[PATCH 0/4] Reduce use of .code16gcc

[David Woodhouse]

This is a gratuitous GCC-ism. For C code actually compiled with GCC we
should be using -m16 where it's available (GCC 4.9+).

And where the only thing marked with .code16gcc is explicit assembler
code, we should just use .code16 and avoid letting the compiler make any
of the assumptions that the difference affects. Which, in fact, we
already do.

(Once upon a time with ancient versions of gas, we needed to
use .code16gcc because some instructions just wouldn't compile
otherwise. That hasn't been true for a while though.)

It still doesn't actually build with clang after this, but it's a bit
closer.

[PATCH 1/4] build: use -m16 where available instead of asm(".code16gcc")

[David Woodhouse]

GCC 4.9 and clang 3.5 support the -m16 option on the command line which
supersedes the hackish ".code16gcc" assembler directive. Use it where
possible.

Signed-off-by: David Woodhouse 
---
 Makefile | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/Makefile b/Makefile
index 78b598e..4c42124 100644
--- a/Makefile
+++ b/Makefile
@@ -62,13 +62,15 @@ COMMONCFLAGS := -I$(OUT) -Isrc -Os -MD -g \
 COMMONCFLAGS += $(call cc-option,$(CC),-nopie,)
 COMMONCFLAGS += $(call cc-option,$(CC),-fno-stack-protector,)
 COMMONCFLAGS += $(call cc-option,$(CC),-fno-stack-protector-all,)
+COMMA := ,

CFLAGS32FLAT := $(COMMONCFLAGS) -DMODE16=0 -DMODESEGMENT=0 -fomit-frame-pointer
 CFLAGSSEG := $(COMMONCFLAGS) -DMODESEGMENT=1 -fno-defer-pop \
     $(call cc-option,$(CC),-fno-jump-tables,-DMANUAL_NO_JUMP_TABLE) \
     $(call cc-option,$(CC),-fno-tree-switch-conversion,)
 CFLAGS32SEG := $(CFLAGSSEG) -DMODE16=0 -fomit-frame-pointer
-CFLAGS16INC := $(CFLAGSSEG) -DMODE16=1 -Wa,src/code16gcc.s \
+CFLAGS16INC := $(CFLAGSSEG) -DMODE16=1 \
+    $(call cc-option,$(CC),-m16,-Wa$(COMMA)src/code16gcc.s) \
     $(call cc-option,$(CC),--param large-stack-frame=4,-fno-inline)
 CFLAGS16 := $(CFLAGS16INC) -fomit-frame-pointer

[PATCH 2/4] smm: Use .code16 not .code16gcc

[David Woodhouse]

There's no need to use .code16gcc where we are writing assembler code
explicitly. It only affects word-size-ambiguous instructions, and we
should just be explicit. And we are.

Signed-off-by: David Woodhouse 
---
 src/fw/smm.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/fw/smm.c b/src/fw/smm.c
index 0f59f20..77e1e63 100644
--- a/src/fw/smm.c
+++ b/src/fw/smm.c
@@ -19,7 +19,7 @@
 extern u8 smm_relocation_start, smm_relocation_end;
 ASM32FLAT(
     ".global smm_relocation_start, smm_relocation_end\n"
-    "  .code16gcc\n"
+    "  .code16\n"

/* code to relocate SMBASE to 0xa0000 */
     "smm_relocation_start:\n"
@@ -46,7 +46,7 @@ ASM32FLAT(
 extern u8 smm_code_start, smm_code_end;
 ASM32FLAT(
     ".global smm_code_start, smm_code_end\n"
-    "  .code16gcc\n"
+    "  .code16\n"
     "smm_code_start:\n"
     "  rsm\n"
     "smm_code_end:\n"

[PATCH 3/4] romlayout: Use .code16 not .code16gcc

[David Woodhouse]

There's no need to use .code16gcc where we are writing assembler code
explicitly. It only affects word-size-ambiguous instructions, and we
should just be explicit. And we are.

