Add support for gcc v3.x compilers.

Suppress __attribute__((externally_visible)) when no -fwhole-program
Add switch hack for compilers without -fno-jump-tables
Define memcpy() function (for compilers without -minline-all-stringops).
Define call16_simpint as a macro (a param needs to be inlined).
Make sure s3_resume is only defined in 32bit mode.

7 files changed, 53 insertions, 34 deletions

diff --git a/Makefile b/Makefile
index ff594e7c..8cb66259 100644
--- a/Makefile
+++ b/Makefile
@@ -22,7 +22,8 @@ cc-option = $(shell if test -z "`$(1) $(2) -S -o /dev/null -xc \
 # Default compiler flags
 COMMONCFLAGS = -Wall -Os -MD -m32 -march=i386 -mregparm=3 \
                -mpreferred-stack-boundary=2 -mrtd -freg-struct-return \
-               -ffreestanding -fwhole-program -fomit-frame-pointer \
+               $(call cc-option,$(CC),-fwhole-program -DWHOLE_PROGRAM,) \
+               -ffreestanding -fomit-frame-pointer \
                -fno-delete-null-pointer-checks -Wno-strict-aliasing \
                -ffunction-sections -fdata-sections \
                -minline-all-stringops
@@ -31,7 +32,8 @@ COMMONCFLAGS += $(call cc-option,$(CC),-fno-stack-protector,)
 COMMONCFLAGS += $(call cc-option,$(CC),-fno-stack-protector-all,)
 
 override CFLAGS = $(COMMONCFLAGS) -g -DMODE16=0
-CFLAGS16INC = $(COMMONCFLAGS) -DMODE16=1 -fno-jump-tables -fno-defer-pop \
+CFLAGS16INC = $(COMMONCFLAGS) -DMODE16=1 -fno-defer-pop \
+              $(call cc-option,$(CC),-fno-jump-tables,-DMANUAL_NO_JUMP_TABLE) \
               $(call cc-option,$(CC),-fno-tree-switch-conversion,) \
               $(call cc-option,$(CC),--param large-stack-frame=4,)
 CFLAGS16 = $(CFLAGS16INC) -g
diff --git a/src/optionroms.c b/src/optionroms.c
index 366e1077..2856217a 100644
--- a/src/optionroms.c
+++ b/src/optionroms.c
@@ -171,7 +171,7 @@ copy_rom(struct rom_header *rom)
     }
     dprintf(4, "Copying option rom (size %d) from %p to %x\n"
             , romsize, rom, next_rom);
-    memcpy4((void*)next_rom, rom, romsize);
+    memcpy((void*)next_rom, rom, romsize);
     return (struct rom_header *)next_rom;
 }
 
diff --git a/src/resume.c b/src/resume.c
index ad868f62..6db6b714 100644
--- a/src/resume.c
+++ b/src/resume.c
@@ -94,6 +94,7 @@ handle_resume(u8 status)
     panic("Unimplemented shutdown status: %02x\n", status);
 }
 
+#if MODE16==0
 void VISIBLE32
 s3_resume()
 {
@@ -123,6 +124,7 @@ s3_resume()
     }
     call16big(&br);
 }
+#endif
 
 // Ughh - some older gcc compilers have a bug which causes VISIBLE32
 // functions to not be exported as global variables.
diff --git a/src/types.h b/src/types.h
index a004474e..aa9e12d0 100644
--- a/src/types.h
+++ b/src/types.h
@@ -21,7 +21,15 @@ union u64_u32_u {
     u64 val;
 };
 
-#define __VISIBLE __attribute__((externally_visible))
+#ifdef MANUAL_NO_JUMP_TABLE
+# define default case 775324556: asm(""); default
+#endif
+
+#ifdef WHOLE_PROGRAM
+# define __VISIBLE __attribute__((externally_visible))
+#else
+# define __VISIBLE
+#endif
 
 #define UNIQSEC __FILE__ "." __stringify(__LINE__)
 
diff --git a/src/util.c b/src/util.c
index 1a67e370..a44b35d4 100644
--- a/src/util.c
+++ b/src/util.c
@@ -52,25 +52,6 @@ __call16_int(struct bregs *callregs, u16 offset)
     call16(callregs);
 }
 
