diff --git a/.gitignore b/.gitignore
index 7758f82f..01c7389f 100644
--- a/.gitignore
+++ b/.gitignore
@@ -119,3 +119,6 @@ IDE/IAR/Release
 build/
 CMakeFiles/
 CMakeCache.txt
+
+# Stage 1
+stage1/loader_stage1.ld
diff --git a/Makefile b/Makefile
index d9ea419d..03dd1f08 100644
--- a/Makefile
+++ b/Makefile
@@ -46,6 +46,10 @@ include arch.mk
 # Parse config options
 include options.mk
 
+OBJS+=$(WOLFCRYPT_OBJS)
+OBJS+=$(PUBLIC_KEY_OBJS)
+OBJS+=$(UPDATE_OBJS)
+
 CFLAGS+= \
   -I"." -I"include/" -I"lib/wolfssl" \
   -D"WOLFSSL_USER_SETTINGS" \
@@ -92,7 +96,7 @@ ifeq ($(TARGET),sim)
 endif
 
 ifeq ($(TARGET),nxp_p1021)
-	MAIN_TARGET:=loader_stage1.bin factory.bin
+	MAIN_TARGET:=factory_wstage1.bin
 endif
 
 ASFLAGS:=$(CFLAGS)
@@ -100,11 +104,10 @@ BOOTLOADER_PARTITION_SIZE?=$$(( $(WOLFBOOT_PARTITION_BOOT_ADDRESS) - $(ARCH_FLAS
 
 all: $(MAIN_TARGET)
 
-stage1: loader_stage1.bin
-loader_stage1: loader_stage1.bin
-loader_stage1.bin:
+stage1: stage1/loader_stage1.bin
+stage1/loader_stage1.bin:
 	@echo "\t[BIN] $@"
-	$(Q)$(MAKE) -C stage1 $@
+	$(Q)$(MAKE) -C $(dir $@) $(notdir $@)
 
 test-lib: $(OBJS)
 	$(Q)$(CC) $(CFLAGS) -o $@ $^
@@ -181,7 +184,15 @@ internal_flash.dd: test-app/image_v1_signed.bin wolfboot.elf $(BINASSEMBLE)
 
 factory.bin: $(BOOT_IMG) wolfboot.bin $(PRIVATE_KEY) test-app/image_v1_signed.bin $(BINASSEMBLE)
 	@echo "\t[MERGE] $@"
-	$(Q)$(BINASSEMBLE) $@ $(WOLFBOOT_ORIGIN) wolfboot.bin \
+	$(Q)$(BINASSEMBLE) $@ \
+		$(WOLFBOOT_ORIGIN) wolfboot.bin \
+		$(WOLFBOOT_PARTITION_BOOT_ADDRESS) test-app/image_v1_signed.bin
+
+factory_wstage1.bin: $(BOOT_IMG) wolfboot.bin $(PRIVATE_KEY) test-app/image_v1_signed.bin $(BINASSEMBLE) stage1/loader_stage1.bin
+	@echo "\t[MERGE] $@"
+	$(Q)$(BINASSEMBLE) $@ \
+		$(WOLFBOOT_STAGE1_FLASH_ADDR) stage1/loader_stage1.bin \
+		$(WOLFBOOT_ORIGIN) wolfboot.bin \
 		$(WOLFBOOT_PARTITION_BOOT_ADDRESS) test-app/image_v1_signed.bin
 
 wolfboot.elf: include/target.h $(OBJS) $(LSCRIPT) FORCE
@@ -207,6 +218,7 @@ $(LSCRIPT): FORCE
 		sed -e "s/@WOLFBOOT_PARTITION_SIZE@/$(WOLFBOOT_PARTITION_SIZE)/g" | \
 		sed -e "s/@WOLFBOOT_PARTITION_UPDATE_ADDRESS@/$(WOLFBOOT_PARTITION_UPDATE_ADDRESS)/g" | \
 		sed -e "s/@WOLFBOOT_PARTITION_SWAP_ADDRESS@/$(WOLFBOOT_PARTITION_SWAP_ADDRESS)/g" | \
+		sed -e "s/@WOLFBOOT_STAGE1_SIZE@/$(WOLFBOOT_STAGE1_SIZE)/g" | \
 		sed -e "s/@WOLFBOOT_STAGE1_LOAD_ADDR@/$(WOLFBOOT_STAGE1_LOAD_ADDR)/g" | \
 		sed -e "s/@WOLFBOOT_STAGE1_FLASH_ADDR@/$(WOLFBOOT_STAGE1_FLASH_ADDR)/g" | \
 		sed -e "s/@WOLFBOOT_STAGE1_BASE_ADDR@/$(WOLFBOOT_STAGE1_BASE_ADDR)/g" \
diff --git a/arch.mk b/arch.mk
index b06bc1f0..bde56984 100644
--- a/arch.mk
+++ b/arch.mk
@@ -222,7 +222,7 @@ ifeq ($(ARCH),PPC)
   endif
 
   # Prune unused functions and data
-  CFLAGS +=-ffunction-sections -fdata-sections
+  CFLAGS+=-ffunction-sections -fdata-sections
   LDFLAGS+=-Wl,--gc-sections
 
