[PATCH] s390: improved machine check handling

Improved machine check handling.  Kernel is now able to receive machine checks
while in kernel mode (system call, interrupt and program check handling).
Also register validation is now performed.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

5 files changed, 215 insertions, 56 deletions

diff --git a/arch/s390/kernel/entry.S b/arch/s390/kernel/entry.S
index c0e09b33febe..5b262b5d869f 100644
--- a/arch/s390/kernel/entry.S
+++ b/arch/s390/kernel/entry.S
@@ -7,6 +7,7 @@
  *    Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com),
  *               Hartmut Penner (hp@de.ibm.com),
  *               Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com),
+ *               Heiko Carstens <heiko.carstens@de.ibm.com>
  */
 
 #include <linux/sys.h>
@@ -49,9 +50,9 @@ SP_ILC       =  STACK_FRAME_OVERHEAD + __PT_ILC
 SP_TRAP      =  STACK_FRAME_OVERHEAD + __PT_TRAP
 SP_SIZE      =  STACK_FRAME_OVERHEAD + __PT_SIZE
 
-_TIF_WORK_SVC = (_TIF_SIGPENDING | _TIF_NEED_RESCHED | \
+_TIF_WORK_SVC = (_TIF_SIGPENDING | _TIF_NEED_RESCHED | _TIF_MCCK_PENDING | \
                  _TIF_RESTART_SVC | _TIF_SINGLE_STEP )
-_TIF_WORK_INT = (_TIF_SIGPENDING | _TIF_NEED_RESCHED)
+_TIF_WORK_INT = (_TIF_SIGPENDING | _TIF_NEED_RESCHED | _TIF_MCCK_PENDING)
 
 STACK_SHIFT = PAGE_SHIFT + THREAD_ORDER
 STACK_SIZE  = 1 << STACK_SHIFT
@@ -121,7 +122,11 @@ STACK_SIZE  = 1 << STACK_SHIFT
         bz      BASED(stack_overflow)
 3:
 #endif
-2:      s       %r15,BASED(.Lc_spsize)  # make room for registers & psw
+2:
+        .endm
+
+        .macro  CREATE_STACK_FRAME psworg,savearea
+        s       %r15,BASED(.Lc_spsize)  # make room for registers & psw
         mvc     SP_PSW(8,%r15),0(%r12)  # move user PSW to stack
         la      %r12,\psworg
         st      %r2,SP_ORIG_R2(%r15)    # store original content of gpr 2
@@ -161,6 +166,13 @@ __switch_to_base:
         be      __switch_to_noper-__switch_to_base(%r1) # we got away w/o bashing TLB's
         lctl    %c9,%c11,__THREAD_per(%r3)      # Nope we didn't
 __switch_to_noper:
+        l       %r4,__THREAD_info(%r2)          # get thread_info of prev
+        tm      __TI_flags+3(%r4),_TIF_MCCK_PENDING # machine check pending?
+        bz      __switch_to_no_mcck-__switch_to_base(%r1)
+        ni      __TI_flags+3(%r4),255-_TIF_MCCK_PENDING # clear flag in prev
+        l       %r4,__THREAD_info(%r3)          # get thread_info of next
+        oi      __TI_flags+3(%r4),_TIF_MCCK_PENDING # set it in next
+__switch_to_no_mcck:
         stm     %r6,%r15,__SF_GPRS(%r15)# store __switch_to registers of prev task
         st      %r15,__THREAD_ksp(%r2)  # store kernel stack to prev->tss.ksp
         l       %r15,__THREAD_ksp(%r3)  # load kernel stack from next->tss.ksp
@@ -185,6 +197,7 @@ system_call:
 sysc_saveall:
         SAVE_ALL_BASE __LC_SAVE_AREA
         SAVE_ALL __LC_SVC_OLD_PSW,__LC_SAVE_AREA,1
+        CREATE_STACK_FRAME __LC_SVC_OLD_PSW,__LC_SAVE_AREA
         lh      %r7,0x8a          # get svc number from lowcore
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
 sysc_vtime:
@@ -234,6 +247,8 @@ sysc_work_loop:
 # One of the work bits is on. Find out which one.
 #
 sysc_work:
+        tm      __TI_flags+3(%r9),_TIF_MCCK_PENDING
+        bo      BASED(sysc_mcck_pending)
         tm      __TI_flags+3(%r9),_TIF_NEED_RESCHED
         bo      BASED(sysc_reschedule)
         tm      __TI_flags+3(%r9),_TIF_SIGPENDING
@@ -253,6 +268,14 @@ sysc_reschedule:
         br      %r1                    # call scheduler
 
