strub: sparc64: unbias the stack address [PR112917]

The stack pointer is biased by 2047 bytes on sparc64, so the range it
delimits is way off.  Unbias the addresses returned by
__builtin_stack_address (), so that the strub builtins, inlined or
not, can function correctly.  I've considered introducing a new target
macro, but using STACK_POINTER_OFFSET seems safe, and it enables the
register save areas to be scrubbed as well.

Because of the large fixed-size outgoing args area next to the
register save area on sparc, we still need __strub_leave to not
allocate its own frame, otherwise it won't be able to clear part of
the frame it should.


for  gcc/ChangeLog

	PR middle-end/112917
	* builtins.cc (expand_bultin_stack_address): Add
	STACK_POINTER_OFFSET.
	* doc/extend.texi (__builtin_stack_address): Adjust.

2 files changed, 54 insertions, 3 deletions

diff --git a/gcc/builtins.cc b/gcc/builtins.cc
index 23cc6b6..0f64fee 100644
--- a/gcc/builtins.cc
+++ b/gcc/builtins.cc
@@ -5451,8 +5451,38 @@ expand_builtin_frame_address (tree fndecl, tree exp)
 static rtx
 expand_builtin_stack_address ()
 {
-  return convert_to_mode (ptr_mode, copy_to_reg (stack_pointer_rtx),
-			  STACK_UNSIGNED);
+  rtx ret = convert_to_mode (ptr_mode, copy_to_reg (stack_pointer_rtx),
+			     STACK_UNSIGNED);
+
+  /* Unbias the stack pointer, bringing it to the boundary between the
+     stack area claimed by the active function calling this builtin,
+     and stack ranges that could get clobbered if it called another
+     function.  It should NOT encompass any stack red zone, that is
+     used in leaf functions.
+
+     On SPARC, the register save area is *not* considered active or
+     used by the active function, but rather as akin to the area in
+     which call-preserved registers are saved by callees.  This
+     enables __strub_leave to clear what would otherwise overlap with
+     its own register save area.
+
+     If the address is computed too high or too low, parts of a stack
+     range that should be scrubbed may be left unscrubbed, scrubbing
+     may corrupt active portions of the stack frame, and stack ranges
+     may be doubly-scrubbed by caller and callee.
+
+     In order for it to be just right, the area delimited by
+     @code{__builtin_stack_address} and @code{__builtin_frame_address
+     (0)} should encompass caller's registers saved by the function,
+     local on-stack variables and @code{alloca} stack areas.
+     Accumulated outgoing on-stack arguments, preallocated as part of
+     a function's own prologue, are to be regarded as part of the
+     (caller) function's active area as well, whereas those pushed or
+     allocated temporarily for a call are regarded as part of the
+     callee's stack range, rather than the caller's.  */
+  ret = plus_constant (ptr_mode, ret, STACK_POINTER_OFFSET);
+
+  return force_reg (ptr_mode, ret);
 }
 
 /* Expand a call to builtin function __builtin_strub_enter.  */
diff --git a/gcc/doc/extend.texi b/gcc/doc/extend.texi
index 766e033..4cc3b61b 100644
--- a/gcc/doc/extend.texi
+++ b/gcc/doc/extend.texi
@@ -12669,7 +12669,28 @@ situations.
 @enddefbuiltin
 
 @deftypefn {Built-in Function} {void *} __builtin_stack_address ()
-This function returns the value of the stack pointer register.
+This function returns the stack pointer register, offset by
+@code{STACK_POINTER_OFFSET}.
+
+Conceptually, the returned address returned by this built-in function is
+the boundary between the stack area allocated for use by its caller, and
+the area that could be modified by a function call, that the caller
+could safely zero-out before or after (but not during) the call
+sequence.
+
+Arguments for a callee may be preallocated as part of the caller's stack
+frame, or allocated on a per-call basis, depending on the target, so
+they may be on either side of this boundary.
+
+Even if the stack pointer is biased, the result is not.  The register
+save area on SPARC is regarded as modifiable by calls, rather than as
+allocated for use by the caller function, since it is never in use while
+the caller function itself is running.
+
+Red zones that only leaf functions could use are also regarded as
+modifiable by calls, rather than as allocated for use by the caller.
+This is only theoretical, since leaf functions do not issue calls, but a
+constant offset makes this built-in function more predictable.
 @end deftypefn
 
 @node Stack Scrubbing