zipl: fix tape boot

The assembly code in tape0.S loads the kernel from tape to offset 0x4000.
After loading the kernel it loads the parm file and stores it to the
address provided by zipl. zipl doesn't know about the 0x4000 load offset,
so this address is inside of the kernel image and will silently corrupt
the loaded image in memory.

Fix this by copying the kernel to the final destination before loading
the parm files and initrd. We can do this because we can strip the first
0x10000 bytes like the zipl C code does for non-tape IPL.

Signed-off-by: Sven Schnelle <svens@linux.ibm.com>
Signed-off-by: Jan Höppner <hoeppner@linux.ibm.com>

zipl/boot/tape0.S

@@ -218,12 +218,25 @@ iplstart:
 	la    %r2,_initrd_addr			# store ramdisk address
 	mvc   INITRD_START-PARMAREA(8,%r12),0(%r2)
 
+# copy kernel to final destination
+
+	l	%r6,.Lkernsize
+	sl	%r6,.Lkernoff
+	l	%r2,.Lkernoff			# 0x10000
+	lr	%r3,%r6
+	l	%r4,.Loffset			# 0x4000
+	a	%r4,.Lkernoff
+	lr	%r5,%r6
+	mvcl	%r2,%r4
+	l	%r12,.Lparmarea
 
 #
 # Load parameter file
 #
 
 	l     %r4,_parm_addr+4
+	chi   %r4,-1
+	be    .Lnopf
 .Lagain1:
 	lr    %r2,%r4				# load parmfile
 	bas   %r14,.Lloader
@@ -284,6 +297,8 @@ iplstart:
 
 .Lagain2:
  	l     %r2,INITRD_START-PARMAREA+4(%r12)	# load ramdisk
+	chi   %r2,-1
+	be    .Lnoinitrd
 	bas   %r14,.Lloader
 
  	st    %r2,INITRD_SIZE-PARMAREA+4(%r12)	# store ramdisk size
@@ -298,86 +313,14 @@ iplstart:
 	clc   0(3,%r2),.L_eof
 	bz    .Lagain2
 
-#
-# Copy kernel image to final position
-#
-
-	la    %r2,_load_psw			# prepare copy code
-	la    %r3,.Lcopypsw
-	mvc   0(8,%r3),0(%r2)			# copy load psw
-
-	l     %r4,_initrd_addr+4
-	ahi   %r4,-0x100
-
-	la    %r2,.Lcopystart			# get copy code addr (from)
-	mvc   0(.Lcopyend-.Lcopystart,%r4),0(%r2)	# copy code
-
-	l     %r1,.Loffset			# temp kernel address (from)
-
-	l     %r2,4(%r3)			# get kernel addr from psw (to)
-	n     %r2,.Lpswmask
-	s     %r2,.Lkernoff
-
-	l     %r3,.Lkernsize			# (length)
-	br    %r4				# branch to copy routine
-
-#
-# Copy kernel, then load PSW
-#
-# R1 = from
-# R2 = to
-# R3 = length
-#
-
-	.align 8
-.Lcopystart:
-	clr   %r1,%r2
-	je    .Lcopyfinish			# kernel is already at dest
-	jh    .Lcopytolower			# need move to lower address
-
-# Copy last to first
-
-	ar    %r1,%r3				# start from last long
-	ar    %r2,%r3
-	lhi   %r4,-4				# set addr increment to -4
-	j     .Lcopygo
-
-# Copy first to last
-
-.Lcopytolower:
-	lhi   %r4,4				# set addr increment to 4
-
-.Lcopygo:
-	ahi   %r3,3				# get size in longs
-	srl   %r3,2
-
-.Lcopyloop:
-	l     %r0,0(%r1)			# copy longs
-	st    %r0,0(%r2)
-	ar    %r1,%r4
-	ar    %r2,%r4
-	brct  %r3,.Lcopyloop
-
-.Lcopyfinish:
-	bras  %r13,.Lcopyload			# set up base register
-.Lcopyload:
-	sr    %r0,%r0
-	la    %r1,.Lcopypsw-.Lcopyload(%r13)
-	mvc   0(8,%r0),0(%r1)			# copy load psw to 0
-
-	lm    %r0,%r15,.Lvmparam-.Lcopyload(%r13) # load saved registers
+.Lnoinitrd:
+	mvc   0(8,0),_load_psw			# copy load psw
 	lpsw  0					# start kernel
 
 	.align  8
-.Lcopypsw:
-	.quad 0
 .Lvmparam:
 	.long 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0	# vm parameter space
 
-.Lcopyend:
-
-.Lpswmask:
-	.long 0x7fffffff
 .Lkernsize:
 	.long 0
 .Lkernoff: