revert seqmemcpy and memcpy reimplementation

arm9/Makefile

@@ -10,7 +10,7 @@ INCDIRS := source source/common source/filesys source/crypto source/fatfs source
 INCLUDE := $(foreach dir,$(INCDIRS),-I"$(shell pwd)/$(dir)")
 
 ASFLAGS += $(SUBARCH) $(INCLUDE)
-CFLAGS  += $(SUBARCH) $(INCLUDE)
+CFLAGS  += $(SUBARCH) $(INCLUDE) -fno-builtin-memcpy
 LDFLAGS += $(SUBARCH) -Wl,-Map,$(TARGET).map
 
 include ../Makefile.common

arm9/source/common/common.h

@@ -40,9 +40,6 @@
 
 #define countof(x) \
     (sizeof(x) / sizeof((x)[0]))
-
-#define seqmemcpy(d,o,s) \
-    for(u32 i = 0; i < (s); i++) ((u8*)(void*)(d))[i] = ((u8*)(void*)(o))[i]
     
 #define bkpt \
     asm("bkpt\n\t")

arm9/source/common/memcpy.s

@@ -0,0 +1,72 @@
+@ memcpy_arm946e-s - hand written reimplementation of memcpy to be sequential
+@ Written in 2019 by luigoalma <luigoalma at gmail dot com>
+@ To the extent possible under law, the author(s) have dedicated all copyright and related and neighboring rights to this software to the public domain worldwide. This software is distributed without any warranty.
+@ For a copy of CC0 Public Domain Dedication, see <https://creativecommons.org/publicdomain/zero/1.0/>.
+	.cpu    arm946e-s
+	.arch   armv5te
+	.arm
+	.section .text.memcpy, "ax", %progbits
+	.align  2
+	.global memcpy
+	.syntax unified
+	.type   memcpy, %function
+memcpy:
+	@ r0 = dest
+	@ r1 = src
+	@ r2 = length
+	@ check if length 0 and return if so
+	cmp     r2, #0
+	bxeq    lr
+	push    {r0,r4-r9,lr}
+	@ pre-fetch data
+	pld     [r1]
+	@ alignment check with word size
+	@ if not aligned but both are in the same misalignment, fix it up
+	@ otherwise jump to basic loop
+	orr     r12, r0, r1
+	ands    r12, r12, #3
+	beq     .L1
+	and     r4, r0, #3
+	and     r5, r1, #3
+	cmp     r4, r5
+	bne     .L6
+	rsb     r4, r4, #4
+.L0:
+	ldrb    r3, [r1], #1
+	strb    r3, [r0], #1
+	subs    r2, r2, #1
+	popeq   {r0,r4-r9,pc}
+	subs    r4, r4, #1
+	bne     .L0
+.L1:
+	@ check if length higher than 32
+	@ if so, do the 32 byte block copy loop,
+	@ until there's nothing left or remainder to copy is less than 32
+	movs    r3, r2, LSR#5
+	beq     .L3
+.L2:
+	ldm     r1!, {r4-r9,r12,lr}
+	stm     r0!, {r4-r9,r12,lr}
+	subs    r3, r3, #1
+	bne     .L2
+	ands    r2, r2, #0x1F
+	popeq   {r0,r4-r9,pc}
+.L3:
+	@ copy in word size the remaining data,
+	@ and finish off with basic loop if can't copy all by word size.
+	movs    r3, r2, LSR#2
+	beq     .L6
+.L4:
+	ldr     r12, [r1], #4
+	str     r12, [r0], #4
+	subs    r3, r3, #1
+	bne     .L4
+	ands    r2, r2, #0x3
+.L5: @ the basic loop
+	popeq   {r0,r4-r9,pc}
+.L6:
+	ldrb    r3, [r1], #1
+	strb    r3, [r0], #1
+	subs    r2, r2, #1
+	b       .L5
+	.size   memcpy, .-memcpy

arm9/source/crypto/rsa.c

@@ -16,6 +16,12 @@
  *   along with this program.  If not, see <http://www.gnu.org/licenses/>.
  */
 