-inline void
-call16_simpint(int nr, u32 *eax, u32 *flags)
-{
-    extern void __force_link_error__call16_simpint_only_in_16bit_mode();
-    if (!MODE16)
-        __force_link_error__call16_simpint_only_in_16bit_mode();
-
-    asm volatile(
-        "stc\n"
-        "int %2\n"
-        "pushfl\n"
-        "popl %1\n"
-        "cli\n"
-        "cld"
-        : "+a"(*eax), "=r"(*flags)
-        : "i"(nr)
-        : "cc", "memory");
-}
-
 // Switch to the extra stack in ebda and call a function.
 inline u32
 stack_hop(u32 eax, u32 edx, u32 ecx, void *func)
@@ -204,15 +185,27 @@ memcpy_far(u16 d_seg, void *d_far, u16 s_seg, const void *s_far, size_t len)
         : "cc", "memory");
 }
 
-// Memcpy that uses 4-byte accesses
-void
-memcpy4(void *d1, const void *s1, size_t len)
+void *
+#undef memcpy
+memcpy(void *d1, const void *s1, size_t len)
+#define memcpy __builtin_memcpy
 {
+    void *d = d1;
+    if (((u32)d1 | (u32)s1 | len) & 3) {
+        // non-aligned memcpy
+        asm volatile(
+            "rep movsb (%%esi),%%es:(%%edi)"
+            : "+c"(len), "+S"(s1), "+D"(d)
+            : : "cc", "memory");
+        return d1;
+    }
+    // Common case - use 4-byte copy
     len /= 4;
     asm volatile(
         "rep movsl (%%esi),%%es:(%%edi)"
-        : "+c"(len), "+S"(s1), "+D"(d1)
+        : "+c"(len), "+S"(s1), "+D"(d)
         : : "cc", "memory");
+    return d1;
 }
 
 void *
diff --git a/src/util.h b/src/util.h
index 1d33c947..23750cd4 100644
--- a/src/util.h
+++ b/src/util.h
@@ -65,6 +65,23 @@ static inline u64 rdtscll(void)
     return val;
 }
 
+#define call16_simpint(nr, peax, pflags) do {                           \
+        extern void __force_link_error__call16_simpint_only_in_16bit_mode(); \
+        if (!MODE16)                                                    \
+            __force_link_error__call16_simpint_only_in_16bit_mode();    \
+                                                                        \
+        asm volatile(                                                   \
+            "stc\n"                                                     \
+            "int %2\n"                                                  \
+            "pushfl\n"                                                  \
+            "popl %1\n"                                                 \
+            "cli\n"                                                     \
+            "cld"                                                       \
+            : "+a"(*peax), "=r"(*pflags)                                \
+            : "i"(nr)                                                   \
+            : "cc", "memory");                                          \
+    } while (0)
+
 // util.c
 inline u32 stack_hop(u32 eax, u32 edx, u32 ecx, void *func);
 u8 checksum_far(u16 buf_seg, void *buf_far, u32 len);
@@ -75,8 +92,7 @@ int strcmp(const char *s1, const char *s2);
 inline void memset_far(u16 d_seg, void *d_far, u8 c, size_t len);
 inline void memset16_far(u16 d_seg, void *d_far, u16 c, size_t len);
 void *memset(void *s, int c, size_t n);
-void memcpy4(void *d1, const void *s1, size_t len);
-#define memcpy(d1, s1, len) __builtin_memcpy((d1), (s1), (len))
+#define memcpy __builtin_memcpy
 inline void memcpy_far(u16 d_seg, void *d_far
                        , u16 s_seg, const void *s_far, size_t len);
 void *memmove(void *d, const void *s, size_t len);
@@ -89,7 +105,6 @@ inline void __call16_int(struct bregs *callregs, u16 offset);
         extern void irq_trampoline_ ##nr ();                    \
         __call16_int((callregs), (u32)&irq_trampoline_ ##nr );  \
     } while (0)
-inline void call16_simpint(int nr, u32 *eax, u32 *flags);
 void usleep(u32 usec);
 int get_keystroke(int msec);
 
diff --git a/tools/test-gcc.sh b/tools/test-gcc.sh
index c354ec3d..55f359fa 100755
--- a/tools/test-gcc.sh
+++ b/tools/test-gcc.sh
@@ -11,10 +11,9 @@ TMPFILE3o=out/tmp_testcompile3.o
 # Test for "-fwhole-program"
 $CC -fwhole-program -S -o /dev/null -xc /dev/null > /dev/null 2>&1
 if [ $? -ne 0 ]; then
-    echo "This version of gcc does not support -fwhole-program." > /dev/fd/2
-    echo "Please upgrade to gcc v4.1 or later" > /dev/fd/2
-    echo -1
-    exit 1
+    echo "  Working around no -fwhole-program" > /dev/fd/2
+    echo 1
+    exit 0
 fi
 
 # Test if "visible" variables are marked global.