part1 of arm64. This patch provide u-boot with arm64 support. Currently it works on Foundation Model for armv8 or Fast Model for armv8.

From: fenghua <fenghua@ligen-virtual-machine.(none)>

This patch provide u-boot with arm64 support. Currently, it works on Foundation Model for armv8 or Fast Model for armv8.

Signed-off-by fenghua <fenghua@ligen-virtual-machine.(none)>

From: fenghua <fenghua@ligen-virtual-machine.(none)>

This patch provide u-boot with arm64 support. Currently, it works on Foundation Model for armv8 or Fast Model for armv8.

Signed-off-by fenghua <fenghua@ligen-virtual-machine.(none)>

---
 arch/arm64/config.mk                 |   32 +++
 arch/arm64/lib/Makefile              |   64 +++++
 arch/arm64/lib/board.c               |  456 ++++++++++++++++++++++++++++++++++
 arch/arm64/lib/bootm.c               |  211 ++++++++++++++++
 arch/arm64/lib/cache.c               |  282 +++++++++++++++++++++
 arch/arm64/lib/crt0.S                |  129 ++++++++++
 arch/arm64/lib/interrupts.c          |  109 ++++++++
 arch/arm64/lib/relocate.S            |   72 ++++++
 arch/arm64/lib/reset.c               |   37 +++
 arch/arm64/lib/timer.c               |   95 +++++++
 board/armltd/dts/vexpress64.dts      |  215 ++++++++++++++++
 board/armltd/vexpress64/Makefile     |   43 ++++
 board/armltd/vexpress64/vexpress64.c |   63 +++++
 boards.cfg                           |    1 +
 common/cmd_bdinfo.c                  |   32 +++
 common/fdt_support.c                 |   66 ++---
 common/image.c                       |    5 +-
 doc/README.arm64                     |   10 +
 examples/standalone/stubs.c          |   13 +
 include/configs/vexpress_aemv8a.h    |  200 +++++++++++++++
 include/image.h                      |    1 +
 lib/asm-offsets.c                    |    2 +-
 22 files changed, 2103 insertions(+), 35 deletions(-)
 create mode 100644 arch/arm64/config.mk
 create mode 100644 arch/arm64/lib/Makefile
 create mode 100644 arch/arm64/lib/board.c
 create mode 100644 arch/arm64/lib/bootm.c
 create mode 100644 arch/arm64/lib/cache.c
 create mode 100644 arch/arm64/lib/crt0.S
 create mode 100644 arch/arm64/lib/interrupts.c

 create mode 100644 arch/arm64/lib/relocate.S
 create mode 100644 arch/arm64/lib/reset.c
 create mode 100644 arch/arm64/lib/timer.c
 create mode 100644 board/armltd/dts/vexpress64.dts
 create mode 100644 board/armltd/vexpress64/Makefile
 create mode 100644 board/armltd/vexpress64/vexpress64.c
 create mode 100644 doc/README.arm64
 create mode 100644 include/configs/vexpress_aemv8a.h

arch/arm64/config.mk new file +13 -0

@@ -0,0 +1,32 @@
+#
+# Copyright (c) 2013	FengHua <[email protected]>
+#
+# See file CREDITS for list of people who contributed to this
+# project.
+#
+# This program is free software; you can redistribute it and/or
+# modify it under the terms of the GNU General Public License as
+# published by the Free Software Foundation; either version 2 of
+# the License, or (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of

common/fdt_support.c +34 -25

index b034c98..9bc5821 100644
@@ -21,6 +21,34 @@
  */
 DECLARE_GLOBAL_DATA_PTR;
 
+/*
+ * Get cells len in bytes
+ *     if #NNNN-cells property is 2 then len is 8
+ *     otherwise len is 4
+ */
+static int get_cells_len(void *blob, char *nr_cells_name)
+{
+	const fdt32_t *cell;
+
+	cell = fdt_getprop(blob, 0, nr_cells_name, NULL);
+	if (cell && fdt32_to_cpu(*cell) == 2)
+		return 8;
+
+	return 4;
+}
+
+/*
+ * Write a 4 or 8 byte big endian cell
+ */
+static void write_cell(u8 *addr, u64 val, int size)
+{
+	int shift = (size - 1) * 8;
+	while (size-- > 0) {
+		*addr++ = (val >> shift) & 0xff;
+		shift -= 8;
+	}
+}
+
 /**
  * fdt_getprop_u32_default - Find a node and return it's property or a default
  *