   OBJS+=src/boot_ppc_start.o src/boot_ppc.o
@@ -491,3 +491,5 @@ endif
 ifeq ($(DEBUG_UART),1)
   CFLAGS+=-DDEBUG_UART
 endif
+
+CFLAGS+=-DWOLFBOOT_ARCH=$(ARCH)
diff --git a/config/examples/nxp-p1021.config b/config/examples/nxp-p1021.config
index 23d544ab..ee20155e 100644
--- a/config/examples/nxp-p1021.config
+++ b/config/examples/nxp-p1021.config
@@ -25,6 +25,9 @@ WOLFTPM?=0
 # Flash Sector (Block) Size (16KB)
 WOLFBOOT_SECTOR_SIZE=0x4000
 
+# Maximum size of wolfBoot stage 1 loader
+WOLFBOOT_STAGE1_SIZE=0x1000
+
 # wolfBoot partition size (128KB)
 BOOTLOADER_PARTITION_SIZE=0x20000
 
@@ -34,7 +37,7 @@ WOLFBOOT_PARTITION_SIZE?=0x1000000
 
 ## NAND Flash Offsets ##
 # Location in Flash for stage 1 loader (XIP from boot ROM)
-WOLFBOOT_STAGE1_FLASH_ADDR = 0x0
+WOLFBOOT_STAGE1_FLASH_ADDR=0x0
 # Location in Flash for wolfBoot
 WOLFBOOT_ORIGIN=0x4000
 # Location in Flash for Application Partition
@@ -44,7 +47,6 @@ WOLFBOOT_PARTITION_UPDATE_ADDRESS?=0x1200000
 # Location in Flash for temporary sector used during updates
 WOLFBOOT_PARTITION_SWAP_ADDRESS?=  0x2200000
 
-
 ## Loader Destinations (DDR) ##
 
 # Address in RAM to load stage 1 loader
diff --git a/config/examples/nxp-t2080.config b/config/examples/nxp-t2080.config
index 8e749353..10d04d34 100644
--- a/config/examples/nxp-t2080.config
+++ b/config/examples/nxp-t2080.config
@@ -1,8 +1,9 @@
 ARCH=PPC
 TARGET=nxp_t2080
-SIGN?=ED25519
-HASH?=SHA256
-DEBUG?=1
+SIGN?=ECC384
+HASH?=SHA384
+DEBUG?=0
+DEBUG_UART?=1
 VTOR?=1
 CORTEX_M0?=0
 NO_ASM?=0
diff --git a/docs/Loader.md b/docs/Loader.md
index 41ba9ca9..7738dc95 100644
--- a/docs/Loader.md
+++ b/docs/Loader.md
@@ -10,10 +10,13 @@ A first stage loader whose purpose is to load wolfBoot from flash to ram and jum
 
 Example: `make WOLFBOOT_STAGE1_LOAD_ADDR=0x1000 stage1`
 
+* `WOLFBOOT_STAGE1_SIZE`: Maximum size of wolfBoot stage 1 loader
+* `WOLFBOOT_STAGE1_FLASH_ADDR`: Location in Flash for stage 1 loader (XIP from boot ROM)
 * `WOLFBOOT_STAGE1_BASE_ADDR`: Address in RAM to load stage 1 loader to
 * `WOLFBOOT_STAGE1_LOAD_ADDR`: Address in RAM to load wolfBoot to
 * `WOLFBOOT_LOAD_ADDRESS`: Address in RAM to load application partition
 
+
 ## update_ram.c
 
 Implementation for RAM based updater
diff --git a/docs/Targets.md b/docs/Targets.md
index 9542658a..9a0848d4 100644
--- a/docs/Targets.md
+++ b/docs/Targets.md
@@ -1081,22 +1081,27 @@ The NXP QorIQ P1021 is a PPC e500v2 based processor (two cores). This has been t
 
 wolfBoot supports loading from external flash using the eLBC FMC (Flash Machine) with NAND.
 
-When each e500 core comes out of reset, its MMU has one 4-Kbyte page defined at `0x0_FFFF_Fnnn`. For NAND boot the first 4KB is loaded to this region with the first offset jump instruction at `0x0_FFFF_FFFC`. The 4KB is mapped to the FCM buffers.
+When each e500 core comes out of reset, its MMU has one 4-Kbyte page defined at `0x0_FFFF_Fnnn`. For NAND boot the first 4KB is loaded to this region with the first offset jump instruction at `0x0_FFFF_FFFC`. The 4KB is mapped to the eLBC FCM buffers.
 
 This device defines the default boot ROM address range to be 8 Mbytes at address `0x0_FF80_0000` to `0x0_FFFF_FFFF`.
 
-cfg_rom_loc[0:3] = 1000 Local bus FCM-8-bit NAND flash small page
-cfg_rom_loc[0:3] = 1010 Local bus FCM-8-bit NAND flash large page
+These pin determine if the boot ROM will use small or large flash page:
+* `cfg_rom_loc[0:3]` = 1000 Local bus FCM-8-bit NAND flash small page
+* `cfg_rom_loc[0:3]` = 1010 Local bus FCM-8-bit NAND flash large page
 
 If the boot sequencer is not enabled, the processor cores exit reset and fetches boot code in default configurations.
 
 A loader must reside in the 4KB page to handle early startup including DDR and then load wolfBoot into DDR for execution.
 
