diff --git a/Makefile b/Makefile
index 28d24d3..4d77fe5 100644
--- a/Makefile
+++ b/Makefile
@@ -23,12 +23,13 @@ kernel.rom: kernel/*.asm kernel/*.inc lib/*.inc
 	nasm -s -o $@ -I lib -I kernel kernel/main.asm && scripts/fix-rom.sh $@
 
 boot/%.bin: boot/%.asm lib/*.inc
-	nasm -s -o $@ -I lib $<
+	nasm -s -o $@ -I boot -I lib $<
 
 fdimage.img: boot/floppy.bin kernel.rom
 	dd if=/dev/zero bs=512 count=$$(( $(FD_CYLINDERS) * $(FD_HEADS) * $(FD_SECTORS) )) of=$@
 	mformat -i $@ -t $(FD_CYLINDERS) -h $(FD_HEADS) -n $(FD_SECTORS) -B boot/floppy.bin ::
 	mcopy -i $@ kernel.rom ::kernel.rom
+	mattrib -i $@ +s ::kernel.rom
 
 clean:
 	rm -f *.com *.bin *.rom *.img *.log boot/*.bin
diff --git a/boot/chs.inc b/boot/chs.inc
new file mode 100644
index 0000000..53c5c4b
--- /dev/null
+++ b/boot/chs.inc
@@ -0,0 +1,47 @@
+fix_chs:
+  mov ah, 0x08
+  int 0x13
+  jc chs_end ; skip if function does not exist
+  inc dh
+  mov [cs:fdc.nos], dh
+  mov ax, cx
+  and ax, 0x003F
+  mov [cs:fdc.spt], ax ; no adjustment because sectors are 1-indexed
+  mov ax, cx
+  xchg al, ah
+  mov cl,6
+  shr ah,cl
+  inc ax ; convert from maximum number (0-based) to total number (1-based) of cylinders
+  mul word [cs:fdc.nos] ; number of tracks = number of cylinders * heads
+  mul word [cs:fdc.spt] ; number of sectors = number of tracks * sectors per track
+  mov [cs:fdc.ts], ax
+  jmp chs_end
+
+; Load a single sector into memory
+; Does not return on error
+; in    dx:ax   sector number (will be trashed)
+;       es:bx   buffer
+load_sector:
+  push cx
+  div word [cs:fdc.spt]    ; ax:temp     = (lba / spt)
+  inc dx                   ; dx:sector   = (lba % spt) + 1
+  mov cl, dl ; sector number
+  xor dx, dx
+  div word [cs:fdc.nos]    ; ax:cylinder = (tmp / heads)
+                           ; dx:head     = (tmp % heads)
+  mov ch, al ; cylinder number
+  mov dh, dl ; head number
+  mov dl, [cs:fdc.drv] ; driver number
+  mov ax, 0x0201 ; ah=0x02 al=0x01
+  int 0x13
+  jc disk_error
+  ; return
+  pop cx
+  ret
+
+disk_error:
+  xchg al, ah
+  call error
+  db "DISK ERROR", 0
+
+chs_end:
\ No newline at end of file
diff --git a/boot/floppy.asm b/boot/floppy.asm
index 0725eae..7a8ce27 100644
--- a/boot/floppy.asm
+++ b/boot/floppy.asm
@@ -1,11 +1,9 @@
 org 0x0000
 
-fdc:		; FDC Descriptor as per ECMA-107
-  jmp _startup
-  times (0x03 - ($-$$)) nop
-.oem:
-  db "2B"	; creating system identifier
-  times (0x0B - ($-$$)) db " " ; space-padded
+jmp 0x07C0:main
+
+times (0x0B - ($-$$)) db 0 ; space-padded
+fdc:
 .ss:
   dw 0x200	; sector size
 .sc:
@@ -43,62 +41,99 @@ fdc:		; FDC Descriptor as per ECMA-107
   db "FAT12"
   times (62 - ($-$$)) db " "
 