@@ -131,9 +159,9 @@ static int fdt_fixup_stdout(void *fdt, int chosenoff)
 
 int fdt_initrd(void *fdt, ulong initrd_start, ulong initrd_end, int force)
 {
-	int   nodeoffset;
+	int   nodeoffset, addr_cell_len;
 	int   err, j, total;
-	fdt32_t  tmp;
+	fdt64_t  tmp;
 	const char *path;
 	uint64_t addr, size;
 

@@ -170,9 +198,11 @@ int fdt_initrd(void *fdt, ulong initrd_start, ulong initrd_end, int force)
 		return err;
 	}
 
+	addr_cell_len = get_cells_len(fdt, "#address-cells");
+
 	path = fdt_getprop(fdt, nodeoffset, "linux,initrd-start", NULL);
 	if ((path == NULL) || force) {
-		tmp = cpu_to_fdt32(initrd_start);
+		write_cell((u8 *)&tmp, initrd_start, addr_cell_len);
 		err = fdt_setprop(fdt, nodeoffset,
 			"linux,initrd-start", &tmp, sizeof(tmp));
 		if (err < 0) {

@@ -181,7 +211,7 @@ int fdt_initrd(void *fdt, ulong initrd_start, ulong initrd_end, int force)
 				fdt_strerror(err));
 			return err;
 		}
-		tmp = cpu_to_fdt32(initrd_end);
+		write_cell((u8 *)&tmp, initrd_end, addr_cell_len);
 		err = fdt_setprop(fdt, nodeoffset,
 			"linux,initrd-end", &tmp, sizeof(tmp));
 		if (err < 0) {

@@ -343,34 +373,6 @@ void do_fixup_by_compat_u32(void *fdt, const char *compat,
 	do_fixup_by_compat(fdt, compat, prop, &tmp, 4, create);
 }
 
-/*
- * Get cells len in bytes
- *     if #NNNN-cells property is 2 then len is 8
- *     otherwise len is 4
- */
-static int get_cells_len(void *blob, char *nr_cells_name)
-{
-	const fdt32_t *cell;
-
-	cell = fdt_getprop(blob, 0, nr_cells_name, NULL);
-	if (cell && fdt32_to_cpu(*cell) == 2)
-		return 8;
-
-	return 4;
-}
-
-/*
- * Write a 4 or 8 byte big endian cell
- */
-static void write_cell(u8 *addr, u64 val, int size)
-{

common/fdt_support.c +1 -1

@@ -422,7 +423,7 @@ ulong getenv_bootm_low(void)
 
 #if defined(CONFIG_SYS_SDRAM_BASE)
 	return CONFIG_SYS_SDRAM_BASE;
-#elif defined(CONFIG_ARM)
+#elif defined(CONFIG_ARM) || defined(CONFIG_ARM64)
 	return gd->bd->bi_dram[0].start;
 #else
 	return 0;

doc/README.arm64 new file +3 -0

+3. Currently, fdt should be in the first 512MB of RAM, so, fdt_high should be handled specially.
+   I define fdt_high as 0xa0000000 when CONFIG_SYS_SDRAM_BASE is 0x80000000.
+

examples/standalone/stubs.c +3 -0

index 8fb1765..4e39b1b 100644
@@ -195,6 +195,19 @@ gd_t *global_data;
 "	l.jr	r13\n"		\
 "	l.nop\n"				\
 	: : "i"(offsetof(gd_t, jt)), "i"(XF_ ## x * sizeof(void *)) : "r13");
+#elif defined(CONFIG_ARM64)
+/*
+ * x18 holds the pointer to the global_data, ip is a call-clobbered

examples/standalone/stubs.c +0 -0

 #else
 /*"	addi	$sp, $sp, -24\n"	\
 "	br	$r16\n"			\*/

include/configs/vexpress_aemv8a.h new file +29 -0

new file mode 100644
index 0000000..3c96987
@@ -0,0 +1,200 @@
+/*
+ * Configuration for Versatile Express. Parts were derived from other ARM
+ *   configurations.
+ *
+ * See file CREDITS for list of people who contributed to this
+ * project.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of
+ * the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.	 See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston,
+ * MA 02111-1307 USA
+ */
+
+#ifndef __VEXPRESS_AEMV8A_H
+#define __VEXPRESS_AEMV8A_H
+
+#define DEBUG
+
+#define CONFIG_SYS_NO_FLASH