+### First Stage Loader (stage 1) for NXP P1021 PPC
+
+A first stage loader is required to load the wolfBoot image into DDR for execution. This is because only 4KB of code space is available on boot. The stage 1 loader must also copy iteslf from the FCM buffer to DDR (or L2SRAM) to allow using of the eLBC to read NAND blocks.
+
 ### Building wolfBoot for NXP P1021 PPC
 
 By default wolfBoot will use `powerpc-linux-gnu-` cross-compiler prefix. These tools can be installed with the Debian package `gcc-powerpc-linux-gnu` (`sudo apt install gcc-powerpc-linux-gnu`).
 
-The `make` creates a `factory.bin` image that can be programmed at `0x00000000`
+The `make` creates a `factory_wstage1.bin` image that can be programmed at `0x00000000`
 
 ```
 cp ./config/examples/nxp-p1021.config .config
@@ -1104,6 +1109,16 @@ make keytools
 make clean && make
 ```
 
+#### Flash Layout for NXP P1021 PPC (default)
+
+| File                         | NAND offset |
+| ---------------------------- | ----------- |
+| stage1/loader_stage1.bin     | 0x00000000  |
+| wolfboot.bin                 | 0x00004000  |
+| test-app/image_v1_signed.bin | 0x00200000  |
+| update                       | 0x01200000  |
+| swap block                   | 0x02200000  |
+
 
 ## NXP QorIQ T2080 PPC
 
diff --git a/hal/nxp_p1021.c b/hal/nxp_p1021.c
index ee8385a3..6295e84f 100644
--- a/hal/nxp_p1021.c
+++ b/hal/nxp_p1021.c
@@ -951,23 +951,6 @@ void hal_init(void)
     (void)svr;
 #endif
 
-#ifdef BUILD_LOADER_STAGE1
-    #ifdef WOLFBOOT_STAGE1_BASE_ADDR
-    /* if this is executing from boot 4KB region (FCM buffer) it must
-     * first be relocated to RAM before the eLBC NAND can be read */
-    if (((size_t)&hal_init & BOOT_ROM_ADDR) == BOOT_ROM_ADDR) {
-        __attribute__((noreturn)) void (*wolfboot_start)(void) =
-            (void*)BOOT_ROM_ADDR + BOOT_ROM_SIZE - 0x04;
-
-        /* relocate 4KB code to DST and jump */
-        memmove((void*)WOLFBOOT_STAGE1_BASE_ADDR, (void*)BOOT_ROM_ADDR,
-            BOOT_ROM_SIZE);
-
-        wolfboot_start(); /* never returns */
-    }
-    #endif
-#endif
-
 #ifdef ENABLE_PCIE
     hal_pcie_init();
 #endif
diff --git a/hal/nxp_p1021.ld b/hal/nxp_p1021.ld
index b995afcc..4675f0fc 100644
--- a/hal/nxp_p1021.ld
+++ b/hal/nxp_p1021.ld
@@ -2,17 +2,17 @@ OUTPUT_ARCH( "powerpc" )
 
 ENTRY( _reset )
 
+HEAP_SIZE = 4K; /* heap not used */
+STACK_SIZE = 128K;
+
 MEMORY
 {
-    /* L2 as SRAM (up to 256KB) */
-    RAM (rwx) : ORIGIN = 0xF8F80000, LENGTH = 64K
-
-    /* L1 as SRAM (up to 16KB) */
-    LRAM (rwx) : ORIGIN = 0xFFD00000, LENGTH = 16K
-
     /* DDR3 - 512MB (offset by destination address and 4KB boot region) */
     DRAM (rwx) : ORIGIN = @WOLFBOOT_STAGE1_LOAD_ADDR@,
                  LENGTH = 0x1FFFFFFF - @WOLFBOOT_STAGE1_LOAD_ADDR@
+
+    /* L1 as SRAM (up to 16KB) */
+    RAM (rwx) : ORIGIN = 0xFFD00000, LENGTH = 16K
 }
 
 SECTIONS
@@ -77,5 +77,5 @@ SECTIONS
     } > DRAM
 }
 
-PROVIDE(_start_heap = ORIGIN(DRAM));
-PROVIDE(_end_stack  = ORIGIN(LRAM) + (LENGTH(LRAM)) );
+PROVIDE(_start_heap = _end);
+PROVIDE(_end_stack  = _end + HEAP_SIZE + STACK_SIZE );
diff --git a/hal/nxp_p1021_stage1.ld b/hal/nxp_p1021_stage1.ld
index eb772c93..c72d2ada 100644
--- a/hal/nxp_p1021_stage1.ld
+++ b/hal/nxp_p1021_stage1.ld
@@ -5,25 +5,21 @@ ENTRY( _reset )
 BASE_ADDR = @WOLFBOOT_STAGE1_BASE_ADDR@;
 
 /* for flashing to NAND it must be 4KB */
-/* if debugging and running from RAM this can be increased */
-MAX_LOADER_SIZE = 0x1000; /* 4KB */
+LOADER_STAGE1_SIZE = @WOLFBOOT_STAGE1_SIZE@;
 
 /* Boot initialization code */
 BOOTSTRAP_TLB = BASE_ADDR;
 
