Handle erroneous DW_AT_call_return_pc

On PPC64, with the test case included in an earlier patch, we found
that "finish" would still not correctly find the return value via
entry values.

The issue is simple.  The compiler emits:

   0x00000000100032b8 <+28>:	bl      0x1000320c <pck__create_large>
   0x00000000100032bc <+32>:	nop
   0x00000000100032c0 <+36>:	li      r9,42

... but the DWARF says:

    <162a>   DW_AT_call_return_pc: 0x100032c0

That is, the declared return PC is one instruction past the actual
return PC.

This patch adds a new arch hook to handle this scenario, and
implements it for PPC64.  Some care is taken so that GDB will continue
to work if this compiler bug is fixed.  A GCC patch is here:

    https://gcc.gnu.org/pipermail/gcc-patches/2023-March/613336.html

No check for 'nop' is done, as subsequent discussion revealed that the
linker might replace this with another instruction.

6 files changed, 90 insertions, 0 deletions

diff --git a/gdb/arch-utils.c b/gdb/arch-utils.c
index 997a292e9ef..1de5981765a 100644
--- a/gdb/arch-utils.c
+++ b/gdb/arch-utils.c
@@ -1105,6 +1105,12 @@ default_dwarf2_omit_typedef_p (struct type *target_type, const char *producer,
   return false;
 }
 
+static CORE_ADDR
+default_update_call_site_pc (struct gdbarch *gdbarch, CORE_ADDR pc)
+{
+  return pc;
+}
+
 /* Non-zero if we want to trace architecture code.  */
 
 #ifndef GDBARCH_DEBUG
diff --git a/gdb/gdbarch-gen.h b/gdb/gdbarch-gen.h
index 56b0a917168..7f83bf4e214 100644
--- a/gdb/gdbarch-gen.h
+++ b/gdb/gdbarch-gen.h
@@ -476,6 +476,19 @@ typedef bool (gdbarch_dwarf2_omit_typedef_p_ftype) (struct type *target_type, co
 extern bool gdbarch_dwarf2_omit_typedef_p (struct gdbarch *gdbarch, struct type *target_type, const char *producer, const char *name);
 extern void set_gdbarch_dwarf2_omit_typedef_p (struct gdbarch *gdbarch, gdbarch_dwarf2_omit_typedef_p_ftype *dwarf2_omit_typedef_p);
 
+/* Update PC when trying to find a call site.  This is useful on
+   architectures where the call site PC, as reported in the DWARF, can be
+   incorrect for some reason.
+
+   The passed-in PC will be an address in the inferior.  GDB will have
+   already failed to find a call site at this PC.  This function may
+   simply return its parameter if it thinks that should be the correct
+   address. */
+
+typedef CORE_ADDR (gdbarch_update_call_site_pc_ftype) (struct gdbarch *gdbarch, CORE_ADDR pc);
+extern CORE_ADDR gdbarch_update_call_site_pc (struct gdbarch *gdbarch, CORE_ADDR pc);
+extern void set_gdbarch_update_call_site_pc (struct gdbarch *gdbarch, gdbarch_update_call_site_pc_ftype *update_call_site_pc);
+
 /* Return true if the return value of function is stored in the first hidden
    parameter.  In theory, this feature should be language-dependent, specified
    by language and its ABI, such as C++.  Unfortunately, compiler may
diff --git a/gdb/gdbarch.c b/gdb/gdbarch.c
index 9406458189f..30199a0b03d 100644
--- a/gdb/gdbarch.c
+++ b/gdb/gdbarch.c
@@ -115,6 +115,7 @@ struct gdbarch
   gdbarch_return_value_as_value_ftype *return_value_as_value = default_gdbarch_return_value;
   gdbarch_get_return_buf_addr_ftype *get_return_buf_addr = default_get_return_buf_addr;
   gdbarch_dwarf2_omit_typedef_p_ftype *dwarf2_omit_typedef_p = default_dwarf2_omit_typedef_p;
+  gdbarch_update_call_site_pc_ftype *update_call_site_pc = default_update_call_site_pc;
   gdbarch_return_in_first_hidden_param_p_ftype *return_in_first_hidden_param_p = default_return_in_first_hidden_param_p;
   gdbarch_skip_prologue_ftype *skip_prologue = nullptr;
   gdbarch_skip_main_prologue_ftype *skip_main_prologue = nullptr;
@@ -372,6 +373,7 @@ verify_gdbarch (struct gdbarch *gdbarch)
     log.puts ("\n\treturn_value_as_value");
   /* Skip verify of get_return_buf_addr, invalid_p == 0 */
   /* Skip verify of dwarf2_omit_typedef_p, invalid_p == 0 */
+  /* Skip verify of update_call_site_pc, invalid_p == 0 */
   /* Skip verify of return_in_first_hidden_param_p, invalid_p == 0 */
   if (gdbarch->skip_prologue == 0)
     log.puts ("\n\tskip_prologue");
@@ -794,6 +796,9 @@ gdbarch_dump (struct gdbarch *gdbarch, struct ui_file *file)
 	      "gdbarch_dump: dwarf2_omit_typedef_p = <%s>\n",
 	      host_address_to_string (gdbarch->dwarf2_omit_typedef_p));
   gdb_printf (file,
+	      "gdbarch_dump: update_call_site_pc = <%s>\n",
+	      host_address_to_string (gdbarch->update_call_site_pc));
+  gdb_printf (file,
 	      "gdbarch_dump: return_in_first_hidden_param_p = <%s>\n",
 	      host_address_to_string (gdbarch->return_in_first_hidden_param_p));
   gdb_printf (file,
@@ -2639,6 +2644,23 @@ set_gdbarch_dwarf2_omit_typedef_p (struct gdbarch *gdbarch,
   gdbarch->dwarf2_omit_typedef_p = dwarf2_omit_typedef_p;
 }
 