-----????-----
???: "Scott Wood" <[email protected]>
????: 2013?8?14? ???
???: fenghua at phytium.com.cn
??: u-boot at lists.denx.de, trini at ti.com
??: Re: [U-Boot] [PATCH] part1 of arm64. This patch provide u-boot with arm64 support. Currently it works on Foundation Model for armv8 or Fast Model for armv8.

On Mon, 2013-08-12 at 00:05 +0800, fenghua at phytium.com.cn wrote:

Quoted 6 lines

From: fenghua <fenghua@ligen-virtual-machine.(none)>

This patch provide u-boot with arm64 support. Currently, it works on Foundation Model for armv8 or Fast Model for armv8.

Signed-off-by fenghua <fenghua@ligen-virtual-machine.(none)>

Could you provide a proper signoff (full name and valid e-mail address)?

Quoted 32 lines

---
 arch/arm64/config.mk                 |   32 +++
 arch/arm64/lib/Makefile              |   64 +++++
 arch/arm64/lib/board.c               |  456 ++++++++++++++++++++++++++++++++++
 arch/arm64/lib/bootm.c               |  211 ++++++++++++++++
 arch/arm64/lib/cache.c               |  282 +++++++++++++++++++++
 arch/arm64/lib/crt0.S                |  129 ++++++++++
 arch/arm64/lib/interrupts.c          |  109 ++++++++
 arch/arm64/lib/relocate.S            |   72 ++++++
 arch/arm64/lib/reset.c               |   37 +++
 arch/arm64/lib/timer.c               |   95 +++++++
 board/armltd/dts/vexpress64.dts      |  215 ++++++++++++++++
 board/armltd/vexpress64/Makefile     |   43 ++++
 board/armltd/vexpress64/vexpress64.c |   63 +++++
 boards.cfg                           |    1 +
 common/cmd_bdinfo.c                  |   32 +++
 common/fdt_support.c                 |   66 ++---
 common/image.c                       |    5 +-
 doc/README.arm64                     |   10 +
 examples/standalone/stubs.c          |   13 +
 include/configs/vexpress_aemv8a.h    |  200 +++++++++++++++
 include/image.h                      |    1 +
 lib/asm-offsets.c                    |    2 +-
 22 files changed, 2103 insertions(+), 35 deletions(-)
 create mode 100644 arch/arm64/config.mk
 create mode 100644 arch/arm64/lib/Makefile
 create mode 100644 arch/arm64/lib/board.c
 create mode 100644 arch/arm64/lib/bootm.c
 create mode 100644 arch/arm64/lib/cache.c
 create mode 100644 arch/arm64/lib/crt0.S
 create mode 100644 arch/arm64/lib/interrupts.c

Quoted 8 lines

 create mode 100644 arch/arm64/lib/relocate.S
 create mode 100644 arch/arm64/lib/reset.c
 create mode 100644 arch/arm64/lib/timer.c
 create mode 100644 board/armltd/dts/vexpress64.dts
 create mode 100644 board/armltd/vexpress64/Makefile
 create mode 100644 board/armltd/vexpress64/vexpress64.c
 create mode 100644 doc/README.arm64
 create mode 100644 include/configs/vexpress_aemv8a.h

It would be nice if the patches were separated into arch support, core support, board support, anything specific to the foundation/fast models, etc.

arch/arm64/config.mk

Quoted 7 lines

arch/arm64/config.mk new file +0 -0

new file mode 100644
index 0000000..e40e983

Does there really need to be a separate arch/arm64? I know this is what Linux does, but on every other 32/64 architecture that decision has been eventually reversed (even on x86, which has about as much difference between 32 and 64 bit as ARM does).

We started working on armv8 U-Boot using a unified arch/arm, and at least got to the point of something that builds, so it doesn't seem impractical. Besides making maintenance easier, it would also make it much easier to review what is being done differently for arm64 (and for any files that do need to be moved or copied, be sure to pass -M -C to git format-patch).