 #
+# _TIF_MCCK_PENDING is set, call handler
+#
+sysc_mcck_pending:
+        l       %r1,BASED(.Ls390_handle_mcck)
+        la      %r14,BASED(sysc_work_loop)
+        br      %r1                     # TIF bit will be cleared by handler
+
+#
 # _TIF_SIGPENDING is set, call do_signal
 #
 sysc_sigpending:     
@@ -430,6 +453,7 @@ pgm_check_handler:
         tm      __LC_PGM_INT_CODE+1,0x80 # check whether we got a per exception
         bnz     BASED(pgm_per)           # got per exception -> special case
         SAVE_ALL __LC_PGM_OLD_PSW,__LC_SAVE_AREA,1
+        CREATE_STACK_FRAME __LC_PGM_OLD_PSW,__LC_SAVE_AREA
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
         tm      SP_PSW+1(%r15),0x01     # interrupting from user ?
         bz      BASED(pgm_no_vtime)
@@ -468,6 +492,7 @@ pgm_per:
 #
 pgm_per_std:
         SAVE_ALL __LC_PGM_OLD_PSW,__LC_SAVE_AREA,1
+        CREATE_STACK_FRAME __LC_PGM_OLD_PSW,__LC_SAVE_AREA
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
         tm      SP_PSW+1(%r15),0x01     # interrupting from user ?
         bz      BASED(pgm_no_vtime2)
@@ -493,6 +518,7 @@ pgm_no_vtime2:
 #
 pgm_svcper:
         SAVE_ALL __LC_SVC_OLD_PSW,__LC_SAVE_AREA,1
+        CREATE_STACK_FRAME __LC_SVC_OLD_PSW,__LC_SAVE_AREA
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
         tm      SP_PSW+1(%r15),0x01     # interrupting from user ?
         bz      BASED(pgm_no_vtime3)
@@ -521,6 +547,7 @@ io_int_handler:
         stck    __LC_INT_CLOCK
         SAVE_ALL_BASE __LC_SAVE_AREA+16
         SAVE_ALL __LC_IO_OLD_PSW,__LC_SAVE_AREA+16,0
+        CREATE_STACK_FRAME __LC_IO_OLD_PSW,__LC_SAVE_AREA+16
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
         tm      SP_PSW+1(%r15),0x01     # interrupting from user ?
         bz      BASED(io_no_vtime)
@@ -578,9 +605,11 @@ io_work:
         lr      %r15,%r1
 #
 # One of the work bits is on. Find out which one.
-# Checked are: _TIF_SIGPENDING and _TIF_NEED_RESCHED
+# Checked are: _TIF_SIGPENDING, _TIF_NEED_RESCHED and _TIF_MCCK_PENDING
 #
 io_work_loop:
+        tm      __TI_flags+3(%r9),_TIF_MCCK_PENDING
+        bo      BASED(io_mcck_pending)
         tm      __TI_flags+3(%r9),_TIF_NEED_RESCHED
         bo      BASED(io_reschedule)
         tm      __TI_flags+3(%r9),_TIF_SIGPENDING
@@ -588,6 +617,14 @@ io_work_loop:
         b       BASED(io_leave)
 
 #
+# _TIF_MCCK_PENDING is set, call handler
+#
+io_mcck_pending:
+        l       %r1,BASED(.Ls390_handle_mcck)
+        l       %r14,BASED(io_work_loop)
+        br      %r1                    # TIF bit will be cleared by handler
+
+#
 # _TIF_NEED_RESCHED is set, call schedule
 #       
 io_reschedule:        
@@ -621,6 +658,7 @@ ext_int_handler:
         stck    __LC_INT_CLOCK
         SAVE_ALL_BASE __LC_SAVE_AREA+16
         SAVE_ALL __LC_EXT_OLD_PSW,__LC_SAVE_AREA+16,0
+        CREATE_STACK_FRAME __LC_EXT_OLD_PSW,__LC_SAVE_AREA+16
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
         tm      SP_PSW+1(%r15),0x01     # interrupting from user ?
         bz      BASED(ext_no_vtime)
@@ -642,19 +680,62 @@ ext_no_vtime:
 
         .globl mcck_int_handler
 mcck_int_handler:
-        STORE_TIMER __LC_ASYNC_ENTER_TIMER
+        spt     __LC_CPU_TIMER_SAVE_AREA        # revalidate cpu timer
+        lm      %r0,%r15,__LC_GPREGS_SAVE_AREA  # revalidate gprs
         SAVE_ALL_BASE __LC_SAVE_AREA+32
-        SAVE_ALL __LC_MCK_OLD_PSW,__LC_SAVE_AREA+32,0
+        la      %r12,__LC_MCK_OLD_PSW
+        tm      __LC_MCCK_CODE,0x80     # system damage?
+        bo      BASED(mcck_int_main)    # yes -> rest of mcck code invalid
+        tm      __LC_MCCK_CODE+5,0x02   # stored cpu timer value valid?
+        bo      BASED(0f)
+        spt     __LC_LAST_UPDATE_TIMER  # revalidate cpu timer
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
-        tm      SP_PSW+1(%r15),0x01     # interrupting from user ?
+        mvc     __LC_LAST_UPDATE_TIMER(8),__LC_ASYNC_ENTER_TIMER
+        mvc     __LC_LAST_UPDATE_TIMER(8),__LC_SYNC_ENTER_TIMER
+        mvc     __LC_LAST_UPDATE_TIMER(8),__LC_EXIT_TIMER
+0:      tm      __LC_MCCK_CODE+2,0x08   # mwp of old psw valid?
+        bno     BASED(mcck_no_vtime)    # no -> skip cleanup critical
+        tm      __LC_MCK_OLD_PSW+1,0x01 # interrupting from user ?
         bz      BASED(mcck_no_vtime)
         UPDATE_VTIME __LC_EXIT_TIMER,__LC_ASYNC_ENTER_TIMER,__LC_USER_TIMER
         UPDATE_VTIME __LC_LAST_UPDATE_TIMER,__LC_EXIT_TIMER,__LC_SYSTEM_TIMER
         mvc     __LC_LAST_UPDATE_TIMER(8),__LC_ASYNC_ENTER_TIMER
 mcck_no_vtime:
 #endif
+0:
+        tm      __LC_MCCK_CODE+2,0x09   # mwp + ia of old psw valid?
+        bno     BASED(mcck_int_main)    # no -> skip cleanup critical
+        tm      __LC_MCK_OLD_PSW+1,0x01 # test problem state bit
+        bnz     BASED(mcck_int_main)    # from user -> load async stack
+        clc     __LC_MCK_OLD_PSW+4(4),BASED(.Lcritical_end)
+        bhe     BASED(mcck_int_main)
+        clc     __LC_MCK_OLD_PSW+4(4),BASED(.Lcritical_start)
+        bl      BASED(mcck_int_main)
+        l       %r14,BASED(.Lcleanup_critical)
+        basr    %r14,%r14
+mcck_int_main:
+        l       %r14,__LC_PANIC_STACK   # are we already on the panic stack?
+        slr     %r14,%r15
+        sra     %r14,PAGE_SHIFT
+        be      BASED(0f)
+        l       %r15,__LC_PANIC_STACK   # load panic stack
+0:      CREATE_STACK_FRAME __LC_MCK_OLD_PSW,__LC_SAVE_AREA+32
+        l       %r9,__LC_THREAD_INFO    # load pointer to thread_info struct
+        la      %r2,SP_PTREGS(%r15)     # load pt_regs
         l       %r1,BASED(.Ls390_mcck)
-        basr    %r14,%r1          # call machine check handler
+        basr    %r14,%r1                # call machine check handler
+        tm      SP_PSW+1(%r15),0x01     # returning to user ?
+        bno     BASED(mcck_return)
+        l       %r1,__LC_KERNEL_STACK   # switch to kernel stack
+        s       %r1,BASED(.Lc_spsize)
+        mvc     SP_PTREGS(__PT_SIZE,%r1),SP_PTREGS(%r15)
+        xc      __SF_BACKCHAIN(4,%r1),__SF_BACKCHAIN(%r1) # clear back chain
+        lr      %r15,%r1
+        stosm   __SF_EMPTY(%r15),0x04   # turn dat on
+        tm      __TI_flags+3(%r9),_TIF_MCCK_PENDING
+        bno     BASED(mcck_return)
+        l       %r1,BASED(.Ls390_handle_mcck)
+        basr    %r14,%r1                # call machine check handler
 mcck_return:
         RESTORE_ALL 0
 
