Fix LOG_X usage crashing the system

Each `LOG_X` call declared a rather huge variable on the stack,
and using it to dump a whole structure often led to a page fault.
Now the temporary variables are `static`, which is a bit wasteful
on the kernel binary size, but easier to maintain.

Additionally bootloader updated to load kernels bigger than 64K.

src/bootloader.asm

@@ -10,7 +10,6 @@ mov sp, 0x7000
 FIRST_PARTITION_SECTOR equ 2048 ; TODO Read MBR.
 STAGE_BUFFER_SEGMENT   equ 0x0900
 STAGE_BUFFER_OFFSET    equ 0x0000
-KERNEL_BUFFER_SEGMENT  equ 0x2000
 
 jmp prepare_kernel
 
@@ -39,9 +38,10 @@ root_start: dw 0
 root_size:  dw 0
 data_start: dw 0
 
-kernel_filename:      db 'KERNEL  BIN'
-kernel_cluster:       dw 0
-kernel_buffer_offset: dw 0
+kernel_filename:        db 'KERNEL  BIN'
+kernel_cluster:         dw 0
+kernel_buffer_segment:  dw 0x2000
+kernel_buffer_offset:   dw 0
 
 read_sectors:
     xor bx, bx
@@ -192,7 +192,7 @@ load_kernel_cluster:
     add ax, [data_start]
     xor cx, cx
     mov cl, [bpb.sectors_per_cluster]
-    mov bx, KERNEL_BUFFER_SEGMENT
+    mov bx, [kernel_buffer_segment]
     mov es, bx
     mov di, [kernel_buffer_offset]
     call read_sectors
@@ -201,6 +201,8 @@ load_kernel_cluster:
     shl ax, 9 ; * 512
     add ax, [kernel_buffer_offset]
     mov [kernel_buffer_offset], ax
+    jnc find_next_kernel_cluster
+    add word [kernel_buffer_segment], 0x1000
 
 find_next_kernel_cluster:
     xor bx, bx