Quoted 18 lines

arch/arm64/config.mk new file +17 -0

@@ -0,0 +1,32 @@
+#
+# Copyright (c) 2013	FengHua <[email protected]>
+#
+# See file CREDITS for list of people who contributed to this
+# project.
+#
+# This program is free software; you can redistribute it and/or
+# modify it under the terms of the GNU General Public License as
+# published by the Free Software Foundation; either version 2 of
+# the License, or (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License

arch/arm64/config.mk line 23

Quoted 6 lines

arch/arm64/config.mk new file +6 -0

+# along with this program; if not, write to the Free Software
+# Foundation, Inc., 59 Temple Place, Suite 330, Boston,
+# MA 02111-1307 USA
+#
+
+CROSS_COMPILE ?= /home/fenghua/DS-5-Workspace/gcc-linaro-aarch64-linux-gnu-4.8-2013.05_linux/bin/aarch64-linux-gnu-

Please don't insert references to paths that only work on your computer.

Plus, what about native builds that don't need a cross compiler?

Quoted 103 lines

common/fdt_support.c +34 -29

index b034c98..9bc5821 100644
@@ -21,6 +21,34 @@
  */
 DECLARE_GLOBAL_DATA_PTR;
 
+/*
+ * Get cells len in bytes
+ *     if #NNNN-cells property is 2 then len is 8
+ *     otherwise len is 4
+ */
+static int get_cells_len(void *blob, char *nr_cells_name)
+{
+	const fdt32_t *cell;
+
+	cell = fdt_getprop(blob, 0, nr_cells_name, NULL);
+	if (cell && fdt32_to_cpu(*cell) == 2)
+		return 8;
+
+	return 4;
+}
+
+/*
+ * Write a 4 or 8 byte big endian cell
+ */
+static void write_cell(u8 *addr, u64 val, int size)
+{
+	int shift = (size - 1) * 8;
+	while (size-- > 0) {
+		*addr++ = (val >> shift) & 0xff;
+		shift -= 8;
+	}
+}
+
 /**
  * fdt_getprop_u32_default - Find a node and return it's property or a default
  *

@@ -131,9 +159,9 @@ static int fdt_fixup_stdout(void *fdt, int chosenoff)
 
 int fdt_initrd(void *fdt, ulong initrd_start, ulong initrd_end, int force)
 {
-	int   nodeoffset;
+	int   nodeoffset, addr_cell_len;
 	int   err, j, total;
-	fdt32_t  tmp;
+	fdt64_t  tmp;
 	const char *path;
 	uint64_t addr, size;
 

@@ -170,9 +198,11 @@ int fdt_initrd(void *fdt, ulong initrd_start, ulong initrd_end, int force)
 		return err;
 	}
 
+	addr_cell_len = get_cells_len(fdt, "#address-cells");
+
 	path = fdt_getprop(fdt, nodeoffset, "linux,initrd-start", NULL);
 	if ((path == NULL) || force) {
-		tmp = cpu_to_fdt32(initrd_start);
+		write_cell((u8 *)&tmp, initrd_start, addr_cell_len);
 		err = fdt_setprop(fdt, nodeoffset,
 			"linux,initrd-start", &tmp, sizeof(tmp));
 		if (err < 0) {

@@ -181,7 +211,7 @@ int fdt_initrd(void *fdt, ulong initrd_start, ulong initrd_end, int force)
 				fdt_strerror(err));
 			return err;
 		}
-		tmp = cpu_to_fdt32(initrd_end);
+		write_cell((u8 *)&tmp, initrd_end, addr_cell_len);
 		err = fdt_setprop(fdt, nodeoffset,
 			"linux,initrd-end", &tmp, sizeof(tmp));
 		if (err < 0) {

@@ -343,34 +373,6 @@ void do_fixup_by_compat_u32(void *fdt, const char *compat,
 	do_fixup_by_compat(fdt, compat, prop, &tmp, 4, create);
 }
 
