[rpms/gcc] rhel-f41-base: - fix -fstack-clash-protection codegen issue on 32 bit x86

[rpms/gcc] rhel-f41-base: - fix -fstack-clash-protection codegen issue on 32 bit x86

[Jeff Law]

          A new commit has been pushed.

          Repo   : rpms/gcc
          Branch : rhel-f41-base
          Commit : fa22fac534c6eb5c499d3fbbaa728a7ff1111b83
          Author : Jeff Law <law@redhat.com>
          Date   : 2018-02-01T09:37:28-07:00
          Stats  : +187/-1 in 2 file(s)
          URL    : https://src.fedoraproject.org/rpms/gcc/c/fa22fac534c6eb5c499d3fbbaa728a7ff1111b83?branch=rhel-f41-base

          Log:
          - fix -fstack-clash-protection codegen issue on 32 bit x86
(#1540221, PR target/84128)

---
diff --git a/gcc.spec b/gcc.spec
index 21eb279..889d67d 100644
--- a/gcc.spec
+++ b/gcc.spec
@@ -4,7 +4,7 @@
 %global gcc_major 8
 # Note, gcc_release must be integer, if you want to add suffixes to
 # %{release}, append them after %{gcc_release} on Release: line.
-%global gcc_release 0.8
+%global gcc_release 0.9
 %global nvptx_tools_gitrev c28050f60193b3b95a18866a96f03334e874e78f
 %global nvptx_newlib_gitrev aadc8eb0ec43b7cd0dd2dfb484bae63c8b05ef24
 %global _unpackaged_files_terminate_build 0
@@ -236,6 +236,7 @@ Patch10: gcc8-foffload-default.patch
 Patch11: gcc8-Wno-format-security.patch
 Patch12: gcc8-rh1512529-aarch64.patch
 Patch13: gcc8-pr84146.patch
+Patch14: gcc8-pr84128.patch
 
 Patch1000: nvptx-tools-no-ptxas.patch
 Patch1001: nvptx-tools-build.patch
@@ -775,6 +776,7 @@ to NVidia PTX capable devices if available.
 %patch11 -p0 -b .Wno-format-security~
 %patch12 -p0 -b .rh1512529-aarch64~
 %patch13 -p0 -b .pr84146~
+%patch14 -p0 -b .pr84128~
 
 cd nvptx-tools-%{nvptx_tools_gitrev}
 %patch1000 -p1 -b .nvptx-tools-no-ptxas~
@@ -3036,6 +3038,10 @@ fi
 %endif
 
 %changelog
+* Thu Feb 1 2018 Jeff Law <law@redhat.com> 8.0.1-0.9
+- fix -fstack-clash-protection codegen issue on 32 bit x86
+  (#1540221, PR target/84128)
+
 * Wed Jan 31 2018 Jakub Jelinek <jakub@redhat.com> 8.0.1-0.8
 - update from the trunk
   - PRs c++/83993, c++/84092, c++/84138, c/84100, fortran/84088,