@@ -742,7 +823,7 @@ cleanup_critical:
         clc     4(4,%r12),BASED(cleanup_table_sysc_work_loop)
         bl      BASED(0f)
         clc     4(4,%r12),BASED(cleanup_table_sysc_work_loop+4)
-        bl      BASED(cleanup_sysc_leave)
+        bl      BASED(cleanup_sysc_return)
 0:
         br      %r14
 
@@ -760,6 +841,7 @@ cleanup_system_call:
         mvc     __LC_SAVE_AREA(16),__LC_SAVE_AREA+16
 0:      st      %r13,__LC_SAVE_AREA+20
         SAVE_ALL __LC_SVC_OLD_PSW,__LC_SAVE_AREA,1
+        CREATE_STACK_FRAME __LC_SVC_OLD_PSW,__LC_SAVE_AREA
         st      %r15,__LC_SAVE_AREA+28
         lh      %r7,0x8a
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
@@ -834,6 +916,8 @@ cleanup_sysc_leave_insn:
  * Symbol constants
  */
 .Ls390_mcck:   .long  s390_do_machine_check
+.Ls390_handle_mcck:
+               .long  s390_handle_mcck
 .Ldo_IRQ:      .long  do_IRQ
 .Ldo_extint:   .long  do_extint
 .Ldo_signal:   .long  do_signal
diff --git a/arch/s390/kernel/entry64.S b/arch/s390/kernel/entry64.S
index 51527ab8c8f9..57ca75d0ad7f 100644
--- a/arch/s390/kernel/entry64.S
+++ b/arch/s390/kernel/entry64.S
@@ -7,6 +7,7 @@
  *    Author(s): Martin Schwidefsky (schwidefsky@de.ibm.com),
  *               Hartmut Penner (hp@de.ibm.com),
  *               Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com),
+ *               Heiko Carstens <heiko.carstens@de.ibm.com>
  */
 
 #include <linux/sys.h>
@@ -52,9 +53,9 @@ SP_SIZE      =  STACK_FRAME_OVERHEAD + __PT_SIZE
 STACK_SHIFT = PAGE_SHIFT + THREAD_ORDER
 STACK_SIZE  = 1 << STACK_SHIFT
 
-_TIF_WORK_SVC = (_TIF_SIGPENDING | _TIF_NEED_RESCHED | \
+_TIF_WORK_SVC = (_TIF_SIGPENDING | _TIF_NEED_RESCHED | _TIF_MCCK_PENDING | \
                  _TIF_RESTART_SVC | _TIF_SINGLE_STEP )
-_TIF_WORK_INT = (_TIF_SIGPENDING | _TIF_NEED_RESCHED)
+_TIF_WORK_INT = (_TIF_SIGPENDING | _TIF_NEED_RESCHED | _TIF_MCCK_PENDING)
 
 #define BASED(name) name-system_call(%r13)
 