-/*
- * Get cells len in bytes
- *     if #NNNN-cells property is 2 then len is 8
- *     otherwise len is 4
- */
-static int get_cells_len(void *blob, char *nr_cells_name)
-{
-	const fdt32_t *cell;
-
-	cell = fdt_getprop(blob, 0, nr_cells_name, NULL);
-	if (cell && fdt32_to_cpu(*cell) == 2)
-		return 8;
-
-	return 4;
-}
-
-/*
- * Write a 4 or 8 byte big endian cell
- */
-static void write_cell(u8 *addr, u64 val, int size)
-{
-	int shift = (size - 1) * 8;
-	while (size-- > 0) {
-		*addr++ = (val >> shift) & 0xff;
-		shift -= 8;

common/fdt_support.c void do_fixup_by_compat_u32(void *fdt, const char *compat, · line 378

Quoted 6 lines

common/fdt_support.c +0 -3

-	}
-}
-
 #ifdef CONFIG_NR_DRAM_BANKS
 #define MEMORY_BANKS_MAX CONFIG_NR_DRAM_BANKS
 #else

What's going on here?

Quoted 10 lines

common/fdt_support.c +1 -1

@@ -422,7 +423,7 @@ ulong getenv_bootm_low(void)
 
 #if defined(CONFIG_SYS_SDRAM_BASE)
 	return CONFIG_SYS_SDRAM_BASE;
-#elif defined(CONFIG_ARM)
+#elif defined(CONFIG_ARM) || defined(CONFIG_ARM64)
 	return gd->bd->bi_dram[0].start;
 #else
 	return 0;

common/fdt_support.c phys_size_t getenv_bootm_size(void) · line 451

Quoted 8 lines

common/fdt_support.c +1 -1

@@ -444,7 +445,7 @@ phys_size_t getenv_bootm_size(void)
 		tmp = 0;
 
 
-#if defined(CONFIG_ARM)
+#if defined(CONFIG_ARM) || defined(CONFIG_ARM64)
 	return gd->bd->bi_dram[0].size - tmp;
 #else
 	return gd->bd->bi_memsize - tmp;

Ignoring the issue of why CONFIG_ARM is not defined here, why is ARM special?

doc/README.arm64 line 5

Quoted 11 lines

doc/README.arm64 new file +5 -0

new file mode 100644
index 0000000..8fef26d
@@ -0,0 +1,10 @@
+Notes:
+
+1. Currenly, u-boot running at EL2.
+
+2. Currently, gcc-aarch64 produce error when compiling with pie and rel_dyn.

arm64 uses rela, not rel. This requires a small change to the relocation code and the linker script.

doc/README.arm64 line 6

Quoted 1 line

doc/README.arm64 new file +1 -0

+   So, GOT is used to relocate u-boot and CONFIG_NEEDS_MANUAL_RELOC is needed.

:-(

Quoted 18 lines

doc/README.arm64 new file +3 -0

+3. Currently, fdt should be in the first 512MB of RAM, so, fdt_high should be handled specially.
+   I define fdt_high as 0xa0000000 when CONFIG_SYS_SDRAM_BASE is 0x80000000.
+

examples/standalone/stubs.c +7 -0

index 8fb1765..4e39b1b 100644
@@ -195,6 +195,19 @@ gd_t *global_data;
 "	l.jr	r13\n"		\
 "	l.nop\n"				\
 	: : "i"(offsetof(gd_t, jt)), "i"(XF_ ## x * sizeof(void *)) : "r13");
+#elif defined(CONFIG_ARM64)
+/*
+ * x18 holds the pointer to the global_data, ip is a call-clobbered
+ * register
+ */
+#define EXPORT_FUNC(x) \
+	asm volatile (			\

examples/standalone/stubs.c gd_t *global_data; · line 210

Quoted 6 lines

examples/standalone/stubs.c +6 -0

+"	.globl " #x "\n"		\
+#x ":\n"				\
+"	ldr	x19, [x18, %0]\n"		\
+"	ldr	x19, [x19, %1]\n"		\
+"	br	x19\n"		\
+	: : "i"(offsetof(gd_t, jt)), "i"(XF_ ## x * sizeof(void *)) : "x19");

Why use a nonvolate register here?

