[Bug target/126394] arm: stack misalignment when leaf procedures read tp via __aeabi_read_tp

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126394

Kalvis Duckmanton <kalvisd at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |kalvisd at gmail dot com

--- Comment #1 from Kalvis Duckmanton <kalvisd at gmail dot com> ---
The following patch should allow leaf functions using thread-local storage to
be detected (and thus allow the stack to be aligned as needed)

diff --git a/external/gpl3/gcc/dist/gcc/config/arm/arm.cc
b/external/gpl3/gcc/dist/gcc/config/arm/arm.cc
index d3f7a08f99c7..fef0dbf4d304 100644
--- a/external/gpl3/gcc/dist/gcc/config/arm/arm.cc
+++ b/external/gpl3/gcc/dist/gcc/config/arm/arm.cc
@@ -9529,6 +9529,8 @@ arm_legitimize_address (rtx x, rtx orig_x, machine_mode
mode)

       x = legitimize_tls_address (x, NULL_RTX);

+      cfun->machine->uses_tls = true;
+
       if (addend)
        {
          x = gen_rtx_PLUS (SImode, x, addend);
@@ -21981,6 +21983,9 @@ arm_output_function_prologue (FILE *f)
               frame_pointer_needed,
               cfun->machine->uses_anonymous_args);

+  asm_fprintf (f, "\t%@ uses_tls = %d, is_leaf = %d\n",
+              cfun->machine->uses_tls, crtl->is_leaf);
+
   if (cfun->machine->lr_save_eliminated)
     asm_fprintf (f, "\t%@ link register save eliminated.\n");

@@ -23162,10 +23167,10 @@ arm_compute_frame_layout (void)
         think of this as a leaf routine.  Even if __aeabi_read_tp
         itself doesn't use the stack, resolving the symbol may take a
         detour through a procedure call to the dynamic linker.  We
-        should really enforce alignment only if the procedure actually
-        uses __aeabi_read_tp (load_tp_soft*) but I don't know how to
-        query that here.  */
-      && !TARGET_SOFT_TP)
+        enforce alignment only if the procedure actually uses
+        __aeabi_read_tp (load_tp_soft*).  */
+      && ! TARGET_SOFT_TP
+      && ! cfun->machine->uses_tls)
     {
       offsets->outgoing_args = offsets->soft_frame;
       offsets->locals_base = offsets->soft_frame;
diff --git a/external/gpl3/gcc/dist/gcc/config/arm/arm.h
b/external/gpl3/gcc/dist/gcc/config/arm/arm.h
index d18077b32847..e71c1090da6f 100644
--- a/external/gpl3/gcc/dist/gcc/config/arm/arm.h
+++ b/external/gpl3/gcc/dist/gcc/config/arm/arm.h
@@ -1651,6 +1651,8 @@ typedef struct GTY(()) machine_function
   /* Set to 1 when pointer authentication operation uses value of SP other
      than the incoming stack pointer value.  */
   int pacspval_needed;
+  /* Set to 1 if this function accesses thread-local storage */
+  int uses_tls;
 }
 machine_function;
 #endif
diff --git a/external/gpl3/gcc/dist/gcc/config/arm/arm.md
b/external/gpl3/gcc/dist/gcc/config/arm/arm.md
index 13a8fbf7a147..83a3727d247d 100644
--- a/external/gpl3/gcc/dist/gcc/config/arm/arm.md
+++ b/external/gpl3/gcc/dist/gcc/config/arm/arm.md
@@ -6523,7 +6523,10 @@
      data and load its address to a register.  Offsets have been split off
      already.  */
   if (arm_tls_referenced_p (operands[1]))
-    operands[1] = legitimize_tls_address (operands[1], tmp);
+    {
+      operands[1] = legitimize_tls_address (operands[1], tmp);
+      cfun->machine->uses_tls = true;
+    }
   else if (flag_pic
           && (CONSTANT_P (operands[1])
               || symbol_mentioned_p (operands[1])