@@ -114,7 +115,11 @@ _TIF_WORK_INT = (_TIF_SIGPENDING | _TIF_NEED_RESCHED)
         jz      stack_overflow
 3:
 #endif
-2:      aghi    %r15,-SP_SIZE           # make room for registers & psw
+2:
+        .endm
+
+        .macro  CREATE_STACK_FRAME psworg,savearea
+        aghi    %r15,-SP_SIZE           # make room for registers & psw
         mvc     SP_PSW(16,%r15),0(%r12) # move user PSW to stack
         la      %r12,\psworg
         stg     %r2,SP_ORIG_R2(%r15)    # store original content of gpr 2
@@ -152,6 +157,13 @@ __switch_to:
         je      __switch_to_noper            # we got away without bashing TLB's
         lctlg   %c9,%c11,__THREAD_per(%r3)      # Nope we didn't
 __switch_to_noper:
+        lg      %r4,__THREAD_info(%r2)              # get thread_info of prev
+        tm      __TI_flags+7(%r4),_TIF_MCCK_PENDING # machine check pending?
+        jz      __switch_to_no_mcck
+        ni      __TI_flags+7(%r4),255-_TIF_MCCK_PENDING # clear flag in prev
+        lg      %r4,__THREAD_info(%r3)              # get thread_info of next
+        oi      __TI_flags+7(%r4),_TIF_MCCK_PENDING # set it in next
+__switch_to_no_mcck:
         stmg    %r6,%r15,__SF_GPRS(%r15)# store __switch_to registers of prev task
         stg     %r15,__THREAD_ksp(%r2)  # store kernel stack to prev->tss.ksp
         lg      %r15,__THREAD_ksp(%r3)  # load kernel stack from next->tss.ksp
@@ -176,6 +188,7 @@ system_call:
 sysc_saveall:
         SAVE_ALL_BASE __LC_SAVE_AREA
         SAVE_ALL __LC_SVC_OLD_PSW,__LC_SAVE_AREA,1
+        CREATE_STACK_FRAME __LC_SVC_OLD_PSW,__LC_SAVE_AREA
         llgh    %r7,__LC_SVC_INT_CODE # get svc number from lowcore
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
 sysc_vtime:
@@ -232,6 +245,8 @@ sysc_work_loop:
 # One of the work bits is on. Find out which one.
 #
 sysc_work:
+        tm      __TI_flags+7(%r9),_TIF_MCCK_PENDING
+        jo      sysc_mcck_pending
         tm      __TI_flags+7(%r9),_TIF_NEED_RESCHED
         jo      sysc_reschedule
         tm      __TI_flags+7(%r9),_TIF_SIGPENDING
@@ -250,6 +265,13 @@ sysc_reschedule:
         jg      schedule            # return point is sysc_return
 
 #
+# _TIF_MCCK_PENDING is set, call handler
+#
+sysc_mcck_pending:
+        larl    %r14,sysc_work_loop
+        jg      s390_handle_mcck    # TIF bit will be cleared by handler
+
+#
 # _TIF_SIGPENDING is set, call do_signal
 #
 sysc_sigpending:     
@@ -474,6 +496,7 @@ pgm_check_handler:
         tm      __LC_PGM_INT_CODE+1,0x80 # check whether we got a per exception
         jnz     pgm_per                  # got per exception -> special case
         SAVE_ALL __LC_PGM_OLD_PSW,__LC_SAVE_AREA,1
+        CREATE_STACK_FRAME __LC_PGM_OLD_PSW,__LC_SAVE_AREA
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
         tm      SP_PSW+1(%r15),0x01     # interrupting from user ?
         jz      pgm_no_vtime
@@ -512,6 +535,7 @@ pgm_per:
 #
 pgm_per_std:
         SAVE_ALL __LC_PGM_OLD_PSW,__LC_SAVE_AREA,1
+        CREATE_STACK_FRAME __LC_PGM_OLD_PSW,__LC_SAVE_AREA
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
         tm      SP_PSW+1(%r15),0x01     # interrupting from user ?
         jz      pgm_no_vtime2
@@ -537,6 +561,7 @@ pgm_no_vtime2:
 #
 pgm_svcper:
         SAVE_ALL __LC_SVC_OLD_PSW,__LC_SAVE_AREA,1
+        CREATE_STACK_FRAME __LC_SVC_OLD_PSW,__LC_SAVE_AREA
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
         tm      SP_PSW+1(%r15),0x01     # interrupting from user ?
         jz      pgm_no_vtime3
@@ -564,6 +589,7 @@ io_int_handler:
         stck    __LC_INT_CLOCK
         SAVE_ALL_BASE __LC_SAVE_AREA+32
         SAVE_ALL __LC_IO_OLD_PSW,__LC_SAVE_AREA+32,0
+        CREATE_STACK_FRAME __LC_IO_OLD_PSW,__LC_SAVE_AREA+32
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
         tm      SP_PSW+1(%r15),0x01     # interrupting from user ?
         jz      io_no_vtime
@@ -621,9 +647,11 @@ io_work:
         lgr     %r15,%r1
 #
 # One of the work bits is on. Find out which one.
-# Checked are: _TIF_SIGPENDING and _TIF_NEED_RESCHED
+# Checked are: _TIF_SIGPENDING, _TIF_NEED_RESCHED and _TIF_MCCK_PENDING
 #
 io_work_loop:
+        tm      __TI_flags+7(%r9),_TIF_MCCK_PENDING
+        jo      io_mcck_pending
         tm      __TI_flags+7(%r9),_TIF_NEED_RESCHED
         jo      io_reschedule
         tm      __TI_flags+7(%r9),_TIF_SIGPENDING
@@ -631,6 +659,13 @@ io_work_loop:
         j       io_leave
 