-_startup:
-  xor ax, ax
-  ; setup stack area growing down from 0x10000
-  mov ss, ax
-  mov sp, ax
-  ; adjust CS
-  jmp 0x07C0:main
-
-; non-fdc variables
-rootdir:
-.start:
-  dw 0
-.length:
-  dw 0
-
 main:
   cli
-
-  ; backup important values
+  ; setup stack area growing down from directly below us
+  xor ax, ax
+  mov ss, ax
+  mov sp, 0x7C00
+  ; setup base segments
+  mov ax, cs
+  mov ds, ax
+  mov es, ax
+  ; save initial environment
   push bx
   push cx
   push dx
 
-  ; setup buffer area directly at 4k
-  mov ax, 0x0100
-  mov ds, ax
-  mov es, ax
+  mov [fdc.drv], dl ; fix drive number in fdc
 
-  mov [cs:fdc.drv], dl ; backup drive number
+%include "chs.inc"
 
-  call fix_chs
-  call calc_sectors
-  call find_first_cluster
-  xor bx, bx
-  call load_cluster
+  mov di, 0x1000 ; loading address for roms
+  call search_rootdir
 
-  ; restore important variables
-  xor ax, ax
-  pop dx
-  pop cx
-  pop bx
-  mov di, [0x001A]
-  mov ax, [di+0x1A]
-  push ds
-  push ax
-  retf
+  mov ax, 0xFFFF
   call error
-  db "ROM ERROR", 0
+  db "THING", 0
+
+search_rootdir:
+  mov ax, [fdc.sf]
+  mul byte [fdc.fn]
+  add ax, [fdc.rsc]
+  mul word [fdc.ss]
+  mov si, ax ; now pointing to start of rootdir
+  mov cx, [fdc.rde]
+  jmp .start
+.next:
+  dec cx
+  jz .end
+  add si, 0x20
+.start:
+  call lazy_load
+  cmp byte [si], 0 ; end of directory
+  je .end
+  cmp byte [si], 0xe5 ; deleted entry
+  je .next
+  cmp word [si+8], 0x4F52 ; first 2 chars of extension: "RO"
+  jne .next
+  mov ax, [si+10] ; 3rd char of extension, file flags
+  and ax, 0x14FF ; zero out flags except directory and system
+  cmp ax, 0x044D ; "M" , system flag set, directory flag unset
+  jne .next
+  mov ax, [si+26]
+  push si
+  push cx
+  call loadrom
+  pop cx
+  pop si
+  jmp .next
+.end:
+  ret
+
+; in	ax	cluster number
+loadrom:
+  call print16
+  ret
+
+; Makes sure the block starting with (or under SI) is
+; already fetched from disk
+; in	ds:si	data expected to be fetched from disk
+; trashes ax, bx and dx
+lazy_load:
+  cmp si, [max_si]
+  jc .end
+  ; calculate next sector number
+  mov ax, [max_si]
+  xor dx, dx
+  div word [fdc.ss]
+  ; write to es:bx
+  xor dx, dx
+  push cs
+  pop es
+  mov bx, [max_si]
+  call load_sector
+  ; update max_si counter
+  mov ax, [max_si]
+  add ax, [fdc.ss]
+  mov [max_si], ax
+  ; loop until requirement is satisfied
+  jmp lazy_load
+.end:
+  ret
 
-; Write AX, a string and bail out
-; Return pointer is the string start
-; String must be 0 suffixed
 error:
   call print16
-  push cs
-  pop ds
   pop si
   mov bx, 0x0000
   mov al, 0x20
@@ -112,132 +147,13 @@ error:
   hlt
   jmp .hlt
 