Quoted 43 lines

examples/standalone/stubs.c +0 -0

 #else
 /*"	addi	$sp, $sp, -24\n"	\
 "	br	$r16\n"			\*/

include/configs/vexpress_aemv8a.h new file +33 -0

new file mode 100644
index 0000000..3c96987
@@ -0,0 +1,200 @@
+/*
+ * Configuration for Versatile Express. Parts were derived from other ARM
+ *   configurations.
+ *
+ * See file CREDITS for list of people who contributed to this
+ * project.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of
+ * the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.	 See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston,
+ * MA 02111-1307 USA
+ */
+
+#ifndef __VEXPRESS_AEMV8A_H
+#define __VEXPRESS_AEMV8A_H
+
+#define DEBUG
+
+#define CONFIG_SYS_NO_FLASH
+
+#define CONFIG_SUPPORT_RAW_INITRD
+
+#define CONFIG_SKIP_LOWLEVEL_INIT

include/configs/vexpress_aemv8a.h line 39

Quoted 6 lines

include/configs/vexpress_aemv8a.h new file +6 -0

+
+/* Cache Definitions */
+//#define CONFIG_SYS_DCACHE_OFF
+//#define CONFIG_SYS_ICACHE_OFF
+
+#define CONFIG_IDENT_STRING		" aarch64"

Shouldn't CONFIG_IDENT_STRING be the target name (i.e. the board name)?

-Scott

Quoted 21 lines

-----????-----
???: "Scott Wood" <[email protected]>
????: 2013?8?14? ???
???: fenghua at phytium.com.cn
??: u-boot at lists.denx.de, trini at ti.com
??: Re: [U-Boot] [PATCH] part1 of arm64. This patch provide u-boot with arm64 support. Currently it works on Foundation Model for armv8 or Fast Model for armv8.

arch/arm64/config.mk

Quoted 5 lines · fenghua at phytium.com.cn

arch/arm64/config.mk new file +0 -0

new file mode 100644
index 0000000..e40e983

Does there really need to be a separate arch/arm64? I know this is what Linux does, but on every other 32/64 architecture that decision has been eventually reversed (even on x86, which has about as much difference between 32 and 64 bit as ARM does).

Quoted 7 lines

We started working on armv8 U-Boot using a unified arch/arm, and at least got to the point of something that builds, so it doesn't seem impractical. Besides making maintenance easier, it would also make it much easier to review what is being done differently for arm64 (and for any files that do need to be moved or copied, be sure to pass -M -C to git format-patch).

The porting is performed as a seperate architecture due to a few considerations,
1. The porting will be simple and clear.

2. There's no any real chip of armv8 until now. Many aspect of armv8 are not completely confirmed.

Of course, it could be merged with ARM at a proper time in the later. Actually, linux kernel follow this mode.

Quoted 18 lines

Quoted 18 lines

arch/arm64/config.mk new file +17 -0

@@ -0,0 +1,32 @@
+#
+# Copyright (c) 2013	FengHua <[email protected]>
+#
+# See file CREDITS for list of people who contributed to this
+# project.
+#
+# This program is free software; you can redistribute it and/or
+# modify it under the terms of the GNU General Public License as
+# published by the Free Software Foundation; either version 2 of
+# the License, or (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License

Quoted 9 lines

arch/arm64/config.mk line 23

Quoted 6 lines

arch/arm64/config.mk new file +6 -0

+# along with this program; if not, write to the Free Software
+# Foundation, Inc., 59 Temple Place, Suite 330, Boston,
+# MA 02111-1307 USA
+#
+
+CROSS_COMPILE ?= /home/fenghua/DS-5-Workspace/gcc-linaro-aarch64-linux-gnu-4.8-2013.05_linux/bin/aarch64-linux-gnu-

Please don't insert references to paths that only work on your computer.

I will fix it.

Quoted 1 line

Plus, what about native builds that don't need a cross compiler?

Currently, there's no real chip of armv8. so,cross compiler is needed.

