AARCH64: Allow for embedding an initramfs

This actually breaks the armv7 port. It has to be revisited in the
future, since it not only is not compilable right now, but also
cannot boot linux properly. TODO

Also introduce a new C-based __memcpy_optimized that does not
cause an exception on some devices when copying the initramfs.
This issue has to be debugged further.

Signed-off-by: Ivaylo Ivanov <ivo.ivanov.ivanov1@gmail.com>

.gitignore

@@ -18,3 +18,4 @@ boot.img
 blob/Image
 blob/dtb
 drivers/*.a
+blob/ramdisk

Kconfig

@@ -36,6 +36,13 @@ menu "Building options"
 		default blob/dtb
 		help
 		  Path to the device tree, which is going to get embedded to uniLoader
+
+	config RAMDISK_PATH
+		string "Path to ramdisk image"
+		depends on EMBED_LINUX
+		default blob/ramdisk
+		help
+		  Path to the ramdisk, which is going to get embedded to uniLoader
 endmenu
 
 source "lib/Kconfig"

Makefile

@@ -103,6 +103,7 @@ export srctree objtree VPATH
 CROSS_COMPILE	?= $(CONFIG_CROSS_COMPILE:"%"=%)
 KERNEL_PATH	?= $(CONFIG_KERNEL_PATH:"%"=%)
 DT_PATH	?= $(CONFIG_DT_PATH:"%"=%)
+RAMDISK_PATH	?= $(CONFIG_RAMDISK_PATH:"%"=%)
 TEXT_BASE	?= $(CONFIG_TEXT_BASE:"%"=%)
 
 KCONFIG_CONFIG	?= .config
@@ -371,8 +372,8 @@ uniLoader-all	:= $(uniLoader-objs) $(uniLoader-libs)
 quiet_cmd_uniLoader = LD      $@.o
       cmd_uniLoader = $(LD) $(main-y) $(arch-libs-y) $(uniLoader-libs) -o $@.o --script=arch/$(ARCH)/linker.lds
 
-arch/$(ARCH)/linker.lds: arch/$(ARCH)/linker.lds.S $(KERNEL_PATH)
-	$(CPP) $< -DTEXT_BASE=$(TEXT_BASE) -DKERNEL_PATH=$(KERNEL_PATH) -DDTB_PATH=$(DT_PATH) -P -o $@
+arch/$(ARCH)/linker.lds: arch/$(ARCH)/linker.lds.S $(KERNEL_PATH) $(RAMDISK_PATH)
+	$(CPP) $< -DTEXT_BASE=$(TEXT_BASE) -DKERNEL_PATH=$(KERNEL_PATH) -DDTB_PATH=$(DT_PATH) -DRAMDISK_PATH=$(RAMDISK_PATH) -P -o $@
 
 uniLoader: $(uniLoader-all)
 	$(call if_changed,uniLoader)

arch/aarch64/linker.lds.S

@@ -9,6 +9,7 @@ TARGET(binary)
 
 INPUT(KERNEL_PATH)
 INPUT(DTB_PATH)
+INPUT(RAMDISK_PATH)
 
 SECTIONS
 {
@@ -41,5 +42,11 @@ SECTIONS
 		KERNEL_PATH
 	}
 
-	kernel_size = SIZEOF(.kernel);
+	.ramdisk ALIGN(4096) : {
+		ramdisk = .;
+		RAMDISK_PATH
+	}
+
+	kernel_size = SIZEOF(.kernel);
+	ramdisk_size = SIZEOF(.ramdisk);
 }

arch/aarch64/start.S

@@ -18,6 +18,8 @@ ENTRY(_start)
 	add     x0, x0, #:lo12:dtb  /* Add offset within the page */
 	adrp    x1, kernel          /* Load page of kernel */
 	add     x1, x1, #:lo12:kernel /* Add offset within the page */
+	adrp    x2, ramdisk          /* Load page of kernel */
+	add     x2, x2, #:lo12:ramdisk /* Add offset within the page */
 
 	bl main
 END(_start)

board/Kconfig

@@ -97,6 +97,10 @@ menu "Device Specific Addresses"
 		default 0x090000000 if SAMSUNG_J5LTE
 		default 0x090000000 if SAMSUNG_GTA4XL
 