 #
+# _TIF_MCCK_PENDING is set, call handler
+#
+io_mcck_pending:
+        larl    %r14,io_work_loop
+        jg      s390_handle_mcck        # TIF bit will be cleared by handler
+
+#
 # _TIF_NEED_RESCHED is set, call schedule
 #       
 io_reschedule:        
@@ -661,6 +696,7 @@ ext_int_handler:
         stck    __LC_INT_CLOCK
         SAVE_ALL_BASE __LC_SAVE_AREA+32
         SAVE_ALL __LC_EXT_OLD_PSW,__LC_SAVE_AREA+32,0
+        CREATE_STACK_FRAME __LC_EXT_OLD_PSW,__LC_SAVE_AREA+32
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
         tm      SP_PSW+1(%r15),0x01     # interrupting from user ?
         jz      ext_no_vtime
@@ -680,18 +716,60 @@ ext_no_vtime:
  */
         .globl mcck_int_handler
 mcck_int_handler:
-        STORE_TIMER __LC_ASYNC_ENTER_TIMER
+        la      %r1,4095                # revalidate r1
+        spt     __LC_CPU_TIMER_SAVE_AREA-4095(%r1)      # revalidate cpu timer
+        lmg     %r0,%r15,__LC_GPREGS_SAVE_AREA-4095(%r1)# revalidate gprs
         SAVE_ALL_BASE __LC_SAVE_AREA+64
-        SAVE_ALL __LC_MCK_OLD_PSW,__LC_SAVE_AREA+64,0
+        la      %r12,__LC_MCK_OLD_PSW
+        tm      __LC_MCCK_CODE,0x80     # system damage?
+        jo      mcck_int_main           # yes -> rest of mcck code invalid
+        tm      __LC_MCCK_CODE+5,0x02   # stored cpu timer value valid?
+        jo      0f
+        spt     __LC_LAST_UPDATE_TIMER
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
-        tm      SP_PSW+1(%r15),0x01     # interrupting from user ?
+        mvc     __LC_LAST_UPDATE_TIMER(8),__LC_ASYNC_ENTER_TIMER
+        mvc     __LC_LAST_UPDATE_TIMER(8),__LC_SYNC_ENTER_TIMER
+        mvc     __LC_LAST_UPDATE_TIMER(8),__LC_EXIT_TIMER
+0:      tm      __LC_MCCK_CODE+2,0x08   # mwp of old psw valid?
+        jno     mcck_no_vtime           # no -> no timer update
+        tm      __LC_MCK_OLD_PSW+1,0x01 # interrupting from user ?
         jz      mcck_no_vtime
         UPDATE_VTIME __LC_EXIT_TIMER,__LC_ASYNC_ENTER_TIMER,__LC_USER_TIMER
         UPDATE_VTIME __LC_LAST_UPDATE_TIMER,__LC_EXIT_TIMER,__LC_SYSTEM_TIMER
         mvc     __LC_LAST_UPDATE_TIMER(8),__LC_ASYNC_ENTER_TIMER
 mcck_no_vtime:
 #endif
-        brasl   %r14,s390_do_machine_check
+0:
+        tm      __LC_MCCK_CODE+2,0x09   # mwp + ia of old psw valid?
+        jno     mcck_int_main           # no -> skip cleanup critical
+        tm      __LC_MCK_OLD_PSW+1,0x01 # test problem state bit
+        jnz     mcck_int_main           # from user -> load kernel stack
+        clc     __LC_MCK_OLD_PSW+8(8),BASED(.Lcritical_end)
+        jhe     mcck_int_main
+        clc     __LC_MCK_OLD_PSW+8(8),BASED(.Lcritical_start)
+        jl      mcck_int_main
+        brasl   %r14,cleanup_critical
+mcck_int_main:
+        lg      %r14,__LC_PANIC_STACK   # are we already on the panic stack?
+        slgr    %r14,%r15
+        srag    %r14,%r14,PAGE_SHIFT
+        jz      0f
+        lg      %r15,__LC_PANIC_STACK   # load panic stack
+0:      CREATE_STACK_FRAME __LC_MCK_OLD_PSW,__LC_SAVE_AREA+64
+        lg      %r9,__LC_THREAD_INFO    # load pointer to thread_info struct
+        la      %r2,SP_PTREGS(%r15)     # load pt_regs
+        brasl   %r14,s390_do_machine_check
+        tm      SP_PSW+1(%r15),0x01     # returning to user ?
+        jno     mcck_return
+        lg      %r1,__LC_KERNEL_STACK   # switch to kernel stack
+        aghi    %r1,-SP_SIZE
+        mvc     SP_PTREGS(__PT_SIZE,%r1),SP_PTREGS(%r15)
+        xc      __SF_BACKCHAIN(8,%r1),__SF_BACKCHAIN(%r1) # clear back chain
+        lgr     %r15,%r1
+        stosm   __SF_EMPTY(%r15),0x04   # turn dat on
+        tm      __TI_flags+7(%r9),_TIF_MCCK_PENDING
+        jno     mcck_return
+        brasl   %r14,s390_handle_mcck
 mcck_return:
         RESTORE_ALL 0
 