 /* Entry point where boot executes from (end of region) */
-BOOTSTRAP_ENTRY = BASE_ADDR + MAX_LOADER_SIZE - 4;
+BOOTSTRAP_ENTRY = BASE_ADDR + LOADER_STAGE1_SIZE - 4;
 
 MEMORY
 {
     /* Boot Location */
-    FLASH (rx) : ORIGIN = BOOTSTRAP_TLB, LENGTH = MAX_LOADER_SIZE
-
-    /* L2 as SRAM (up to 256KB) */
-    RAM (rwx) : ORIGIN = BASE_ADDR + MAX_LOADER_SIZE, LENGTH = 128K - MAX_LOADER_SIZE
+    FLASH (rx) : ORIGIN = BOOTSTRAP_TLB, LENGTH = LOADER_STAGE1_SIZE
 
     /* L1 as SRAM (up to 16KB) */
-    L1RAM (rwx) : ORIGIN = 0xffd00000, LENGTH = 16K
+    RAM (rwx) : ORIGIN = 0xffd00000, LENGTH = 16K
 }
 
 SECTIONS
@@ -84,4 +80,4 @@ SECTIONS
 }
 
 PROVIDE(_start_heap = _end);
-PROVIDE(_end_stack  = ORIGIN(L1RAM) + (LENGTH(L1RAM)) );
+PROVIDE(_end_stack  = ORIGIN(RAM) + (LENGTH(RAM)) );
diff --git a/hal/nxp_ppc.h b/hal/nxp_ppc.h
index db11257d..d4e9e54e 100644
--- a/hal/nxp_ppc.h
+++ b/hal/nxp_ppc.h
@@ -35,10 +35,6 @@
 /* Relocate CCSRBAR */
 #define CCSRBAR 0xFFE00000
 
-/* boot address */
-#define BOOT_ROM_ADDR 0xFFFFF000UL
-#define BOOT_ROM_SIZE (4*1024)
-
 /* Memory used for transferring blocks to/from NAND.
  * Maps to eLBC FCM internal 8KB region (by hardware) */
 #define FLASH_BASE_ADDR 0xFC000000
@@ -65,6 +61,10 @@
 #error Please define MMU version and CCSRBAR for platform
 #endif
 
+/* boot address */
+#define BOOT_ROM_ADDR 0xFFFFF000UL
+#define BOOT_ROM_SIZE (4*1024)
+
 #define RESET_VECTOR (BOOT_ROM_ADDR + (BOOT_ROM_SIZE - 4))
 
 #ifndef CCSRBAR_DEF
diff --git a/include/image.h b/include/image.h
index 8f2563bf..0e970734 100644
--- a/include/image.h
+++ b/include/image.h
@@ -34,6 +34,15 @@ extern "C" {
 #include "target.h"
 #include "wolfboot/wolfboot.h"
 
+#ifdef EXT_FLASH
+#include "hal.h"
+#endif
+
+#if defined(EXT_ENCRYPTED) && (defined(__WOLFBOOT) || defined(UNIT_TEST))
+#include "encrypt.h"
+#endif
+
+
 int wolfBot_get_dts_size(void *dts_addr);
 
 
@@ -576,11 +585,9 @@ uint16_t wolfBoot_find_header(uint8_t *haystack, uint16_t type, uint8_t **ptr);
         ((pn == PART_UPDATE || pn == PART_DTS_UPDATE)?UPDATE_EXT: \
             ((pn == PART_SWAP)?SWAP_EXT:0)))
 # define PART_IS_EXT(x)  PARTN_IS_EXT(((x)->part))
-#include "hal.h"
 
 
 #if defined(EXT_ENCRYPTED) && (defined(__WOLFBOOT) || defined(UNIT_TEST))
-#include "encrypt.h"
 #define ext_flash_check_write ext_flash_encrypt_write
 #define ext_flash_check_read ext_flash_decrypt_read
 #else
@@ -588,7 +595,8 @@ uint16_t wolfBoot_find_header(uint8_t *haystack, uint16_t type, uint8_t **ptr);
 #define ext_flash_check_read ext_flash_read
 #endif
 
-static inline int wb_flash_erase(struct wolfBoot_image *img, uint32_t off, uint32_t size)
+static inline int wb_flash_erase(struct wolfBoot_image *img, uint32_t off,
+    uint32_t size)
 {
     if (PART_IS_EXT(img))
         return ext_flash_erase((uintptr_t)(img->hdr) + off, size);
@@ -596,7 +604,8 @@ static inline int wb_flash_erase(struct wolfBoot_image *img, uint32_t off, uint3
         return hal_flash_erase((uintptr_t)(img->hdr) + off, size);
 }
 
-static inline int wb_flash_write(struct wolfBoot_image *img, uint32_t off, const void *data, uint32_t size)
+static inline int wb_flash_write(struct wolfBoot_image *img, uint32_t off,
+    const void *data, uint32_t size)
 {
     if (PART_IS_EXT(img))
         return ext_flash_check_write((uintptr_t)(img->hdr) + off, data, size);
@@ -604,18 +613,22 @@ static inline int wb_flash_write(struct wolfBoot_image *img, uint32_t off, const
         return hal_flash_write((uintptr_t)(img->hdr) + off, data, size);
 }
 