+/*
+ * To ensure correct functionality, the builtin memcpy should perform a sequential copy.
+ * If not, it should be replaced with a different reimplementation that does for sure act sequential.
+ * Or alternatively, it's own memcpy with it's own name, say perhaps seqmemcpy, and memcpy calls replaced.
+ */
+
 #include "common.h"
 #include "rsa.h"
 #include "sha.h"
@@ -76,7 +82,7 @@ bool RSA_setKey2048(u8 keyslot, const u8 *const mod, u32 exp)
 	REG_RSA_EXP[(0x100>>2) - 1] = exp;
 
 	if(slot->REG_RSA_SLOTSIZE != RSA_SLOTSIZE_2048) return false;
-	seqmemcpy((void*)REG_RSA_MOD, mod, 0x100);
+	memcpy((void*)REG_RSA_MOD, mod, 0x100);
 
 	return true;
 }
@@ -88,11 +94,11 @@ bool RSA_decrypt2048(void *const decSig, const void *const encSig)
 	if(!(rsaSlots[keyslot].REG_RSA_SLOTCNT & RSA_KEY_STAT_SET)) return false;
 
 	REG_RSA_CNT |= RSA_INPUT_NORMAL | RSA_INPUT_BIG;
-	seqmemcpy((void*)REG_RSA_TXT, encSig, 0x100);
+	memcpy((void*)REG_RSA_TXT, encSig, 0x100);
 
 	REG_RSA_CNT |= RSA_ENABLE;
 	rsaWaitBusy();
-	seqmemcpy(decSig, (void*)REG_RSA_TXT, 0x100);
+	memcpy(decSig, (void*)REG_RSA_TXT, 0x100);
 
 	return true;
 }

arm9/source/crypto/sha.c

@@ -1,5 +1,15 @@
+/*
+ * To ensure correct functionality, the builtin memcpy should perform a sequential copy.
+ * If not, it should be replaced with a different reimplementation that does for sure act sequential.
+ * Or alternatively, it's own memcpy with it's own name, say perhaps seqmemcpy, and memcpy calls replaced, as well the line "*((_sha_block*)REG_SHAINFIFO) = *((const _sha_block*)src32);" inside sha_update, to call explicitly seqmemcpy like, "seqmemcpy(REG_SHAINFIFO, src32, 0x40);".
+ */
 #include "sha.h"
 
+typedef struct
+{
+	u32 data[16];
+} _sha_block;
+
 void sha_init(u32 mode)
 {
     while(*REG_SHACNT & 1);
@@ -12,12 +22,12 @@ void sha_update(const void* src, u32 size)
     
     while(size >= 0x40) {
         while(*REG_SHACNT & 1);
-        seqmemcpy((void*)REG_SHAINFIFO, src32, 0x40);
+        *((_sha_block*)REG_SHAINFIFO) = *((const _sha_block*)src32);
         src32 += 16;
         size -= 0x40;
     }
     while(*REG_SHACNT & 1);
-    if(size) seqmemcpy((void*)REG_SHAINFIFO, src32, size);
+    if(size) memcpy((void*)REG_SHAINFIFO, src32, size);
 }
 
 void sha_get(void* res) {
@@ -26,7 +36,7 @@ void sha_get(void* res) {
     *REG_SHACNT = (*REG_SHACNT & ~SHA_NORMAL_ROUND) | SHA_FINAL_ROUND;
     while(*REG_SHACNT & SHA_FINAL_ROUND);
     while(*REG_SHACNT & 1);
-    if (hash_size) seqmemcpy(res, (void*)REG_SHAHASH, hash_size);
+    if (hash_size) memcpy(res, (void*)REG_SHAHASH, hash_size);
 }
 
 void sha_quick(void* res, const void* src, u32 size, u32 mode) {