+CORE_ADDR
+gdbarch_update_call_site_pc (struct gdbarch *gdbarch, CORE_ADDR pc)
+{
+  gdb_assert (gdbarch != NULL);
+  gdb_assert (gdbarch->update_call_site_pc != NULL);
+  if (gdbarch_debug >= 2)
+    gdb_printf (gdb_stdlog, "gdbarch_update_call_site_pc called\n");
+  return gdbarch->update_call_site_pc (gdbarch, pc);
+}
+
+void
+set_gdbarch_update_call_site_pc (struct gdbarch *gdbarch,
+				 gdbarch_update_call_site_pc_ftype update_call_site_pc)
+{
+  gdbarch->update_call_site_pc = update_call_site_pc;
+}
+
 int
 gdbarch_return_in_first_hidden_param_p (struct gdbarch *gdbarch, struct type *type)
 {
diff --git a/gdb/gdbarch_components.py b/gdb/gdbarch_components.py
index 62cd430cbae..07010b13311 100644
--- a/gdb/gdbarch_components.py
+++ b/gdb/gdbarch_components.py
@@ -929,6 +929,23 @@ Return 0 by default""",
 
 Method(
     comment="""
+Update PC when trying to find a call site.  This is useful on
+architectures where the call site PC, as reported in the DWARF, can be
+incorrect for some reason.
+
+The passed-in PC will be an address in the inferior.  GDB will have
+already failed to find a call site at this PC.  This function may
+simply return its parameter if it thinks that should be the correct
+address.""",
+    type="CORE_ADDR",
+    name="update_call_site_pc",
+    params=[("CORE_ADDR", "pc")],
+    predefault="default_update_call_site_pc",
+    invalid=False,
+)
+
+Method(
+    comment="""
 Return true if the return value of function is stored in the first hidden
 parameter.  In theory, this feature should be language-dependent, specified
 by language and its ABI, such as C++.  Unfortunately, compiler may
diff --git a/gdb/rs6000-tdep.c b/gdb/rs6000-tdep.c
index afeea02a84d..ee18b495d4c 100644
--- a/gdb/rs6000-tdep.c
+++ b/gdb/rs6000-tdep.c
@@ -7471,6 +7471,24 @@ rs6000_program_breakpoint_here_p (gdbarch *gdbarch, CORE_ADDR address)
   return false;
 }
 
+/* Implement the update_call_site_pc arch hook.  */
+
+static CORE_ADDR
+ppc64_update_call_site_pc (struct gdbarch *gdbarch, CORE_ADDR pc)
+{
+  /* Some versions of GCC emit:
+
+     .  bl function
+     .  nop
+     .  ...
+
+     but emit DWARF where the DW_AT_call_return_pc points to
+     instruction after the 'nop'.  Note that while the compiler emits
+     a 'nop', the linker might put some other instruction there -- so
+     we just unconditionally check the next instruction.  */
+  return pc + 4;
+}
+
 /* Initialize the current architecture based on INFO.  If possible, re-use an
    architecture from ARCHES, which is a list of architectures already created
    during this debugging session.
@@ -8257,6 +8275,7 @@ rs6000_gdbarch_init (struct gdbarch_info info, struct gdbarch_list *arches)
       set_gdbarch_return_value (gdbarch, ppc64_sysv_abi_return_value);
       set_gdbarch_get_return_buf_addr (gdbarch,
 				       ppc64_sysv_get_return_buf_addr);
+      set_gdbarch_update_call_site_pc (gdbarch, ppc64_update_call_site_pc);
     }
   else
     set_gdbarch_return_value (gdbarch, ppc_sysv_abi_return_value);
diff --git a/gdb/symtab.c b/gdb/symtab.c
index 9e9798676cb..6b6905d5489 100644
--- a/gdb/symtab.c
+++ b/gdb/symtab.c
@@ -349,6 +349,19 @@ compunit_symtab::find_call_site (CORE_ADDR pc) const
   struct call_site call_site_local (unrelocated_pc, nullptr, nullptr);
   void **slot
     = htab_find_slot (m_call_site_htab, &call_site_local, NO_INSERT);
+  if (slot != nullptr)
+    return (call_site *) *slot;
+
+  /* See if the arch knows another PC we should try.  On some
+     platforms, GCC emits a DWARF call site that is offset from the
+     actual return location.  */
+  struct gdbarch *arch = objfile ()->arch ();
+  CORE_ADDR new_pc = gdbarch_update_call_site_pc (arch, pc);
+  if (pc == new_pc)
+    return nullptr;
+
+  call_site new_call_site_local (new_pc - delta, nullptr, nullptr);
+  slot = htab_find_slot (m_call_site_htab, &new_call_site_local, NO_INSERT);
   if (slot == nullptr)
     return nullptr;