-static inline int wb_flash_write_verify_word(struct wolfBoot_image *img, uint32_t off, uint32_t word)
+static inline int wb_flash_write_verify_word(struct wolfBoot_image *img,
+    uint32_t off, uint32_t word)
 {
     int ret;
     volatile uint32_t copy;
     if (PART_IS_EXT(img))
     {
-        ext_flash_check_read((uintptr_t)(img->hdr) + off, (void *)&copy, sizeof(uint32_t));
+        ext_flash_check_read((uintptr_t)(img->hdr) + off, (void *)&copy,
+            sizeof(uint32_t));
         while (copy != word) {
-            ret = ext_flash_check_write((uintptr_t)(img->hdr) + off, (void *)&word, sizeof(uint32_t));
+            ret = ext_flash_check_write((uintptr_t)(img->hdr) + off,
+                (void *)&word, sizeof(uint32_t));
             if (ret < 0)
                 return ret;
-            ext_flash_check_read((uintptr_t)(img->hdr) + off, (void *)&copy, sizeof(uint32_t));
+            ext_flash_check_read((uintptr_t)(img->hdr) + off, (void *)&copy,
+                sizeof(uint32_t));
         }
     } else {
         volatile uint32_t *pcopy = (volatile uint32_t*)(img->hdr + off);
@@ -631,8 +644,10 @@ static inline int wb_flash_write_verify_word(struct wolfBoot_image *img, uint32_
 
 # define PART_IS_EXT(x) (0)
 # define PARTN_IS_EXT(x) (0)
-# define wb_flash_erase(im, of, siz)  hal_flash_erase(((uintptr_t)(((im)->hdr)) + of), siz)
-# define wb_flash_write(im, of, dat, siz)  hal_flash_write(((uintptr_t)((im)->hdr)) + of, dat, siz)
+# define wb_flash_erase(im, of, siz) \
+    hal_flash_erase(((uintptr_t)(((im)->hdr)) + of), siz)
+# define wb_flash_write(im, of, dat, siz) \
+    hal_flash_write(((uintptr_t)((im)->hdr)) + of, dat, siz)
 
 #endif /* EXT_FLASH */
 
@@ -645,9 +660,10 @@ static inline int wb_flash_write_verify_word(struct wolfBoot_image *img, uint32_
 #define UBOOT_FDT_MAGIC	    0xEDFE0DD0UL
 
 #ifndef EXT_ENCRYPTED
-#define WOLFBOOT_MAX_SPACE (WOLFBOOT_PARTITION_SIZE - (TRAILER_SKIP + sizeof(uint32_t) + (WOLFBOOT_PARTITION_SIZE + 1 / (WOLFBOOT_SECTOR_SIZE * 8))))
+#define WOLFBOOT_MAX_SPACE (WOLFBOOT_PARTITION_SIZE - \
+    (TRAILER_SKIP + sizeof(uint32_t) + \
+    (WOLFBOOT_PARTITION_SIZE + 1 / (WOLFBOOT_SECTOR_SIZE * 8))))
 #else
-#include "encrypt.h"
 #define WOLFBOOT_MAX_SPACE (WOLFBOOT_PARTITION_SIZE - ENCRYPT_TMP_SECRET_OFFSET)
 #endif
 
diff --git a/include/user_settings.h b/include/user_settings.h
index 1a97c79d..52a44358 100644
--- a/include/user_settings.h
+++ b/include/user_settings.h
@@ -34,6 +34,12 @@
 #define WOLFCRYPT_ONLY
 #define SIZEOF_LONG_LONG 8
 
+#define CTYPE_USER /* don't let wolfCrypt types.h include ctype.h */
+extern int toupper(int c);
+extern int tolower(int c);
+#define XTOUPPER(c)     toupper((c))
+#define XTOLOWER(c)     tolower((c))
+
 #ifdef USE_FAST_MATH
 #   define WC_NO_HARDEN
 #endif
diff --git a/options.mk b/options.mk
index f2853262..4fbef934 100644
--- a/options.mk
+++ b/options.mk
@@ -376,8 +376,9 @@ else
   endif
 endif
 
-ifeq ($(V),0)
-  Q=@
+Q?=@
+ifeq ($(V),1)
+  Q=
 endif
 
 ifeq ($(NO_MPU),1)
@@ -415,9 +416,6 @@ ifeq ($(WOLFBOOT_HUGE_STACK),1)
   CFLAGS+=-DWOLFBOOT_HUGE_STACK
 endif
 
-OBJS+=$(PUBLIC_KEY_OBJS)
-OBJS+=$(UPDATE_OBJS)
-
 ifeq ($(WOLFTPM),1)
   OBJS += lib/wolfTPM/src/tpm2.o \
     lib/wolfTPM/src/tpm2_packet.o \
@@ -466,7 +464,6 @@ ifeq ($(HASH),SHA3)
 endif
 
 CFLAGS+=-DIMAGE_HEADER_SIZE=$(IMAGE_HEADER_SIZE)
-OBJS+=$(WOLFCRYPT_OBJS)
 