@@ -775,7 +853,7 @@ cleanup_critical:
         clc     8(8,%r12),BASED(cleanup_table_sysc_work_loop)
         jl      0f
         clc     8(8,%r12),BASED(cleanup_table_sysc_work_loop+8)
-        jl      cleanup_sysc_leave
+        jl      cleanup_sysc_return
 0:
         br      %r14
 
@@ -793,6 +871,7 @@ cleanup_system_call:
         mvc     __LC_SAVE_AREA(32),__LC_SAVE_AREA+32
 0:      stg     %r13,__LC_SAVE_AREA+40
         SAVE_ALL __LC_SVC_OLD_PSW,__LC_SAVE_AREA,1
+        CREATE_STACK_FRAME __LC_SVC_OLD_PSW,__LC_SAVE_AREA
         stg     %r15,__LC_SAVE_AREA+56
         llgh    %r7,__LC_SVC_INT_CODE
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
diff --git a/arch/s390/kernel/process.c b/arch/s390/kernel/process.c
index 7aea25d6e300..9f3dff6c0b72 100644
--- a/arch/s390/kernel/process.c
+++ b/arch/s390/kernel/process.c
@@ -91,13 +91,12 @@ void do_monitor_call(struct pt_regs *regs, long interruption_code)
                             (void *)(long) smp_processor_id());
 }
 
+extern void s390_handle_mcck(void);
 /*
  * The idle loop on a S390...
  */
 void default_idle(void)
 {
-        psw_t wait_psw;
-        unsigned long reg;
         int cpu, rc;
 
         local_irq_disable();
@@ -125,38 +124,17 @@ void default_idle(void)
                 cpu_die();
 #endif
 
-        /* 
-         * Wait for external, I/O or machine check interrupt and
-         * switch off machine check bit after the wait has ended.
-         */
-        wait_psw.mask = PSW_KERNEL_BITS | PSW_MASK_MCHECK | PSW_MASK_WAIT |
-                PSW_MASK_IO | PSW_MASK_EXT;
-#ifndef CONFIG_ARCH_S390X
-        asm volatile (
-                "    basr %0,0\n"
-                "0:  la   %0,1f-0b(%0)\n"
-                "    st   %0,4(%1)\n"
-                "    oi   4(%1),0x80\n"
-                "    lpsw 0(%1)\n"
-                "1:  la   %0,2f-1b(%0)\n"
-                "    st   %0,4(%1)\n"
-                "    oi   4(%1),0x80\n"
-                "    ni   1(%1),0xf9\n"
-                "    lpsw 0(%1)\n"
-                "2:"
-                : "=&a" (reg) : "a" (&wait_psw) : "memory", "cc" );
-#else /* CONFIG_ARCH_S390X */
-        asm volatile (
-                "    larl  %0,0f\n"
-                "    stg   %0,8(%1)\n"
-                "    lpswe 0(%1)\n"
-                "0:  larl  %0,1f\n"
-                "    stg   %0,8(%1)\n"
-                "    ni    1(%1),0xf9\n"
-                "    lpswe 0(%1)\n"
-                "1:"
-                : "=&a" (reg) : "a" (&wait_psw) : "memory", "cc" );
-#endif /* CONFIG_ARCH_S390X */
+        local_mcck_disable();
+        if (test_thread_flag(TIF_MCCK_PENDING)) {
+                local_mcck_enable();
+                local_irq_enable();
+                s390_handle_mcck();
+                return;
+        }
+
+        /* Wait for external, I/O or machine check interrupt. */
+        __load_psw_mask(PSW_KERNEL_BITS | PSW_MASK_WAIT |
+                        PSW_MASK_IO | PSW_MASK_EXT);
 }
 