-; Read CHS data from BIOS and fix fdc values
-fix_chs:
-  mov ah, 0x08
-  int 0x13
-  jc .end ; skip if function does not exist
-  inc dh
-  mov [cs:fdc.nos], dh
-  mov ax, cx
-  and ax, 0x003F
-  mov [cs:fdc.spt], ax ; no adjustment because sectors are 1-indexed
-  mov ax, cx
-  xchg al, ah
-  mov cl,6
-  shr ah,cl
-  inc ax ; convert from maximum number (0-based) to total number (1-based) of cylinders
-  mul word [cs:fdc.nos] ; number of tracks = number of cylinders * heads
-  mul word [cs:fdc.spt] ; number of sectors = number of tracks * sectors per track
-  mov [cs:fdc.ts], ax
-.end:
-  ret
-
-calc_sectors:
-  mov bx, [cs:fdc.ss]  ; bytes per sector
-  mov cl, 5
-  shr bx, cl           ; div by 2^5, 32 bytes per entry
-  mov ax, [cs:fdc.rde] ; number of entries
-  xor dx, dx
-  div bx
-  mov [cs:rootdir.length], ax
-  mov ax, [cs:fdc.sf]  ; sectors per fat
-  mul byte [cs:fdc.fn] ; number of fats
-  add ax, [cs:fdc.rsc] ; reserved sectors
-  mov [cs:rootdir.start], ax
-  ret
-
-; Load the root directory
-find_first_cluster:
-  mov ax, [cs:rootdir.start]
-  mov cx, [cs:rootdir.length]
-  xor bx, bx
-.blkloop:
-  call loadblk
-  add bx, [cs:fdc.ss]
-  add ax, 1
-  loop .blkloop
-  push cs
-  pop es
-  mov cx, [cs:fdc.rde]
-  xor bx, bx
-.dirloop:
-  push cx
-  mov cx, 0x000B
-  mov si, bx
-  mov di, bootfilename
-  repe cmpsb
-  pop cx
-  jz .found
-  add bx, 0x0020 ; 32 bytes per directory entry
-  loop .dirloop
-  call error
-  db "NOT FOUND", 0
-.found:
-  push ds
-  pop es
-  add bx, 26
-  mov ax, [bx]
-  ret
-
-; Load a single FAT cluster into memory
-; in    ax      cluster number
-;       es:bx   buffer
-;
-load_cluster:
-  xor ch, ch
-  mov cl, [cs:fdc.sc] ; sectors per cluster
-  xor dx, dx
-  dec ax ; somehow 2 is first cluster
-  div word cx
-  add ax, [cs:rootdir.start]
-  add ax, [cs:rootdir.length]
-.loop:
-  call loadblk
-  inc ax
-  add bx, [cs:fdc.ss]
-  loop .loop
-  ret
-
-; Load a single block into memory
-; Does not return on error
-; in    ax      sector number
-;       es:bx	buffer
-loadblk:
-  push ax
-  push cx
-  push dx
-  xor dx, dx
-  div word [cs:fdc.spt]    ; ax:temp     = (lba / spt)
-  inc dx                ; dx:sector   = (lba % spt) + 1
-  mov cl, dl ; sector number
-  xor dx, dx
-  div word [cs:fdc.nos]    ; ax:cylinder = (tmp / heads)
-                        ; dx:head     = (tmp % heads)
-  mov ch, al ; cylinder number
-  mov dh, dl ; head number
-  mov dl, [cs:fdc.drv] ; driver number
-  mov ax, 0x0201 ; ah=0x02 al=0x01
-  int 0x13
-  jc .error
-  pop dx
-  pop cx
-  pop ax
-  ret
-.error:
-  xchg al, ah
-  call error
-  db "DISK ERROR", 0
-
 %include "print.asm"
 
-times (0x1F0 - ($-$$)) db 0
+times (0x1FA - ($-$$)) db 0
 
-bootfilename:
-  db "KERNEL"
-  times (0x1F8 - ($-$$)) db " "
-  db "ROM"
+dw 0x2B2B
 
-times (0x1FE - ($-$$)) db 0
+max_si:
+dw 0x200
 
 dw 0xAA55