 # check if both encryption and self update are on
 #
diff --git a/src/boot_ppc_start.S b/src/boot_ppc_start.S
index 3044ab9e..30b8de64 100644
--- a/src/boot_ppc_start.S
+++ b/src/boot_ppc_start.S
@@ -412,7 +412,7 @@ setup_l2:
         stw     r4, 0(r6) /* write L2SC0 MMR to flash invalidate L2 cache and locks */
 l2loop:
         sync
-        lwz     r5, 0(6r) /* get current L2SC0 MMR value */
+        lwz     r5, 0(r6) /* get current L2SC0 MMR value */
         and.    r5, r5, r4 /* compare to mask to see if complete */
         bne     l2loop
         isync
diff --git a/src/boot_renesas.c b/src/boot_renesas.c
old mode 100755
new mode 100644
diff --git a/src/loader.c b/src/loader.c
old mode 100755
new mode 100644
diff --git a/src/string.c b/src/string.c
index f3aa7ce1..15b30eee 100644
--- a/src/string.c
+++ b/src/string.c
@@ -37,6 +37,14 @@ size_t strlen(const char *s); /* forward declaration */
     #endif
 #endif
 
+#if !defined(__IAR_SYSTEMS_ICC__) && !defined(PLATFORM_X86_64_EFI)
+/* for RAMFUNCTION */
+#include "image.h"
+#endif
+
+#if !defined(BUILD_LOADER_STAGE1) || \
+    (defined(PRINTF_ENABLED) && defined(DEBUG_UART))
+
 int islower(int c)
 {
     return (c >= 'a' && c <= 'z');
@@ -137,16 +145,6 @@ int strncasecmp(const char *s1, const char *s2, size_t n)
     return diff;
 }
 
-size_t strlen(const char *s)
-{
-    size_t i = 0;
-
-    while (s[i] != 0)
-        i++;
-
-    return i;
-}
-
 char *strncat(char *dest, const char *src, size_t n)
 {
     size_t i = 0;
@@ -179,22 +177,6 @@ int strncmp(const char *s1, const char *s2, size_t n)
     return diff;
 }
 
-#if  !defined(__IAR_SYSTEMS_ICC__) && !defined(PLATFORM_X86_64_EFI)
-#include "image.h"
-void RAMFUNCTION *memcpy(void *dst, const void *src, size_t n)
-{
-    size_t i;
-    const char *s = (const char *)src;
-    char *d = (char *)dst;
-
-    for (i = 0; i < n; i++) {
-        d[i] = s[i];
-    }
-
-    return dst;
-}
-#endif /* IAR */
-
 char *strncpy(char *dst, const char *src, size_t n)
 {
     size_t i;
@@ -222,7 +204,6 @@ char *strcpy(char *dst, const char *src)
     return dst;
 }
 
-
 int memcmp(const void *_s1, const void *_s2, size_t n)
 {
     int diff = 0;
@@ -239,6 +220,35 @@ int memcmp(const void *_s1, const void *_s2, size_t n)
     return diff;
 }
 
+#endif /* !BUILD_LOADER_STAGE1 || (PRINTF_ENABLED && DEBUG_UART) */
+
+#if !defined(BUILD_LOADER_STAGE1) || defined(DEBUG_UART)
+size_t strlen(const char *s)
+{
+    size_t i = 0;
+
+    while (s[i] != 0)
+        i++;
+
+    return i;
+}
+#endif
+
+#if  !defined(__IAR_SYSTEMS_ICC__) && !defined(PLATFORM_X86_64_EFI)
+void RAMFUNCTION *memcpy(void *dst, const void *src, size_t n)
+{
+    size_t i;
+    const char *s = (const char *)src;
+    char *d = (char *)dst;
+
+    for (i = 0; i < n; i++) {
+        d[i] = s[i];
+    }
+
+    return dst;
+}
+#endif /* IAR */
+
 #ifndef __IAR_SYSTEMS_ICC__
 void *memmove(void *dst, const void *src, size_t n)
 {
@@ -258,6 +268,7 @@ void *memmove(void *dst, const void *src, size_t n)
 }
 #endif
 
+
 #if defined(PRINTF_ENABLED) && defined(DEBUG_UART)
 void uart_writenum(int num, int base)
 {
diff --git a/stage1/Makefile b/stage1/Makefile
old mode 100755
new mode 100644
index d9dc4b69..395bf63a
--- a/stage1/Makefile
+++ b/stage1/Makefile
@@ -12,7 +12,7 @@ WOLFBOOT_ROOT?=$(PWD)
 CFLAGS:=-D"__WOLFBOOT"
 CFLAGS+=-Werror -Wextra
 LDFLAGS:=
-LSCRIPT:=../config/loader_stage1.ld
+LSCRIPT:=loader_stage1.ld
 LSCRIPT_IN:=../hal/$(TARGET)_stage1.ld
 LSCRIPT_FLAGS:=
 LD_START_GROUP:=-Wl,--start-group
@@ -24,7 +24,7 @@ DEBUG_UART?=0
 OBJS:= \
 	./hal/$(TARGET).o \
 	./src/string.o \
-	./src/loader_stage1.o
+	loader_stage1.o
 
 WOLFCRYPT_OBJS:=
 PUBLIC_KEY_OBJS:=
@@ -55,12 +55,21 @@ TARGET_H_TEMPLATE:=../include/target.h.in
 ASFLAGS:=$(CFLAGS)
 BOOTLOADER_PARTITION_SIZE?=$$(( $(WOLFBOOT_PARTITION_BOOT_ADDRESS) - $(ARCH_FLASH_OFFSET)))
 
