[PATCH v2 0/1] MR4030: configure: Use -mpreferred-stack-boundary=2 on i386.

[Zebediah Figura (＠zfigura)]

--
  v2: configure: Use -mpreferred-stack-boundary=2 on i386.
 
https://gitlab.winehq.org/wine/wine/-/merge_requests/4030

[PATCH v2 1/1] configure: Use -mpreferred-stack-boundary=2 on i386.

[Zebediah Figura]

From: Zebediah Figura <zfigura(a)codeweavers.com>

gcc currently assumes the stack alignment for i686-w64-mingw32 is 16, which is
essentially wrong. It works around this in most cases by applying -mstackrealign
when SSE is in use. However, this doesn't address the case of manually aligned
types (i.e. DECLSPEC_ALIGN), and it also for some reason doesn't seem to apply
when -mavx512f is in use, which seems to be an independent gcc bug.

The correct solution on the gcc end is not to assume that the stack is 4-byte
aligned. -mpreferred-stack-boundary=2 achieves exactly this, and there is motion
upstream to address this by effectively making this setting the default. Since
this will take time to propagate downstream, though, it seems prudent to address
this bug locally by essentially applying the same fix downstream.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=55007
Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=55899
---
 configure    | 34 ++++++++++++++++++++++++++++++++++
 configure.ac |  1 +
 2 files changed, 35 insertions(+)

diff --git a/configure b/configure
index 2ffbfd8520c..294ab23ba1f 100755
--- a/configure
+++ b/configure
@@ -11448,6 +11448,40 @@ printf "%s\n" "$ac_res" >&6; }
 if eval test \"x\$"$as_ac_var"\" = x"yes"
 then :
   as_fn_append ${wine_arch}_EXTRACFLAGS " -fno-omit-frame-pointer"
+fi }
+            { as_ac_var=`printf "%s\n" "ac_cv_${wine_arch}_cflags_-mpreferred-stack-boundary=2" | $as_tr_sh`
+{ printf "%s\n" "$as_me:${as_lineno-$LINENO}: checking whether $CC supports -mpreferred-stack-boundary=2" >&5
+printf %s "checking whether $CC supports -mpreferred-stack-boundary=2... " >&6; }
+if eval test \${$as_ac_var+y}
+then :
+  printf %s "(cached) " >&6
+else $as_nop
+  ac_wine_try_cflags_saved=$CFLAGS
+ac_wine_try_cflags_saved_exeext=$ac_exeext
+CFLAGS="$CFLAGS -nostdlib -nodefaultlibs -mpreferred-stack-boundary=2"
+ac_exeext=".exe"
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+void *__os_arm64x_dispatch_ret = 0;
+int __cdecl mainCRTStartup(void) { return 0; }
+_ACEOF
+if ac_fn_c_try_link "$LINENO"
+then :
+  eval "$as_ac_var=yes"
+else $as_nop
+  eval "$as_ac_var=no"
+fi
+rm -f core conftest.err conftest.$ac_objext conftest.beam \
+    conftest$ac_exeext conftest.$ac_ext
+CFLAGS=$ac_wine_try_cflags_saved
+ac_exeext=$ac_wine_try_cflags_saved_exeext
+fi
+eval ac_res=\$$as_ac_var
+	       { printf "%s\n" "$as_me:${as_lineno-$LINENO}: result: $ac_res" >&5
+printf "%s\n" "$ac_res" >&6; }
+if eval test \"x\$"$as_ac_var"\" = x"yes"
+then :
+  as_fn_append ${wine_arch}_EXTRACFLAGS " -mpreferred-stack-boundary=2"
 fi }
             { as_ac_var=`printf "%s\n" "ac_cv_${wine_arch}_cflags_-Wl,--disable-stdcall-fixup" | $as_tr_sh`
 { printf "%s\n" "$as_me:${as_lineno-$LINENO}: checking whether $CC supports -Wl,--disable-stdcall-fixup" >&5
diff --git a/configure.ac b/configure.ac
index aba623ee6e7..6551f22042c 100644
--- a/configure.ac
+++ b/configure.ac
@@ -969,6 +969,7 @@ This is an error since --enable-archs=$wine_arch was requested.])])
 
     case $wine_arch in
       i386) WINE_TRY_PE_CFLAGS([-fno-omit-frame-pointer])
+            WINE_TRY_PE_CFLAGS([-mpreferred-stack-boundary=2])
             WINE_TRY_PE_CFLAGS([-Wl,--disable-stdcall-fixup],
                                [AS_VAR_APPEND([${wine_arch}_LDFLAGS],[" -Wl,--disable-stdcall-fixup"])]) ;;
       x86_64) WINE_TRY_PE_CFLAGS([-Wformat-overflow])
-- 
GitLab

https://gitlab.winehq.org/wine/wine/-/merge_requests/4030

[Zebediah Figura (＠zfigura)]

I see now that my original patch comment was also rather misleading, implying that we were working around bugs that should be fixed a proper way, which isn't really the case. I've reworded the commit message to explain better now.

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/4030#note_52815