x86-32: Fix non-standard stack frame layout

This patch fixes the non-standard stack frame layout on 32-bit. Prolog is
changed as follows:

Before:

  [main]   0xa736b790:   57                         push   %edi
  [main]   0xa736b791:   56                         push   %esi
  [main]   0xa736b792:   53                         push   %ebx
  [main]   0xa736b793:   55                         push   %ebp
  [main]   0xa736b794:   89 e5                      mov    %esp,%ebp
  [main]   0xa736b796:   83 ec 10                   sub    $0x10,%esp

After:

  [main]   0xa73db4a8:   55                         push   %ebp
  [main]   0xa73db4a9:   89 e5                      mov    %esp,%ebp
  [main]   0xa73db4ab:   83 ec 10                   sub    $0x10,%esp
  [main]   0xa73db4ae:   57                         push   %edi
  [main]   0xa73db4af:   56                         push   %esi
  [main]   0xa73db4b0:   53                         push   %ebx

Signed-off-by: Pekka Enberg <penberg@kernel.org>

5 files changed, 8 insertions, 25 deletions

diff --git a/arch/x86/emit_32.c b/arch/x86/emit_32.c
index 40004c06..f133bdac 100644
--- a/arch/x86/emit_32.c
+++ b/arch/x86/emit_32.c
@@ -621,14 +621,14 @@ static void emit_restore_callee_save_regs(struct buffer *buf)
 void emit_prolog(struct buffer *buf, struct stack_frame *frame,
 					unsigned long frame_size)
 {
-	emit_save_callee_save_regs(buf);
-
 	__emit_push_reg(buf, MACH_REG_EBP);
 	__emit_mov_reg_reg(buf, MACH_REG_ESP, MACH_REG_EBP);
 
 	if (frame_size)
 		__emit_sub_imm_reg(buf, frame_size, MACH_REG_ESP);
 
+	emit_save_callee_save_regs(buf);
+
 	if (opt_debug_stack)
 		__emit_push_imm(buf, STACK_FRAME_REDZONE_END);
 }
@@ -667,22 +667,18 @@ void emit_epilog(struct buffer *buf)
 	if (opt_debug_stack)
 		emit_stack_redzone_check(buf);
 
-	emit_leave(buf);
 	emit_restore_regs(buf);
+	emit_leave(buf);
 	emit_ret(buf);
 }
 
-/*
- * FIXME: We use different stack layout on 32-bit and 64-bit so prolog, epilog,
- * and unwind sequences are slightly different.
- */
 void emit_unwind(struct buffer *buf)
 {
 	if (opt_debug_stack)
 		emit_stack_redzone_check(buf);
 
-	emit_leave(buf);
 	emit_restore_regs(buf);
+	emit_leave(buf);
 	__emit_jmp(buf, (unsigned long)&unwind);
 }
 
diff --git a/arch/x86/emit_64.c b/arch/x86/emit_64.c
index f78bb665..081ab8df 100644
--- a/arch/x86/emit_64.c
+++ b/arch/x86/emit_64.c
@@ -1265,10 +1265,6 @@ void emit_epilog(struct buffer *buf)
 	emit_ret(buf);
 }
 
-/*
- * FIXME: We use different stack layout on 32-bit and 64-bit so prolog, epilog,
- * and unwind sequences are slightly different.
- */
 void emit_unwind(struct buffer *buf)
 {
 	if (opt_debug_stack)
diff --git a/arch/x86/include/arch/stack-frame.h b/arch/x86/include/arch/stack-frame.h
index fc0ab27c..d19e58e4 100644
--- a/arch/x86/include/arch/stack-frame.h
+++ b/arch/x86/include/arch/stack-frame.h
@@ -22,9 +22,6 @@ struct native_stack_frame {
 
 struct jit_stack_frame {
 	void *prev; /* previous stack frame link */
-	unsigned long old_ebx;
-	unsigned long old_esi;
-	unsigned long old_edi;
 	unsigned long return_address;
 	unsigned long args[0];
 } __attribute__((packed));
diff --git a/arch/x86/stack-frame.c b/arch/x86/stack-frame.c
index c7db807e..236c2960 100644
--- a/arch/x86/stack-frame.c
+++ b/arch/x86/stack-frame.c
@@ -40,10 +40,7 @@
 #include "arch/stack-frame.h"
 
 /*
- * The three callee-saved registers are unconditionally stored on the stack
- * after EIP and EBP (see emit_prolog() for details). Therefore, there are five
- * 32-bit stack slots before the first argument to a function as illustrated by
- * the following diagram:
+ * The stack frame layout is illustrated by the following diagram:
  *
  *     :              :  ^
  *     :              :  |  Higher memory addresses
@@ -52,9 +49,6 @@
  *     :     ...      :
  *     |    Arg 1     | <-- Start offset of arguments
  *     |   Old EIP    |
- *     |     EDI      |
- *     |     ESI      |
- *     |     EBX      |
  *     |   Old EBP    | <-- New EBP
  *     |   Local 1    |
  *     :     ...      :
diff --git a/test/unit/arch-x86/encode-test.c b/test/unit/arch-x86/encode-test.c
index 56f8d09d..e417d890 100644
--- a/test/unit/arch-x86/encode-test.c
+++ b/test/unit/arch-x86/encode-test.c
@@ -948,7 +948,7 @@ void test_encoding_reg_memdisp(void)
 void test_encoding_memlocal(void)
 {
 #ifdef CONFIG_32_BIT
-	uint8_t encoding[] = { 0xff, 0xb5, 0x14, 0x00, 0x00, 0x00 };
+	uint8_t encoding[] = { 0xff, 0xb5, 0x08, 0x00, 0x00, 0x00 };
 #else
 	uint8_t encoding[] = { 0xff, 0xb5, 0x10, 0x00, 0x00, 0x00 };
 #endif
@@ -981,7 +981,7 @@ void test_encoding_memlocal(void)
 void test_encoding_fstp_memlocal(void)
 {
 #ifdef CONFIG_32_BIT
-	uint8_t encoding[] = { 0xd9, 0x9d, 0x14, 0x00, 0x00, 0x00 };
+	uint8_t encoding[] = { 0xd9, 0x9d, 0x08, 0x00, 0x00, 0x00 };
 #else
 	uint8_t encoding[] = { 0xd9, 0x9d, 0x10, 0x00, 0x00, 0x00 };
 #endif
@@ -1014,7 +1014,7 @@ void test_encoding_fstp_memlocal(void)
 void test_encoding_fstp_64_memlocal(void)
 {
 #ifdef CONFIG_32_BIT
-	uint8_t encoding[] = { 0xdd, 0x9d, 0x14, 0x00, 0x00, 0x00 };
+	uint8_t encoding[] = { 0xdd, 0x9d, 0x08, 0x00, 0x00, 0x00 };
 #else
 	uint8_t encoding[] = { 0xdd, 0x9d, 0x10, 0x00, 0x00, 0x00 };
 #endif