+ifeq ($(WOLFBOOT_STAGE1_SIZE),)
+	# default is end of 4KB region (0x0FFC)
+	WOLFBOOT_STAGE1_SIZE=0x1000
+endif
+
 CFLAGS+=\
-	-DBUILD_LOADER_STAGE1 -DNO_PRINTF_UART \
+	-DBUILD_LOADER_STAGE1 \
+	-DWOLFBOOT_STAGE1_SIZE=$(WOLFBOOT_STAGE1_SIZE) \
 	-DWOLFBOOT_STAGE1_LOAD_ADDR=$(WOLFBOOT_STAGE1_LOAD_ADDR) \
 	-DWOLFBOOT_STAGE1_FLASH_ADDR=$(WOLFBOOT_STAGE1_FLASH_ADDR) \
 	-DWOLFBOOT_STAGE1_BASE_ADDR=$(WOLFBOOT_STAGE1_BASE_ADDR)
 
+# For printf support (disable NO_PRINTF_UART) and increase WOLFBOOT_STAGE1_SIZE
+CFLAGS+=-DNO_PRINTF_UART
+
 BUILD_DIR=.
 LS1_OBJS=$(addprefix $(BUILD_DIR)/, $(notdir $(OBJS)))
 vpath %.c $(dir ../src)
@@ -103,6 +112,7 @@ $(LSCRIPT): FORCE
 		sed -e "s/@WOLFBOOT_PARTITION_SIZE@/$(WOLFBOOT_PARTITION_SIZE)/g" | \
 		sed -e "s/@WOLFBOOT_PARTITION_UPDATE_ADDRESS@/$(WOLFBOOT_PARTITION_UPDATE_ADDRESS)/g" | \
 		sed -e "s/@WOLFBOOT_PARTITION_SWAP_ADDRESS@/$(WOLFBOOT_PARTITION_SWAP_ADDRESS)/g" | \
+		sed -e "s/@WOLFBOOT_STAGE1_SIZE@/$(WOLFBOOT_STAGE1_SIZE)/g" | \
 		sed -e "s/@WOLFBOOT_STAGE1_LOAD_ADDR@/$(WOLFBOOT_STAGE1_LOAD_ADDR)/g" | \
 		sed -e "s/@WOLFBOOT_STAGE1_FLASH_ADDR@/$(WOLFBOOT_STAGE1_FLASH_ADDR)/g" | \
 		sed -e "s/@WOLFBOOT_STAGE1_BASE_ADDR@/$(WOLFBOOT_STAGE1_BASE_ADDR)/g" \
@@ -110,7 +120,7 @@ $(LSCRIPT): FORCE
 
 %.hex:%.elf
 	@echo "\t[ELF2HEX] $@"
-	@$(OBJCOPY) -O ihex $^ $@
+	$(Q)@$(OBJCOPY) -O ihex $^ $@
 
 ../include/target.h: $(TARGET_H_TEMPLATE) FORCE
 	@cat $(TARGET_H_TEMPLATE) | \
@@ -150,7 +160,7 @@ $(BUILD_DIR)/%.o: ../hal/%.S
 
 clean:
 	$(Q)rm -f *.o
-	$(Q)rm -f loader_stage1.bin loader_stage1.elf loader_stage1.map ../config/loader_stage1.ld
+	$(Q)rm -f loader_stage1.bin loader_stage1.elf loader_stage1.map loader_stage1.ld
 
 FORCE:
 
diff --git a/src/loader_stage1.c b/stage1/loader_stage1.c
similarity index 58%
rename from src/loader_stage1.c
rename to stage1/loader_stage1.c
index fd2ff593..d7e577f7 100644
--- a/src/loader_stage1.c
+++ b/stage1/loader_stage1.c
@@ -28,20 +28,29 @@
 #include "spi_flash.h"
 #include "printf.h"
 #include "wolfboot/wolfboot.h"
+#include "string.h"
+
+#include <stdint.h>
 
 #ifndef EXT_FLASH
     #error The stage1 loader only supports use with external flash
 #endif
 
-#ifndef LOADER_STAGE1_SIZE
-    #define LOADER_STAGE1_SIZE (4*1024)
-#endif
-#ifndef LOADER_STAGE1_START_ADDR
-    /* default is end of 4KB region (0x0FFC) */
-    #define LOADER_STAGE1_START_ADDR \
-        (WOLFBOOT_STAGE1_LOAD_ADDR + LOADER_STAGE1_SIZE - 0x4)
+#ifndef WOLFBOOT_STAGE1_SIZE
+    #define WOLFBOOT_STAGE1_SIZE (4*1024)
 #endif
 
+#ifndef WOLFBOOT_STAGE1_START_ADDR
+    /* default is end of 4KB region (0x0FFC) */
+    #define WOLFBOOT_STAGE1_START_ADDR \
+        (WOLFBOOT_STAGE1_LOAD_ADDR + WOLFBOOT_STAGE1_SIZE - 0x4)
+#endif
+
+#ifdef BUILD_LOADER_STAGE1
+
+#if defined(WOLFBOOT_ARCH) && WOLFBOOT_ARCH == PPC
+#include "hal/nxp_ppc.h"
+#endif
 
 int main(void)
 {
@@ -58,15 +67,31 @@ int main(void)
     uart_write("Loader Stage 1\r\n", 16);
 #endif
 
