add initial bootloader work

5 files changed, 186 insertions, 1 deletions

diff --git a/Makefile b/Makefile
index b1b15c4..da43892 100644
--- a/Makefile
+++ b/Makefile
@@ -17,13 +17,37 @@ kernel.elf : boot.o kernel.o uart.o
 kernel7.img : kernel.elf
 	arm-none-eabi-objcopy $^ -O binary $@
 
+loader_stage2.o : loader_stage2.c
+	arm-none-eabi-gcc $(CFLAGS) -c $^ -o $@
+
+loader_stage2.elf : loader_stage2.o uart.o
+	arm-none-eabi-gcc -T loader_stage2.ld -o $@ -ffreestanding -O2 -nostdlib $^ -lgcc
+
+loader_stage2.img : loader_stage2.elf
+	arm-none-eabi-objcopy $^ -O binary $@
+
+loader_stage2_embeddable.o : loader_stage2.img
+	arm-none-eabi-objcopy -I binary -O elf32-littlearm -B arm --rename-section .data=.rodata $^ $@
+
+loader_stage1.o : loader_stage1.c
+	arm-none-eabi-gcc $(CFLAGS) -c $^ -o $@
+
+loader.elf : boot.o loader_stage1.o uart.o loader_stage2_embeddable.o
+	arm-none-eabi-gcc -T loader_stage1.ld -o $@ -ffreestanding -O2 -nostdlib $^ -lgcc
+
+loader.img : loader.elf
+	arm-none-eabi-objcopy $^ -O binary $@
+
 qemu-elf : kernel.elf
 	qemu-system-arm -m 256 -M raspi2 -serial stdio -kernel $^
 
 qemu-bin : kernel7.img
 	qemu-system-arm -m 256 -M raspi2 -serial stdio -kernel $^
 
+qemu-loader : loader.img
+	qemu-system-arm -m 256 -M raspi2 -serial stdio -kernel $^
+
 clean :
-	-rm kernel7.img kernel.elf boot.o kernel.o uart.o
+	-rm *.img *.elf *.o
 