+	config RAMDISK_ENTRY
+		hex "Ramdisk Entry Address"
+		default 0x082000000 if SAMSUNG_DREAMLTE
+
 	config FRAMEBUFFER_BASE
 		hex "Framebuffer Base Address (for SimpleFB)"
 		depends on SIMPLE_FB

include/main.h

@@ -9,6 +9,7 @@
 #define MAIN_H_
 
 extern unsigned long kernel_size;
+extern unsigned long ramdisk_size;
 extern void load_kernel(void* dtb, void* x1, void* x2, void* x3, void* kernel);
 extern void soc_init(void);
 

lib/unic/string.h

@@ -30,6 +30,49 @@ void writel (unsigned int value, void* address);
 // C-driven optimized functions
 void *memset (void *m, int c, size_t n);
 
+/* Nonzero if either X or Y is not aligned on a "long" boundary.  */
+#define UNALIGNED_CPY(X, Y) (((long)X & (sizeof (long) - 1)) | \
+			    ((long)Y & (sizeof (long) - 1)))
+
+/* How many bytes are copied each iteration of the 4X unrolled loop.  */
+#define BIGBLOCKSIZE (sizeof(long) << 2)
+
+static void *__optimized_memcpy (void *dst0, const void *src0, size_t len0)
+{
+	char *dst = dst0;
+	const char *src = src0;
+	long *aligned_dst;
+	const long *aligned_src;
+
+	/* If the size is small, or either SRC or DST is unaligned,
+	then punt into the byte copy loop.  This should be rare.  */
+	if (!(((len0) < BIGBLOCKSIZE)) && !UNALIGNED_CPY (src, dst)) {
+		aligned_dst = (long*)dst;
+		aligned_src = (long*)src;
+		/* Copy 4X long words at a time if possible.  */
+		while (len0 >= BIGBLOCKSIZE) {
+			*aligned_dst++ = *aligned_src++;
+			*aligned_dst++ = *aligned_src++;
+			*aligned_dst++ = *aligned_src++;
+			*aligned_dst++ = *aligned_src++;
+			len0 -= BIGBLOCKSIZE;
+		}
+		/* Copy one long word at a time if possible.  */
+		while (len0 >= LBLOCKSIZE) {
+			*aligned_dst++ = *aligned_src++;
+			len0 -= LBLOCKSIZE;
+		}
+
+	/* Pick up any residual with a byte copier.  */
+	dst = (char*)aligned_dst;
+	src = (char*)aligned_src;
+	}
+
+	while (len0--)
+		*dst++ = *src++;
+	return dst0;
+}
+
 // Assembly-driven functions
 #ifdef __aarch64__
 void *memcpy (void *__restrict, const void *__restrict, size_t);

main/main.c

@@ -23,7 +23,7 @@ struct board_data board = {
 	}
 };
 
-void main(void* dt, void* kernel)
+void main(void* dt, void* kernel, void* ramdisk)
 {
 	/* Initialize SoC and Board specific peripherals/quirks */
 	init_board_funcs(&board);
@@ -46,6 +46,7 @@ void main(void* dt, void* kernel)
 	printk(-1, "Booting linux...\n");
 
 	memcpy((void*)CONFIG_PAYLOAD_ENTRY, kernel, (unsigned long) &kernel_size);
+	__optimized_memcpy((void*)CONFIG_RAMDISK_ENTRY, ramdisk, (unsigned long) &ramdisk_size);
 	load_kernel(dt, 0, 0, 0, (void*)CONFIG_PAYLOAD_ENTRY);
 
 	/* We shouldn't get there */