diff --git a/gcc8-pr84128.patch b/gcc8-pr84128.patch
new file mode 100644
index 0000000..454fc09
--- /dev/null
+++ b/gcc8-pr84128.patch
@@ -0,0 +1,180 @@
+
+	PR target/84128
+	* config/i386/i386.c (release_scratch_register_on_entry): Add new
+	OFFSET and RELEASE_VIA_POP arguments.  Use SP+OFFSET to restore
+	the scratch if RELEASE_VIA_POP is false.
+	(ix86_adjust_stack_and_probe_stack_clash): Un-constify SIZE.
+	If we have to save a temporary register, decrement SIZE appropriately.
+	Pass new arguments to release_scratch_register_on_entry.
+	(ix86_adjust_stack_and_probe): Likewise.
+	(ix86_emit_probe_stack_range): Pass new arguments to
+	release_scratch_register_on_entry.
+
+	PR target/84128
+	* gcc.target/i386/pr84128.c: New test.
+
+diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
+index fef34a1..3196ac4 100644
+--- gcc/config/i386/i386.c
++++ gcc/config/i386/i386.c
+@@ -12567,22 +12567,39 @@ get_scratch_register_on_entry (struct scratch_reg *sr)
+     }
+ }
+ 
+-/* Release a scratch register obtained from the preceding function.  */
++/* Release a scratch register obtained from the preceding function.
++
++   If RELEASE_VIA_POP is true, we just pop the register off the stack
++   to release it.  This is what non-Linux systems use with -fstack-check.
++
++   Otherwise we use OFFSET to locate the saved register and the
++   allocated stack space becomes part of the local frame and is
++   deallocated by the epilogue.  */
+ 
+ static void
+-release_scratch_register_on_entry (struct scratch_reg *sr)
++release_scratch_register_on_entry (struct scratch_reg *sr, HOST_WIDE_INT offset,
++				   bool release_via_pop)
+ {
+   if (sr->saved)
+     {
+-      struct machine_function *m = cfun->machine;
+-      rtx x, insn = emit_insn (gen_pop (sr->reg));
++      if (release_via_pop)
++	{
++	  struct machine_function *m = cfun->machine;
++	  rtx x, insn = emit_insn (gen_pop (sr->reg));
+ 
+-      /* The RTX_FRAME_RELATED_P mechanism doesn't know about pop.  */
+-      RTX_FRAME_RELATED_P (insn) = 1;
+-      x = gen_rtx_PLUS (Pmode, stack_pointer_rtx, GEN_INT (UNITS_PER_WORD));
+-      x = gen_rtx_SET (stack_pointer_rtx, x);
+-      add_reg_note (insn, REG_FRAME_RELATED_EXPR, x);
+-      m->fs.sp_offset -= UNITS_PER_WORD;
++	  /* The RX FRAME_RELATED_P mechanism doesn't know about pop.  */
++	  RTX_FRAME_RELATED_P (insn) = 1;
++	  x = gen_rtx_PLUS (Pmode, stack_pointer_rtx, GEN_INT (UNITS_PER_WORD));
++	  x = gen_rtx_SET (stack_pointer_rtx, x);
++	  add_reg_note (insn, REG_FRAME_RELATED_EXPR, x);
++	  m->fs.sp_offset -= UNITS_PER_WORD;
++	}
++      else
++	{
++	  rtx x = gen_rtx_PLUS (Pmode, stack_pointer_rtx, GEN_INT (offset));
++	  x = gen_rtx_SET (sr->reg, gen_rtx_MEM (word_mode, x));
++	  emit_insn (x);
++	}
+     }
+ }
+ 
+@@ -12597,7 +12614,7 @@ release_scratch_register_on_entry (struct scratch_reg *sr)
+    pushed on the stack.  */
+ 
+ static void
+-ix86_adjust_stack_and_probe_stack_clash (const HOST_WIDE_INT size,
++ix86_adjust_stack_and_probe_stack_clash (HOST_WIDE_INT size,
+ 					 const bool int_registers_saved)
+ {
+   struct machine_function *m = cfun->machine;
+@@ -12713,6 +12730,12 @@ ix86_adjust_stack_and_probe_stack_clash (const HOST_WIDE_INT size,
+       struct scratch_reg sr;
+       get_scratch_register_on_entry (&sr);
+ 
++      /* If we needed to save a register, then account for any space
++	 that was pushed (we are not going to pop the register when
++	 we do the restore).  */
++      if (sr.saved)
++	size -= UNITS_PER_WORD;
++
+       /* Step 1: round SIZE down to a multiple of the interval.  */
+       HOST_WIDE_INT rounded_size = size & -probe_interval;
+ 
+@@ -12761,7 +12784,9 @@ ix86_adjust_stack_and_probe_stack_clash (const HOST_WIDE_INT size,
+ 				   m->fs.cfa_reg == stack_pointer_rtx);
+       dump_stack_clash_frame_info (PROBE_LOOP, size != rounded_size);
+ 
+-      release_scratch_register_on_entry (&sr);
++      /* This does not deallocate the space reserved for the scratch
++	 register.  That will be deallocated in the epilogue.  */
++      release_scratch_register_on_entry (&sr, size, false);
+     }
+ 
+   /* Make sure nothing is scheduled before we are done.  */
+@@ -12774,7 +12799,7 @@ ix86_adjust_stack_and_probe_stack_clash (const HOST_WIDE_INT size,
+    pushed on the stack.  */
+ 
+ static void
+-ix86_adjust_stack_and_probe (const HOST_WIDE_INT size,
++ix86_adjust_stack_and_probe (HOST_WIDE_INT size,
+ 			     const bool int_registers_saved)
+ {
+   /* We skip the probe for the first interval + a small dope of 4 words and
+@@ -12847,6 +12872,11 @@ ix86_adjust_stack_and_probe (const HOST_WIDE_INT size,
+ 
+       get_scratch_register_on_entry (&sr);
+ 
++      /* If we needed to save a register, then account for any space
++	 that was pushed (we are not going to pop the register when
++	 we do the restore).  */
++      if (sr.saved)
++	size -= UNITS_PER_WORD;
+ 
+       /* Step 1: round SIZE to the previous multiple of the interval.  */
+ 
+@@ -12906,7 +12936,9 @@ ix86_adjust_stack_and_probe (const HOST_WIDE_INT size,
+ 						    (get_probe_interval ()
+ 						     + dope))));
+ 
+-      release_scratch_register_on_entry (&sr);
++      /* This does not deallocate the space reserved for the scratch
++	 register.  That will be deallocated in the epilogue.  */
++      release_scratch_register_on_entry (&sr, size, false);
+     }
+ 
+   /* Even if the stack pointer isn't the CFA register, we need to correctly
+@@ -13055,7 +13087,7 @@ ix86_emit_probe_stack_range (HOST_WIDE_INT first, HOST_WIDE_INT size,
+ 						       sr.reg),
+ 					 rounded_size - size));
+ 
+-      release_scratch_register_on_entry (&sr);
++      release_scratch_register_on_entry (&sr, size, true);
+     }
+ 
+   /* Make sure nothing is scheduled before we are done.  */
+
+diff --git a/gcc/testsuite/gcc.target/i386/pr84128.c b/gcc/testsuite/gcc.target/i386/pr84128.c
+new file mode 100644
+index 0000000..a8323fd6
+--- /dev/null
++++ gcc/testsuite/gcc.target/i386/pr84128.c
+@@ -0,0 +1,30 @@
++/* { dg-do run } */
++/* { dg-options "-O2 -march=i686 -mtune=generic -fstack-clash-protection" } */
++/* { dg-require-effective-target ia32 } */
++
++__attribute__ ((noinline, noclone, weak, regparm (3)))
++int
++f1 (long arg0, int (*pf) (long, void *))
++{
++  unsigned char buf[32768];
++  return pf (arg0, buf);
++}
++
++__attribute__ ((noinline, noclone, weak))
++int
++f2 (long arg0, void *ignored)
++{
++  if (arg0 != 17)
++    __builtin_abort ();
++  return 19;
++}
++
++int
++main (void)
++{
++  if (f1 (17, f2) != 19)
++    __builtin_abort ();
++  return 0;
++}
++
++