+#ifdef BOOT_ROM_ADDR
+    /* if this is executing from boot 4KB region (FCM buffer) it must
+     * first be relocated to RAM before the eLBC NAND can be read */
+    if (((uintptr_t)&hal_init & BOOT_ROM_ADDR) == BOOT_ROM_ADDR) {
+        wolfboot_start = (void*)WOLFBOOT_STAGE1_BASE_ADDR + BOOT_ROM_SIZE - 0x04;
+
+        /* relocate 4KB code to DST and jump */
+        memmove((void*)WOLFBOOT_STAGE1_BASE_ADDR, (void*)BOOT_ROM_ADDR,
+            BOOT_ROM_SIZE);
+
+        wolfboot_start(); /* never returns */
+    }
+#endif
+
     ret = ext_flash_read(
-        (uintptr_t)WOLFBOOT_ORIGIN,       /* flash offset */
-        (uint8_t*)WOLFBOOT_STAGE1_LOAD_ADDR, /* ram destination */
-        BOOTLOADER_PARTITION_SIZE         /* boot-loader partition (entire) */
+        (uintptr_t)WOLFBOOT_ORIGIN,         /* flash offset */
+        (uint8_t*)WOLFBOOT_STAGE1_LOAD_ADDR,/* ram destination */
+        BOOTLOADER_PARTITION_SIZE           /* boot-loader partition (entire) */
     );
     if (ret >= 0) {
-        wolfboot_start = (void*)LOADER_STAGE1_START_ADDR;
+        wolfboot_start = (void*)WOLFBOOT_STAGE1_START_ADDR;
         wolfboot_start(); /* never returns */
     }
 
     return 0;
 }
+
+#endif /* BUILD_LOADER_STAGE1 */
diff --git a/test-app/Makefile b/test-app/Makefile
index de4c4ec0..4cdc2f86 100644
--- a/test-app/Makefile
+++ b/test-app/Makefile
@@ -68,10 +68,6 @@ ifeq ($(ARCH),ARM_BE)
   CFLAGS+=-DSTACK_PAINTING=0
 endif
 
-ifeq ($(V),0)
-  Q=@
-endif
-
 ifeq ($(ENCRYPT),1)
     CFLAGS+=-D"EXT_ENCRYPTED=1"
 endif
@@ -264,26 +260,26 @@ standalone:CFLAGS+=-D"TEST_APP_STANDALONE"
 standalone:LDFLAGS:=-T standalone.ld -Wl,-gc-sections -Wl,-Map=image.map
 
 image.bin: image.elf
-	$(Q)echo "\t[BIN] $@"
+	@echo "\t[BIN] $@"
 	$(Q)$(OBJCOPY) --gap-fill $(FILL_BYTE) -O binary $^ $@
 
 image.elf: $(APP_OBJS) $(LSCRIPT)
-	$(Q)echo "\t[LD] $@"
+	@echo "\t[LD] $@"
 	$(Q)$(LD) $(LDFLAGS) $(APP_OBJS) $(OUTPUT_FLAG) $@
 
 
 standalone: image.bin
 
 ../src/libwolfboot.o: ../src/libwolfboot.c FORCE
-	$(Q)echo "\t[CC-$(ARCH)] $@"
+	@echo "\t[CC-$(ARCH)] $@"
 	$(Q)$(CC) $(CFLAGS) -c $(OUTPUT_FLAG) $@ ../src/libwolfboot.c
 
 %.o:%.c
-	$(Q)echo "\t[CC-$(ARCH)] $@"
+	@echo "\t[CC-$(ARCH)] $@"
 	$(Q)$(CC) $(CFLAGS) -c $(OUTPUT_FLAG) $@ $^
 
 %.o:%.S
-	$(Q)echo "\t[AS-$(ARCH)] $@"
+	@echo "\t[AS-$(ARCH)] $@"
 	$(Q)$(CC) $(CFLAGS) -c $(OUTPUT_FLAG) $@ $^
 
 clean:
diff --git a/tools/bin-assemble/Makefile b/tools/bin-assemble/Makefile
index 99645a10..2fda4157 100644
--- a/tools/bin-assemble/Makefile
+++ b/tools/bin-assemble/Makefile
@@ -8,11 +8,6 @@ EXE=bin-assemble
 
 LIBS=
 
-Q?=@
-ifeq ($(V),1)
-   Q=
-endif
-
 $(EXE): $(EXE).o
 	$(Q)$(CC) -o $@ $^ $(CFLAGS) $(LIBS)
 
diff --git a/tools/uart-flash-server/Makefile b/tools/uart-flash-server/Makefile
index 1d7ec762..fe0b5b1f 100644
--- a/tools/uart-flash-server/Makefile
+++ b/tools/uart-flash-server/Makefile
@@ -4,10 +4,6 @@
 
 CC=gcc
 CFLAGS+=-DBUILD_TOOL -Wall -g -ggdb -I../../include -I../../hal -Wextra
-Q?=@
-ifeq ($(V),1)
-   Q=
-endif
 
 EXE=ufserver