567,6 +23571,8 @@ prepare_one_comp_unit (struct dwarf2_cu *cu, struct die_info *comp_unit_die,
 
   /* Set the language we're debugging.  */
   attr = dwarf2_attr (comp_unit_die, DW_AT_language, cu);
+
+  enum language lang;
   if (cu->producer != nullptr
       && strstr (cu->producer, "IBM XL C for OpenCL") != NULL)
     {
@@ -23574,19 +23580,34 @@ prepare_one_comp_unit (struct dwarf2_cu *cu, struct die_info *comp_unit_die,
 	 attribute is not standardised yet.  As a workaround for the
 	 language detection we fall back to the DW_AT_producer
 	 string.  */
-      cu->per_cu->lang = language_opencl;
+      lang = language_opencl;
     }
   else if (cu->producer != nullptr
 	   && strstr (cu->producer, "GNU Go ") != NULL)
     {
       /* Similar hack for Go.  */
-      cu->per_cu->lang = language_go;
+      lang = language_go;
     }
   else if (attr != nullptr)
-    cu->per_cu->lang = dwarf_lang_to_enum_language (attr->constant_value (0));
+    lang = dwarf_lang_to_enum_language (attr->constant_value (0));
   else
-    cu->per_cu->lang = pretend_language;
-  cu->language_defn = language_def (cu->per_cu->lang);
+    lang = pretend_language;
+
+  {
+#if CXX_STD_THREAD
+    std::lock_guard<std::mutex> guard (per_cu_long_lock);
+#endif
+    /* Assign cu->per_cu->lang lazily.  Note: we're not doing here:
+         if (cu->per_cu->lang == language_unknown)
+           cu->per_cu->lang = lang;
+         else
+           gdb_assert (cu->per_cu->lang == lang);
+       because language may go from unknown to minimal to c.  */
+    if (cu->per_cu->lang != lang)
+      cu->per_cu->lang = lang;
+  }
+
+  cu->language_defn = language_def (lang);
 }
 
 /* See read.h.  */
-- 
2.35.3