Quoted 102 lines

Quoted 102 lines

common/fdt_support.c +34 -29

index b034c98..9bc5821 100644
@@ -21,6 +21,34 @@
  */
 DECLARE_GLOBAL_DATA_PTR;
 
+/*
+ * Get cells len in bytes
+ *     if #NNNN-cells property is 2 then len is 8
+ *     otherwise len is 4
+ */
+static int get_cells_len(void *blob, char *nr_cells_name)
+{
+	const fdt32_t *cell;
+
+	cell = fdt_getprop(blob, 0, nr_cells_name, NULL);
+	if (cell && fdt32_to_cpu(*cell) == 2)
+		return 8;
+
+	return 4;
+}
+
+/*
+ * Write a 4 or 8 byte big endian cell
+ */
+static void write_cell(u8 *addr, u64 val, int size)
+{
+	int shift = (size - 1) * 8;
+	while (size-- > 0) {
+		*addr++ = (val >> shift) & 0xff;
+		shift -= 8;
+	}
+}
+
 /**
  * fdt_getprop_u32_default - Find a node and return it's property or a default
  *

@@ -131,9 +159,9 @@ static int fdt_fixup_stdout(void *fdt, int chosenoff)
 
 int fdt_initrd(void *fdt, ulong initrd_start, ulong initrd_end, int force)
 {
-	int   nodeoffset;
+	int   nodeoffset, addr_cell_len;
 	int   err, j, total;
-	fdt32_t  tmp;
+	fdt64_t  tmp;
 	const char *path;
 	uint64_t addr, size;
 

@@ -170,9 +198,11 @@ int fdt_initrd(void *fdt, ulong initrd_start, ulong initrd_end, int force)
 		return err;
 	}
 
+	addr_cell_len = get_cells_len(fdt, "#address-cells");
+
 	path = fdt_getprop(fdt, nodeoffset, "linux,initrd-start", NULL);
 	if ((path == NULL) || force) {
-		tmp = cpu_to_fdt32(initrd_start);
+		write_cell((u8 *)&tmp, initrd_start, addr_cell_len);
 		err = fdt_setprop(fdt, nodeoffset,
 			"linux,initrd-start", &tmp, sizeof(tmp));
 		if (err < 0) {

@@ -181,7 +211,7 @@ int fdt_initrd(void *fdt, ulong initrd_start, ulong initrd_end, int force)
 				fdt_strerror(err));
 			return err;
 		}
-		tmp = cpu_to_fdt32(initrd_end);
+		write_cell((u8 *)&tmp, initrd_end, addr_cell_len);
 		err = fdt_setprop(fdt, nodeoffset,
 			"linux,initrd-end", &tmp, sizeof(tmp));
 		if (err < 0) {

@@ -343,34 +373,6 @@ void do_fixup_by_compat_u32(void *fdt, const char *compat,
 	do_fixup_by_compat(fdt, compat, prop, &tmp, 4, create);
 }
 
-/*
- * Get cells len in bytes
- *     if #NNNN-cells property is 2 then len is 8
- *     otherwise len is 4
- */
-static int get_cells_len(void *blob, char *nr_cells_name)
-{
-	const fdt32_t *cell;
-
-	cell = fdt_getprop(blob, 0, nr_cells_name, NULL);
-	if (cell && fdt32_to_cpu(*cell) == 2)
-		return 8;
-
-	return 4;
-}
-
-/*
- * Write a 4 or 8 byte big endian cell
- */
-static void write_cell(u8 *addr, u64 val, int size)
-{
-	int shift = (size - 1) * 8;
-	while (size-- > 0) {
-		*addr++ = (val >> shift) & 0xff;
-		shift -= 8;

Quoted 9 lines

common/fdt_support.c void do_fixup_by_compat_u32(void *fdt, const char *compat, · line 378

Quoted 6 lines

common/fdt_support.c +0 -3

-	}
-}
-
 #ifdef CONFIG_NR_DRAM_BANKS
 #define MEMORY_BANKS_MAX CONFIG_NR_DRAM_BANKS
 #else

What's going on here?

Whitout this the initrd represented by "linux,initrd-start" could only be placed under 4G address space. This part fix it to 64 bit address space.

Quoted 6 lines

Quoted 6 lines

doc/README.arm64 new file +0 -0

new file mode 100644
index 0000000..8fef26d
@@ -0,0 +1,10 @@

Quoted 9 lines

doc/README.arm64 line 5

Quoted 5 lines

doc/README.arm64 new file +5 -0

+Notes:
+
+1. Currenly, u-boot running at EL2.
+
+2. Currently, gcc-aarch64 produce error when compiling with pie and rel_dyn.

arm64 uses rela, not rel. This requires a small change to the relocation code and the linker script.

I also try to use rela but it not works. maybe something is wrong with the code.