From 323384a01ed8500f48081f88754d1495d63eff4c Mon Sep 17 00:00:00 2001
From: batt <batt@38d2e660-2303-0410-9eaa-f027e97ec537>
Date: Thu, 14 Jan 2010 14:21:26 +0000
Subject: [PATCH] Optimization: Use a trampoline to start ARM processes instead
 of wasting one stack location and saving lr twice.

git-svn-id: https://src.develer.com/svnoss/bertos/trunk@3134 38d2e660-2303-0410-9eaa-f027e97ec537
---
 bertos/cpu/arm/hw/switch_ctx_arm.S | 20 ++++++++++++--------
 bertos/cpu/frame.h                 | 13 +++++++------
 2 files changed, 19 insertions(+), 14 deletions(-)

diff --git a/bertos/cpu/arm/hw/switch_ctx_arm.S b/bertos/cpu/arm/hw/switch_ctx_arm.S
index 39e9f46a..e610ac97 100644
--- a/bertos/cpu/arm/hw/switch_ctx_arm.S
+++ b/bertos/cpu/arm/hw/switch_ctx_arm.S
@@ -43,12 +43,6 @@
 asm_switch_context:
 	mrs	r2, cpsr               /* Save status. */
 
-	/*
-	 * To work correctly the lr should point to proc_exit, so when process finish
-	 * could be return
-	 */
-	push {lr}                  /* Add one element on stack, for next switching */
-
 	stmfd	sp!, {r2, r4-r11, lr}  /* Save registers. */
 
 	str     sp, [r1]               /* Save old stack pointer. */
@@ -57,5 +51,15 @@ asm_switch_context:
 	ldmfd	sp!, {r2, r4-r11, lr}  /* Load new registers. */
 	msr	cpsr, r2               /* restore flags reg. */
 
-	pop {r0}                  /* pop on stack address of the process that we want to jump */
-	mov	pc, r0
+	mov	pc, lr
+
+
+/* proc_entry trampoline needed because ARM does not pop return addresses
+from the stack, but uses lr instead.*/
+.globl asm_proc_entry
+asm_proc_entry:
+	mov	lr, pc
+	/* In r11 we have the real process entry point as set up by CPU_CREATE_NEW_STACK */
+	bx	r11
+	bl	proc_exit
+
diff --git a/bertos/cpu/frame.h b/bertos/cpu/frame.h
index 2244b3b1..8cb7a813 100644
--- a/bertos/cpu/frame.h
+++ b/bertos/cpu/frame.h
@@ -62,6 +62,8 @@
 	#define CPU_STACK_GROWS_UPWARD 0
 	#define CPU_SP_ON_EMPTY_SLOT   0
 
+
+	EXTERN_C void asm_proc_entry(void);
 	/**
 	 * Initialization value for registers in stack frame.
 	 * For the CPSR register, the initial value is set to:
@@ -72,12 +74,11 @@
 	 */
 	#define CPU_CREATE_NEW_STACK(stack, entry, exit) \
 	do { \
-		/* Process entry point */ \
+		/* LR (asm proc_entry trampoline) */ \
+		CPU_PUSH_CALL_FRAME(stack, asm_proc_entry); \
+		/* R11 (Process entry point) DO NOT CHANGE: asm_proc_entry expects \
+		 * to find the actual process entry point in R11 */ \
 		CPU_PUSH_CALL_FRAME(stack, entry); \
-		/* LR (proc_exit) */ \
-		CPU_PUSH_CALL_FRAME(stack, exit); \
-		/* R11 */ \
-		CPU_PUSH_WORD(stack, 0x11111111); \
 		/* R10 */ \
 		CPU_PUSH_WORD(stack, 0x10101010); \
 		/* R9 */ \
@@ -259,7 +260,7 @@
 
 /**
  * Default macro for creating a new Process stack
- */ 
+ */
 #ifndef CPU_CREATE_NEW_STACK
 
 	#define CPU_CREATE_NEW_STACK(stack, entry, exit) \
-- 
2.25.1