Signed-off-by: David Woodhouse 
---
 src/romlayout.S | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

diff --git a/src/romlayout.S b/src/romlayout.S
index 57e8bcc..9c2719e 100644
--- a/src/romlayout.S
+++ b/src/romlayout.S
@@ -21,7 +21,7 @@
 // %edx = return location (in 32bit mode)
 // Clobbers: ecx, flags, segment registers, cr0, idt/gdt
         DECLFUNC transition32
-        .code16gcc
+        .code16
 transition32:
         movl %eax, %ecx

@@ -102,7 +102,7 @@ transition16big:

ljmpw $SEG32_MODE16BIG_CS, $1f

-        .code16gcc
+        .code16
 1:
         // Disable protected mode
         movl %cr0, %eax
@@ -145,7 +145,7 @@ __call16big:
         jmp transition16big

// Make call.
-        .code16gcc
+        .code16
 1:      movl $_zonelow_seg, %edx        // Adjust %ds, %ss, and %esp
         movl %edx, %ds
         movzwl StackSeg, %edx
@@ -177,7 +177,7 @@ __call16big:
 // Far call a 16bit function from 16bit mode with a specified cpu register state
 // %eax = address of struct bregs, %edx = segment of struct bregs
 // Clobbers: %e[bc]x, %e[ds]i, flags
-        .code16gcc
+        .code16
         DECLFUNC __farcall16
 __farcall16:
         // Save %edx/%eax, %ebp
@@ -372,7 +372,7 @@ entry_pcibios32:
         popfl
         lretl

-        .code16gcc
+        .code16
         DECLFUNC entry_pcibios16
 entry_pcibios16:
         ENTRY_ARG handle_pcibios
@@ -421,7 +421,7 @@ entry_elf:
         movl $BUILD_STACK_ADDR, %esp
         ljmpl $SEG32_MODE32_CS, $_cfunc32flat_handle_post

-        .code16gcc
+        .code16

// UEFI Compatibility Support Module (CSM) entry point
         EXPORTFUNC entry_csm
@@ -453,7 +453,7 @@ entry_csm:
 __csm_return:
         movl $1f, %edx
         jmp transition16big
-        .code16gcc
+        .code16

// Switch back to original stack
 1:      movzwl BREGS_code+2(%eax), %edx

[PATCH 4/4] vgaentry: Use .code16 not .code16gcc

[David Woodhouse]

There's no need to use .code16gcc where we are writing assembler code
explicitly. It only affects word-size-ambiguous instructions, and we
should just be explicit. And we are.

Signed-off-by: David Woodhouse 
---
 vgasrc/vgaentry.S | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/vgasrc/vgaentry.S b/vgasrc/vgaentry.S
index 11197f1..3c2c885 100644
--- a/vgasrc/vgaentry.S
+++ b/vgasrc/vgaentry.S
@@ -15,7 +15,7 @@
  ****************************************************************/

.section .rom.header
-        .code16gcc
+        .code16
         .global _rom_header, _rom_header_size, _rom_header_checksum
 _rom_header:
         .word 0xaa55

[Kevin O'Connor]

On Tue, Jun 03, 2014 at 05:25:57PM +0100, David Woodhouse wrote:
> This is a gratuitous GCC-ism. For C code actually compiled with GCC we
> should be using -m16 where it's available (GCC 4.9+).
>
> And where the only thing marked with .code16gcc is explicit assembler
> code, we should just use .code16 and avoid letting the compiler make any
> of the assumptions that the difference affects. Which, in fact, we
> already do.
>
> (Once upon a time with ancient versions of gas, we needed to
> use .code16gcc because some instructions just wouldn't compile
> otherwise. That hasn't been true for a while though.)
>
> It still doesn't actually build with clang after this, but it's a bit
> closer.

Thanks.  Looks good to me.

I updated the series to apply on top of my pending patches and added
it to my queue:

https://github.com/KevinOConnor/seabios/commits/testing

-Kevin

[Kevin O'Connor]

On Tue, Jun 03, 2014 at 03:58:50PM -0400, Kevin O'Connor wrote:
> On Tue, Jun 03, 2014 at 05:25:57PM +0100, David Woodhouse wrote:
> > This is a gratuitous GCC-ism. For C code actually compiled with GCC we
> > should be using -m16 where it's available (GCC 4.9+).
> >
> > And where the only thing marked with .code16gcc is explicit assembler
> > code, we should just use .code16 and avoid letting the compiler make any
> > of the assumptions that the difference affects. Which, in fact, we
> > already do.
> >
> > (Once upon a time with ancient versions of gas, we needed to
> > use .code16gcc because some instructions just wouldn't compile
> > otherwise. That hasn't been true for a while though.)
> >
> > It still doesn't actually build with clang after this, but it's a bit
> > closer.
>
> Thanks.  Looks good to me.
>
> I updated the series to apply on top of my pending patches and added
> it to my queue:

FYI, I committed this (slightly modified) patch series.

-Kevin