 .PHONY: all qemu-elf qemu-bin clean
diff --git a/loader_stage1.c b/loader_stage1.c
new file mode 100644
index 0000000..9492747
--- /dev/null
+++ b/loader_stage1.c
@@ -0,0 +1,30 @@
+#include <stddef.h>
+#include <stdint.h>
+#include <uart.h>
+#include <global.h>
+
+char *const stage2_addr = ((void*) 0x4000);
+
+// there's one tricky thing about embedding file in executable;
+// mainly, symbols are visible to c code as extern chars, but the actual
+// values are their adresses... see the code below
+extern char
+  _binary_loader_stage2_img_start,
+  _binary_loader_stage2_img_end,
+  _binary_loader_stage2_img_size;
+
+void kernel_main(uint32_t r0, uint32_t r1, uint32_t atags)
+{
+  // uart_init() call to be moved to stage2...
+  uart_init();
+  uart_puts("Hello, bootloader World!\r\n");
+
+  // stage2 of the bootloader is a blob embedded in executable;
+  // copy it over to it's destination place
+  // TODO implement a memcpy() somewhere and use it instead of loops
+  for (size_t i = 0; i < (size_t) &_binary_loader_stage2_img_size; i++)
+    stage2_addr[i] = (&_binary_loader_stage2_img_start)[i];
+
+  // jump to stage2
+  ((void(*)(uint32_t, uint32_t, uint32_t))stage2_addr)(r0, r1, atags);
+}
diff --git a/loader_stage1.ld b/loader_stage1.ld
new file mode 100644
index 0000000..ce11095
--- /dev/null
+++ b/loader_stage1.ld
@@ -0,0 +1,47 @@
+ENTRY(_start)
+ 
+SECTIONS
+{
+    /* Starts at LOADER_ADDR. */
+    /* Warning! Internet says RPis in 32-bit mode load binary at 0x8000! */
+    /* My experiments do, however, show, that qemu emulating RPi2 */
+    /* loads it at 0x10000! (took some pain to find out) */
+    . = 0x10000;
+    /* For AArch64, use . = 0x80000; Unless this too is wrong */
+    __start = .;
+    __text_start = .;
+    .text :
+    {
+        KEEP(*(.text.boot))
+        *(.text)
+    }
+    . = ALIGN(4096); /* align to page size */
+    __text_end = .;
+ 
+    __rodata_start = .;
+    .rodata :
+    {
+        *(.rodata)
+    }
+    . = ALIGN(4096); /* align to page size */
+    __rodata_end = .;
+ 
+    __data_start = .;
+    .data :
+    {
+        *(.data)
+    }
+    . = ALIGN(4096); /* align to page size */
+    __data_end = .;
+ 
+    __bss_start = .;
+    .bss :
+    {
+        bss = .;
+        *(.bss)
+    }
+    . = ALIGN(4096); /* align to page size */
+    __bss_end = .;
+    __bss_size = __bss_end - __bss_start;
+    __end = .;
+}
diff --git a/loader_stage2.c b/loader_stage2.c
new file mode 100644
index 0000000..98ce571
--- /dev/null
+++ b/loader_stage2.c
@@ -0,0 +1,40 @@
+#include <stddef.h>
+#include <stdint.h>
+#include <uart.h>
+#include <global.h>
+
+void *const kernel_load_addr = ((void*) 0x8000);
+
+void __attribute__((section(".text.stage2main")))
+stage2(uint32_t r0, uint32_t r1, uint32_t atags)
+{
+  // Declare as unused
+  (void) r0;
+  (void) r1;
+  (void) atags;
+
+  uart_puts("hello stage2!\n\r");
+
+  // get kernel size via uart (little endian)
+  uint32_t b0, b1, b2, b3;
+  
+  b0 = uart_getc();
+  b1 = uart_getc();
+  b2 = uart_getc();
+  b3 = uart_getc();
+  
+  uint32_t kernel_size = b0 | (b1 << 8) | (b2 << 16) | (b3 << 24);
+
+  // load kernel at kernel_load_addr
+  char *dst = kernel_load_addr, *end = dst + kernel_size;
+
+  while (dst < end)
+    *(dst++) = uart_getc();
+
+  // jump to kernel
+  // TODO also forward arguments (r0, r1, atags)
+  asm volatile("bx %0" :: "r" (kernel_load_addr) : "memory");
+}
+
+void *const _start = ((void*) stage2); // for linker script
+
diff --git a/loader_stage2.ld b/loader_stage2.ld
new file mode 100644
index 0000000..8f215e9
--- /dev/null
+++ b/loader_stage2.ld
@@ -0,0 +1,44 @@
+ENTRY(_start)
+ 
+SECTIONS
+{
+    /* stage2 bootloader gets loaded at 0x4000 */
+    . = 0x4000;
+    __start = .;
+    __text_start = .;
+    .text :
+    {
+        /* have entry point at the beginning */
+        KEEP(*(.text.stage2main))
+        *(.text)
+    }
+    . = ALIGN(4096); /* align to page size */
+    __text_end = .;
+ 
+    __rodata_start = .;
+    .rodata :
+    {
+        *(.rodata)
+    }
+    . = ALIGN(4096); /* align to page size */
+    __rodata_end = .;
+ 
+    __data_start = .;
+    .data :
+    {
+        *(.data)
+    }
+    . = ALIGN(4096); /* align to page size */
+    __data_end = .;
+ 
+    __bss_start = .;
+    .bss :
+    {
+        bss = .;
+        *(.bss)
+    }
+    . = ALIGN(4096); /* align to page size */
+    __bss_end = .;
+    __bss_size = __bss_end - __bss_start;
+    __end = .;
+}