From dbf72e95efd507d5a6255c25aee055a0a3c1350e Mon Sep 17 00:00:00 2001
From: David Guillen Fandos
Date: Tue, 22 Jun 2021 00:09:44 +0200
Subject: Fix the no-caller-saves bug for MIPS

Seems that ABI mandates that we allocate space for arg0..4 even if we do
pass them as registers. For some reason write_io_register<> functions
write in that stack area (1 word) corrupting the s0 saved register.
This seems to be a new gcc behaviour?
---
 Makefile        | 21 ++-------------------
 psp/mips_stub.S | 44 ++++++++++++++++++++++----------------------
 2 files changed, 24 insertions(+), 41 deletions(-)

diff --git a/Makefile b/Makefile
index d127e92..cf0bfd1 100644
--- a/Makefile
+++ b/Makefile
@@ -375,7 +375,7 @@ else ifeq ($(platform), mips32)
 	SHARED := -shared -nostdlib -Wl,--version-script=link.T
 	fpic := -fPIC -DPIC
 	CFLAGS += -fomit-frame-pointer -ffast-math -march=mips32 -mtune=mips32r2 -mhard-float
-	CFLAGS += -fno-caller-saves -DMIPS_HAS_R2_INSTS
+	CFLAGS += -DMIPS_HAS_R2_INSTS
 	HAVE_DYNAREC := 1
 	CPU_ARCH := mips
 
@@ -385,7 +385,6 @@ else ifeq ($(platform), mips64n32)
 	SHARED := -shared -nostdlib -Wl,--version-script=link.T
 	fpic := -fPIC -DPIC
 	CFLAGS += -fomit-frame-pointer -ffast-math -march=mips64 -mabi=n32 -mhard-float
-	CFLAGS += -fno-caller-saves
 	HAVE_DYNAREC := 1
 	CPU_ARCH := mips
 
@@ -394,7 +393,7 @@ else ifeq ($(platform), emscripten)
 	TARGET := $(TARGET_NAME)_libretro_$(platform).bc
 	STATIC_LINKING = 1
 
-# GCW0
+# GCW0 (OD and OD Beta)
 else ifeq ($(platform), gcw0)
 	TARGET := $(TARGET_NAME)_libretro.so
 	CC = /opt/gcw0-toolchain/usr/bin/mipsel-linux-gcc
@@ -407,22 +406,6 @@ else ifeq ($(platform), gcw0)
 	HAVE_DYNAREC := 1
 	CPU_ARCH := mips
 
-# GCW0 (OpenDingux Beta)
-else ifeq ($(platform), gcw0-odbeta)
-	TARGET := $(TARGET_NAME)_libretro.so
-	CC = /opt/gcw0-toolchain/usr/bin/mipsel-linux-gcc
-	CXX = /opt/gcw0-toolchain/usr/bin/mipsel-linux-g++
-	AR = /opt/gcw0-toolchain/usr/bin/mipsel-linux-ar
-	SHARED := -shared -nostdlib -Wl,--version-script=link.T
-	fpic := -fPIC -DPIC
-	CFLAGS += -fomit-frame-pointer -ffast-math -march=mips32 -mtune=mips32r2 -mhard-float
-	# The ASM code and/or MIPS dynarec of GPSP does not respect
-	# MIPS calling conventions, so we must use '-fno-caller-saves'
-	# for the OpenDingux Beta build
-	CFLAGS += -fno-caller-saves -DMIPS_HAS_R2_INSTS
-	HAVE_DYNAREC := 1
-	CPU_ARCH := mips
-
 # Windows
 else
 	TARGET := $(TARGET_NAME)_libretro.dll
diff --git a/psp/mips_stub.S b/psp/mips_stub.S
index 59cf2a0..7b9bcb0 100644
--- a/psp/mips_stub.S
+++ b/psp/mips_stub.S
@@ -293,18 +293,18 @@ mips_cheat_hook:
 # ARM regs must be saved before branching here
 return_to_main:
   REG_L $28, GP_SAVE($16)            # Restore previous state
-  REG_L $s0, 0*SZREG($sp)
-  REG_L $s1, 1*SZREG($sp)
-  REG_L $s2, 2*SZREG($sp)
-  REG_L $s3, 3*SZREG($sp)
-  REG_L $s4, 4*SZREG($sp)
-  REG_L $s5, 5*SZREG($sp)
-  REG_L $s6, 6*SZREG($sp)
-  REG_L $s7, 7*SZREG($sp)
-  REG_L $fp, 8*SZREG($sp)
-  REG_L $ra, 9*SZREG($sp)
+  REG_L $s0,  4*SZREG($sp)
+  REG_L $s1,  5*SZREG($sp)
+  REG_L $s2,  6*SZREG($sp)
+  REG_L $s3,  7*SZREG($sp)
+  REG_L $s4,  8*SZREG($sp)
+  REG_L $s5,  9*SZREG($sp)
+  REG_L $s6, 10*SZREG($sp)
+  REG_L $s7, 11*SZREG($sp)
+  REG_L $fp, 12*SZREG($sp)
+  REG_L $ra, 13*SZREG($sp)
   jr $ra                          # Return to main
-  addiu $sp, $sp, 80              # Restore stack pointer (delay slot)
+  addiu $sp, $sp, 112             # Restore stack pointer (delay slot)
 
 # Perform an indirect branch.
 
@@ -590,17 +590,17 @@ asr_shift_high:
 
 execute_arm_translate_internal:
 
-  addiu $sp, $sp, -80             # Store the main thread context
-  REG_S $s0, 0*SZREG($sp)
-  REG_S $s1, 1*SZREG($sp)
-  REG_S $s2, 2*SZREG($sp)
-  REG_S $s3, 3*SZREG($sp)
-  REG_S $s4, 4*SZREG($sp)
-  REG_S $s5, 5*SZREG($sp)
-  REG_S $s6, 6*SZREG($sp)
-  REG_S $s7, 7*SZREG($sp)
-  REG_S $fp, 8*SZREG($sp)
-  REG_S $ra, 9*SZREG($sp)
+  addiu $sp, $sp, -112            # Store the main thread context
+  REG_S $s0,  4*SZREG($sp)
+  REG_S $s1,  5*SZREG($sp)
+  REG_S $s2,  6*SZREG($sp)
+  REG_S $s3,  7*SZREG($sp)
+  REG_S $s4,  8*SZREG($sp)
+  REG_S $s5,  9*SZREG($sp)
+  REG_S $s6, 10*SZREG($sp)
+  REG_S $s7, 11*SZREG($sp)
+  REG_S $fp, 12*SZREG($sp)
+  REG_S $ra, 13*SZREG($sp)
 
   move $16, $5
   REG_S $28, GP_SAVE($16)
-- 
cgit v1.2.3