From 23a683adf803eef405d248cc9c2a7eb08a7300e2 Mon Sep 17 00:00:00 2001
From: Victor Stinner <vstinner@redhat.com>
Date: Fri, 12 Apr 2019 21:27:37 +0200
Subject: [PATCH] bpo-36618: Add -fmax-type-align=8 flag for clang (GH-12809)

Add -fmax-type-align=8 to CFLAGS when clang compiler is detected.

The pymalloc memory allocator aligns memory on 8 bytes. On x86-64,
clang expects alignment on 16 bytes by default and so uses MOVAPS
instruction which can lead to segmentation fault. Instruct clang that
Python is limited to alignemnt on 8 bytes to use MOVUPS instruction
instead: slower but don't trigger a SIGSEGV if the memory is not
aligned on 16 bytes.

Sadly, the flag must be expected to CFLAGS and not just
CFLAGS_NODIST, since third party C extensions can have the same
issue.
---
 .../2019-04-12-19-49-10.bpo-36618.gcI9iq.rst  |  8 ++++
 configure                                     | 42 +++++++++++++------
 configure.ac                                  | 41 ++++++++++++------
 3 files changed, 65 insertions(+), 26 deletions(-)
 create mode 100644 Misc/NEWS.d/next/Build/2019-04-12-19-49-10.bpo-36618.gcI9iq.rst

diff --git a/Misc/NEWS.d/next/Build/2019-04-12-19-49-10.bpo-36618.gcI9iq.rst b/Misc/NEWS.d/next/Build/2019-04-12-19-49-10.bpo-36618.gcI9iq.rst
new file mode 100644
index 00000000000..597dd67b46e
--- /dev/null
+++ b/Misc/NEWS.d/next/Build/2019-04-12-19-49-10.bpo-36618.gcI9iq.rst
@@ -0,0 +1,8 @@
+Add ``-fmax-type-align=8`` to CFLAGS when clang compiler is detected. The
+pymalloc memory allocator aligns memory on 8 bytes. On x86-64, clang expects
+alignment on 16 bytes by default and so uses MOVAPS instruction which can
+lead to segmentation fault. Instruct clang that Python is limited to
+alignemnt on 8 bytes to use MOVUPS instruction instead: slower but don't
+trigger a SIGSEGV if the memory is not aligned on 16 bytes. Sadly, the flag
+must be expected to ``CFLAGS`` and not just ``CFLAGS_NODIST``, since third
+party C extensions can have the same issue.
diff --git a/configure b/configure
index 72589fdb78c..ac1e66a96bb 100755
--- a/configure
+++ b/configure
@@ -6813,6 +6813,19 @@ esac
 # compiler and platform.  BASECFLAGS tweaks need to be made even if the
 # user set OPT.
 
+case $CC in
+    *clang*)
+        cc_is_clang=1
+        ;;
+    *)
+        if $CC --version 2>&1 | grep -q clang
+        then
+            cc_is_clang=1
+        else
+            cc_is_clang=
+        fi
+esac
+
 # tweak OPT based on compiler and platform, only if the user didn't set
 # it on the command line
 
@@ -6826,19 +6839,6 @@ then
            WRAP="-fwrapv"
         fi
 
-        case $CC in
-            *clang*)
-                cc_is_clang=1
-                ;;
-            *)
-                if $CC --version 2>&1 | grep -q clang
-                then
-                    cc_is_clang=1
-                else
-                    cc_is_clang=
-                fi
-        esac
-
         if test -n "${cc_is_clang}"
         then
             # Clang also needs -fwrapv
@@ -6879,6 +6879,21 @@ then
     esac
 fi
 
+if test -n "${cc_is_clang}"
+then
+    # bpo-36618: Add -fmax-type-align=8 to CFLAGS when clang compiler is
+    # detected. The pymalloc memory allocator aligns memory on 8 bytes. On
+    # x86-64, clang expects alignment on 16 bytes by default and so uses MOVAPS
+    # instruction which can lead to segmentation fault. Instruct clang that
+    # Python is limited to alignemnt on 8 bytes to use MOVUPS instruction
+    # instead: slower but don't trigger a SIGSEGV if the memory is not aligned
+    # on 16 bytes.
+    #
+    # Sadly, the flag must be expected to CFLAGS and not just CFLAGS_NODIST,
+    # since third party C extensions can have the same issue.
+    CFLAGS="$CFLAGS -fmax-type-align=8"
+fi
+
 
 
 
@@ -10200,6 +10215,7 @@ fi
 
 
 
+
 if test "x$ac_cv_env_PKG_CONFIG_set" != "xset"; then
 	if test -n "$ac_tool_prefix"; then
   # Extract the first word of "${ac_tool_prefix}pkg-config", so it can be a program name with args.
diff --git a/configure.ac b/configure.ac
index 30e8587cb08..863b34245ad 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1464,6 +1464,19 @@ esac
 # compiler and platform.  BASECFLAGS tweaks need to be made even if the
 # user set OPT.
 
+case $CC in
+    *clang*)
+        cc_is_clang=1
+        ;;
+    *)
+        if $CC --version 2>&1 | grep -q clang
+        then
+            cc_is_clang=1
+        else
+            cc_is_clang=
+        fi
+esac
+
 # tweak OPT based on compiler and platform, only if the user didn't set
 # it on the command line
 AC_SUBST(OPT)
@@ -1477,19 +1490,6 @@ then
            WRAP="-fwrapv"
         fi
 
-        case $CC in
-            *clang*)
-                cc_is_clang=1
-                ;;
-            *)
-                if $CC --version 2>&1 | grep -q clang
-                then
-                    cc_is_clang=1
-                else
-                    cc_is_clang=
-                fi
-        esac
-
         if test -n "${cc_is_clang}"
         then
             # Clang also needs -fwrapv
@@ -1530,6 +1530,21 @@ then
     esac
 fi
 
+if test -n "${cc_is_clang}"
+then
+    # bpo-36618: Add -fmax-type-align=8 to CFLAGS when clang compiler is
+    # detected. The pymalloc memory allocator aligns memory on 8 bytes. On
+    # x86-64, clang expects alignment on 16 bytes by default and so uses MOVAPS
+    # instruction which can lead to segmentation fault. Instruct clang that
+    # Python is limited to alignemnt on 8 bytes to use MOVUPS instruction
+    # instead: slower but don't trigger a SIGSEGV if the memory is not aligned
+    # on 16 bytes.
+    #
+    # Sadly, the flag must be expected to CFLAGS and not just CFLAGS_NODIST,
+    # since third party C extensions can have the same issue.
+    CFLAGS="$CFLAGS -fmax-type-align=8"
+fi
+
 AC_SUBST(BASECFLAGS)
 AC_SUBST(CFLAGS_NODIST)
 AC_SUBST(LDFLAGS_NODIST)