 void cpu_idle(void)
diff --git a/arch/s390/kernel/setup.c b/arch/s390/kernel/setup.c
index df83215beac3..eb7be0ad7175 100644
--- a/arch/s390/kernel/setup.c
+++ b/arch/s390/kernel/setup.c
@@ -414,7 +414,8 @@ setup_lowcore(void)
         lc->program_new_psw.mask = PSW_KERNEL_BITS;
         lc->program_new_psw.addr =
                 PSW_ADDR_AMODE | (unsigned long)pgm_check_handler;
-        lc->mcck_new_psw.mask = PSW_KERNEL_BITS;
+        lc->mcck_new_psw.mask =
+                PSW_KERNEL_BITS & ~PSW_MASK_MCHECK & ~PSW_MASK_DAT;
         lc->mcck_new_psw.addr =
                 PSW_ADDR_AMODE | (unsigned long) mcck_int_handler;
         lc->io_new_psw.mask = PSW_KERNEL_BITS;
@@ -424,12 +425,18 @@ setup_lowcore(void)
         lc->kernel_stack = ((unsigned long) &init_thread_union) + THREAD_SIZE;
         lc->async_stack = (unsigned long)
                 __alloc_bootmem(ASYNC_SIZE, ASYNC_SIZE, 0) + ASYNC_SIZE;
-#ifdef CONFIG_CHECK_STACK
         lc->panic_stack = (unsigned long)
                 __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, 0) + PAGE_SIZE;
-#endif
         lc->current_task = (unsigned long) init_thread_union.thread_info.task;
         lc->thread_info = (unsigned long) &init_thread_union;
+#ifndef CONFIG_ARCH_S390X
+        if (MACHINE_HAS_IEEE) {
+                lc->extended_save_area_addr = (__u32)
+                        __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, 0);
+                /* enable extended save area */
+                ctl_set_bit(14, 29);
+        }
+#endif
 #ifdef CONFIG_ARCH_S390X
         if (MACHINE_HAS_DIAG44)
                 lc->diag44_opcode = 0x83000044;
diff --git a/arch/s390/kernel/smp.c b/arch/s390/kernel/smp.c
index 93c71fef99dc..50c335067cfe 100644
--- a/arch/s390/kernel/smp.c
+++ b/arch/s390/kernel/smp.c
@@ -773,13 +773,24 @@ void __init smp_prepare_cpus(unsigned int max_cpus)
 
                 *(lowcore_ptr[i]) = S390_lowcore;
                 lowcore_ptr[i]->async_stack = stack + (ASYNC_SIZE);
-#ifdef CONFIG_CHECK_STACK
                 stack = __get_free_pages(GFP_KERNEL,0);
                 if (stack == 0ULL)
                         panic("smp_boot_cpus failed to allocate memory\n");
                 lowcore_ptr[i]->panic_stack = stack + (PAGE_SIZE);
+#ifndef __s390x__
+                if (MACHINE_HAS_IEEE) {
+                        lowcore_ptr[i]->extended_save_area_addr =
+                                (__u32) __get_free_pages(GFP_KERNEL,0);
+                        if (lowcore_ptr[i]->extended_save_area_addr == 0)
+                                panic("smp_boot_cpus failed to "
+                                      "allocate memory\n");
+                }
 #endif
         }
+#ifndef __s390x__
+        if (MACHINE_HAS_IEEE)
+                ctl_set_bit(14, 29); /* enable extended save area */
+#endif
         set_prefix((u32)(unsigned long) lowcore_ptr[smp_processor_id()]);
 
         for_each_cpu(cpu)